test(tty): stabilize no-accumulation test — delta vs pre-loop snapshot of unrelated system sleeps (mount helper sleeps race the window)

test(tty): canonical tool hash parity (bash.ts + await-operation.ts) — same contract as hook parity
feat(tty): MiMo harness process-ownership contract — group-killed bash execution + explicit detachment
2026-09-19 07:24:10 +01:00 · 2026-09-19 07:18:07 +01:00 · 2026-09-19 07:16:35 +01:00 · 2026-09-19 06:39:11 +01:00 · 2026-09-19 06:33:48 +01:00 · 2026-09-19 06:12:38 +01:00
275 changed files with 108556 additions and 2425 deletions
@@ -213,38 +213,17 @@ jobs:
              env.write(f"RELEASE_VERSION={NEW}\n")
              env.write("AUTO_BUMPED=true\n")
          PYBUMP
-      - name: Create release from merged pull request
+      - name: Create evidence-derived release (SCAFFOLD-TRANCHE-009B)
        env:
          GITEA_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-          PR_TITLE: ${{ gitea.event.pull_request.title }}
-          PR_BODY: ${{ gitea.event.pull_request.body }}
-          RELEASE_API_URL: ${{ gitea.server_url }}/api/v1/repos/${{ gitea.repository }}/releases
-          LIST_API_URL: ${{ gitea.server_url }}/api/v1/repos/${{ gitea.repository }}/releases?limit=50
+          RELEASE_SERVER_URL: ${{ gitea.server_url }}
+          RELEASE_REPO: ${{ gitea.repository }}
        run: |
-          # Skip idempotently if a release for this tag already exists (retry/replay safety).
-          if curl --fail --silent "$LIST_API_URL" \
-            -H "Authorization: token $GITEA_TOKEN" \
-            | python3 -c 'import json,sys,os; v=os.environ["RELEASE_VERSION"]; exists=[r for r in json.load(sys.stdin) if r.get("tag_name")=="v"+v]; sys.exit(0 if exists else 1)'; then
-            echo "Release v$RELEASE_VERSION already exists — skipping creation"
-            exit 0
-          fi
-          python3 - <<'PY' > /tmp/rune-release.json
-          import json, os
-          version = os.environ["RELEASE_VERSION"]
-          body = f"## {os.environ['PR_TITLE']}\n\n{os.environ.get('PR_BODY', '')}"
-          print(json.dumps({
-              "tag_name": f"v{version}",
-              "target_commitish": "master",
-              "name": f"Rune v{version}",
-              "body": body,
-              "draft": False,
-              "prerelease": False,
-          }))
-          PY
-          curl --fail-with-body --silent --show-error \
-            -X POST \
-            -H "Authorization: token $GITEA_TOKEN" \
-            -H "Content-Type: application/json" \
-            --data-binary @/tmp/rune-release.json \
-            "$RELEASE_API_URL"
-          rm -f /tmp/rune-release.json
+          # ONE note-generation authority: the tested release_notes module.
+          # The old PR-title/body paste path (which produced empty '##'
+          # bodies on direct-push releases) is REMOVED — governed evidence
+          # from the scaffold store is the source, not PR prose.
+          python3 scripts/release_history_cli.py init \
+            --version "$RELEASE_VERSION" \
+            --server "$RELEASE_SERVER_URL" --repo "$RELEASE_REPO" \
+            --store store/scaffold/current.json
@@ -286,12 +286,61 @@ jobs:
        run: |
          NEW="$(echo "${{ gitea.event.pull_request.head.ref }}" | sed 's|chore/auto-bump-||')"
          python3 scripts/check_version_surfaces.py check --root . --version "$NEW"
-      - name: Publish protected-branch gate (full pipeline path)
-        if: ${{ !startsWith(gitea.event.pull_request.head.ref, 'chore/auto-bump') }}
+      - name: Publish protected-branch gate (full pipeline path — all needs green)
+        # RELEASE-SAFETY FIX (2026-09-13): the previous version of this step
+        # read /tmp/need_* files that no step ever writes and echoed success
+        # UNCONDITIONALLY — required-checks published green while type-check
+        # and unit-tests were red, and deploy promoted the red commits
+        # (observed live: 24413e4, fd6e9b4, de30cd9 promoted 0.66.28/29 with
+        # failing type-check). The verdict now comes from the needs context
+        # itself: this step runs ONLY when every required job succeeded.
+        if: >-
+          ${{ !startsWith(gitea.event.pull_request.head.ref, 'chore/auto-bump')
+              && needs.lint.result == 'success'
+              && needs['unit-tests'].result == 'success'
+              && needs.canary.result == 'success'
+              && needs['best-practices'].result == 'success'
+              && needs.documentation.result == 'success'
+              && needs['type-check'].result == 'success'
+              && needs['gitea-offline'].result == 'success' }}
+        run: echo "All required CI checks passed"
+      - name: Refuse promotion when any required check failed (fail-closed)
+        # Companion to the step above: if anything is not exactly success
+        # (failure, cancelled, skipped on a non-bump path), the required-checks
+        # context reports FAILURE — deploy_qualify then refuses promotion.
+        # Fail-closed: not verified → not deployable.
+        if: >-
+          ${{ !startsWith(gitea.event.pull_request.head.ref, 'chore/auto-bump')
+              && (needs.lint.result != 'success'
+                  || needs['unit-tests'].result != 'success'
+                  || needs.canary.result != 'success'
+                  || needs['best-practices'].result != 'success'
+                  || needs.documentation.result != 'success'
+                  || needs['type-check'].result != 'success'
+                  || needs['gitea-offline'].result != 'success') }}
        run: |
-          # All needs must have succeeded (skipped deps = this is a bump PR, handled above)
-          FAILED=""
-          for job in lint unit-tests canary best-practices documentation type-check gitea-offline; do
-            RESULT=$(cat /tmp/need_$job 2>/dev/null || echo "")
-          done
-          echo "All required CI checks passed"
+          echo "::error::Required CI checks not green — refusing to publish required-checks success"
+          echo "lint=${{ needs.lint.result }} unit-tests=${{ needs['unit-tests'].result }} canary=${{ needs.canary.result }} best-practices=${{ needs['best-practices'].result }} documentation=${{ needs.documentation.result }} type-check=${{ needs['type-check'].result }} gitea-offline=${{ needs['gitea-offline'].result }}"
+          exit 1
+
+  notify-event-bridge:
+    # T16 §16C: post the CI completion WAKE to the local event bridge so
+    # event-driven waiters (deploy.yml → ci_event_wait.py) resume on the
+    # event instead of sleep-polling. ALWAYS runs (success or failure —
+    # a red CI must wake the waiter just as loudly as a green one). The
+    # wake is best-effort and never decides the verdict: the waiter
+    # re-reads the canonical Gitea state via deploy_qualify.py.
+    name: CI Pipeline / notify-event-bridge
+    runs-on: rune-ci-desktop
+    needs: [required-checks]
+    if: always()
+    steps:
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683
+        with:
+          fetch-depth: 1
+      - name: Wake event bridge (wake only — never completion truth)
+        env:
+          CI_SHA: ${{ gitea.sha }}
+          CI_RESULT: ${{ needs.required-checks.result }}
+          RUNE_EVENT_BRIDGE_URL: ${{ vars.RUNE_EVENT_BRIDGE_URL || 'http://127.0.0.1:8791' }}
+        run: python3 scripts/notify_event_bridge.py
@@ -19,38 +19,74 @@ jobs:
        with:
          fetch-depth: 0

-      - name: Wait for CI qualification on this exact SHA
+      - name: Wait for CI qualification on this exact SHA (event-driven, T16 §16C)
+        # T16 §16C/§16D: the old sleep-30 → query → repeat loop is gone.
+        # The waiter registers a correlation on the W/R event bridge and
+        # yields in select() until ci.yml's notify-event-bridge job posts
+        # the completion WAKE; the verdict is then re-read from canonical
+        # Gitea (deploy_qualify.py). The 900s deadline is a FAILURE
+        # DETECTOR (lost-event recovery), never the scheduler.
        env:
          GITEA_SERVER: ${{ gitea.server_url }}
          GITEA_REPO: ${{ gitea.repository }}
          GITEA_TOKEN: ${{ secrets.ROBOT_TOKEN }}
+          # Shared bridge data root: the deploy step runs as the runner
+          # user while the daemon runs as nick — the rendezvous state
+          # (waits/events/resumes/leases) must be shared. §16C live
+          # finding: defaulting to the invoking user's home made the
+          # controller unable to create its FIFO (PermissionError) and
+          # the deploy failed fast+truthfully on both 0.66.63 pushes.
+          RUNE_WR_ROOT: /mnt/workdrive/wr-bridge
+          RUNE_WR_CODE: /home/nick/.local/share/mimo-wait-resume
        run: |
          DEPLOY_SHA=$(git rev-parse HEAD)
          export DEPLOY_SHA
-          echo "Qualifying $DEPLOY_SHA"
-          DEADLINE=$((SECONDS + 900))
-          while [ $SECONDS -lt $DEADLINE ]; do
-            STATUS=$(python3 scripts/deploy_qualify.py)
-            echo "required-checks: $STATUS"
-            if [ "$STATUS" = "success" ]; then exit 0; fi
-            if [ "$STATUS" = "failure" ]; then
-              echo "::error::CI failed on $DEPLOY_SHA — deployment refused"
-              exit 1
-            fi
-            sleep 30
-          done
-          echo "::error::CI qualification timeout on $DEPLOY_SHA — deployment refused"
-          exit 1
+          echo "Qualifying $DEPLOY_SHA (event-driven)"
+          # READINESS BEFORE ACTION (§16U): fail with a clear message if
+          # the bridge rendezvous is not writable in THIS user's context,
+          # instead of a stack trace at FIFO creation time.
+          if ! touch /mnt/workdrive/wr-bridge/resumes/.write-probe 2>/dev/null; then
+            echo "::error::RUNE_WR_ROOT not writable by $(id -un) — bridge data root misconfigured"
+            exit 1
+          fi
+          rm -f /mnt/workdrive/wr-bridge/resumes/.write-probe
+          python3 scripts/ci_event_wait.py --sha "$DEPLOY_SHA" --deadline 900

-      - name: Stage, verify, promote
+      - name: Stage, verify, promote (monotonic)
        env:
          GITEA_TOKEN: ${{ secrets.ROBOT_TOKEN }}
+          GITEA_SERVER: ${{ gitea.server_url }}
+          GITEA_REPO: ${{ gitea.repository }}
        run: |
          python3 scripts/deploy_stage.py

-      - name: Post-deploy runtime verification (the consuming path)
+      - name: Promotion truthfulness check
+        id: promotion_truth
        run: |
          SHA=$(git rev-parse HEAD)
+          CURRENT_TARGET=$(readlink /mnt/workdrive/rune-code/current || true)
+          if [ "$CURRENT_TARGET" = "/mnt/workdrive/rune-code/${SHA:0:12}" ]; then
+            echo "PROMOTED: current == candidate ($SHA)"
+            echo "promoted=true" >> "$GITEA_OUTPUT"
+          else
+            # Not promoted. deploy_stage exit 0 + non-matching current =
+            # NOOP_ALREADY_CURRENT or STALE_SUPERSEDED — truthful non-error
+            # terminal states (a newer build already won). Do NOT claim
+            # promotion; the runtime-verification step below only runs for
+            # real promotions.
+            echo "NOT PROMOTED: current -> $CURRENT_TARGET (superseded/noop — current is authoritative)"
+            echo "promoted=false" >> "$GITEA_OUTPUT"
+          fi
+
+      - name: Post-deploy runtime verification (only for real promotions)
+        if: ${{ success() }}
+        run: |
+          SHA=$(git rev-parse HEAD)
+          CURRENT_TARGET=$(readlink /mnt/workdrive/rune-code/current || true)
+          if [ "$CURRENT_TARGET" != "/mnt/workdrive/rune-code/${SHA:0:12}" ]; then
+            echo "SKIP runtime verification: this job did not promote (superseded/noop)"
+            exit 0
+          fi
          # Exactly what the launcher execs: cd <current>/src, import rune.
          RUNE_HOME=/mnt/workdrive/rune RUNE_MEMORY_DIR=/home/nick/.config/opencode/memory \
            /home/nick/.venvs/memory/bin/python3 -c "
@@ -62,3 +98,28 @@ jobs:
            assert rune.__version__ == info['version']
            print('RUNTIME VERIFIED:', info)
            "
+      - name: Release-history convergence fallback (SCAFFOLD-TRANCHE-0012C)
+        # The PRIMARY convergence mechanism is deploy_stage.py's out-of-lock
+        # publisher (publishes from CURRENT deployment authority, §14B/§14C).
+        # This step is a non-blocking FALLBACK: if the primary publish failed
+        # (e.g. missing Gitea config), this retries. NEVER fails the job —
+        # documentation failure never outweighs deployment validity.
+        if: ${{ success() && steps.promotion_truth.outputs.promoted == 'true' }}
+        continue-on-error: true
+        env:
+          GITEA_TOKEN: ${{ secrets.ROBOT_TOKEN }}
+          RELEASE_SERVER_URL: ${{ gitea.server_url }}
+          RELEASE_REPO: ${{ gitea.repository }}
+        run: |
+          VERSION=$(python3 -c "import sys; sys.path.insert(0,'src'); import rune; print(rune.__version__)" 2>/dev/null || echo "")
+          if [ -z "$VERSION" ]; then
+            echo "SKIP: cannot determine version"
+            exit 0
+          fi
+          python3 scripts/release_history_cli.py refresh \
+            --version "$VERSION" \
+            --server "$RELEASE_SERVER_URL" --repo "$RELEASE_REPO" \
+            --store "$GITHUB_WORKSPACE/store/scaffold/current.json" \
+            --release-sha "$(git rev-parse HEAD)" \
+            --deployed-sha "$(git rev-parse HEAD)" \
+          || echo "::warning::release-history fallback refresh FAILED — in-lock hook may have already converged; retry converges"
@@ -22,9 +22,12 @@ jobs:
    steps:
      - uses: actions/checkout@v4
      - name: Run all canaries
+        # T16 §16Q reconciliation: the glob looped shared/canary_*.py —
+        # real canaries live in src/rune/ (canary_native.MODULES is the
+        # registry authority). The old job silently no-opped.
        run: |
          FAILED=0
-          for canary in shared/canary_*.py; do
+          for canary in src/rune/canary_*.py; do
            if [ -f "$canary" ]; then
              echo "Running $canary..."
              python3 "$canary" || FAILED=$((FAILED + 1))
@@ -1,6 +1,6 @@
 # Rune

-**Current release: v0.66.31**
+**Current release: v0.86.0**

 **Rune is a persistent personal agent designed to accumulate experience, develop reusable competence, and remain recognisably the same agent even when its underlying foundation model changes.**

@@ -0,0 +1,196 @@
+"""completion_file.py — REAL_ASYNC_BRIDGE: durable completion-file delivery
+with inotify wake. Replaces FIFO+lease for process-wait delivery.
+
+DESIGN:
+  Producer (bridge/daemon/notify):
+    1. Validates operation identity
+    2. Atomically writes a completion file:
+         resumes/<operation_id>.done
+       containing the event/verdict JSON (write to tmp, rename for atomicity)
+
+  Waiter (ci_event_wait / any continuation consumer):
+    1. Checks if the completion file already exists (event-before-waiter)
+    2. If not: registers inotify watch on the resumes/ directory
+    3. Re-checks after watch registration (closes the registration race)
+    4. Blocks in kernel on inotify — zero CPU, zero timer
+    5. On file-created event: reads the file, verifies authority, returns
+
+  The event is a WAKE SIGNAL. Authority still decides terminal truth.
+  The completion file is durable — survives daemon and waiter restarts.
+
+WHY NOT FIFO:
+  FIFO is one-shot (token consumed), requires lease management, and
+  fails under session-busy guards. A completion file is atomic, durable,
+  and works across restarts. inotify wakes the process with zero CPU
+  cost and zero latency between file creation and wake.
+
+WHY NOT POLLING:
+  Polling introduces an arbitrary interval. inotify is a kernel callback
+  that fires exactly when the filesystem changes — the latency tracks
+  the real completion event, not a configured interval.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import select
+import struct
+import time
+from pathlib import Path
+
+
+def _wr_bridge() -> Path:
+    # WR_ROOT = daemon/infra convention; RUNE_WR_ROOT = repo/scripts
+    # convention. Both honored (WR_ROOT wins) so the daemon, the waiter,
+    # and any isolated test root always resolve the SAME directory.
+    root = os.environ.get("WR_ROOT") or os.environ.get("RUNE_WR_ROOT") or "/mnt/workdrive/wr-bridge"
+    return Path(root)
+
+
+def _resumes_dir() -> Path:
+    return _wr_bridge() / "resumes"
+
+
+# inotify constants (Linux)
+IN_CREATE = 0x100
+IN_CLOSE_WRITE = 0x8
+IN_MOVED_TO = 0x80
+
+
+def completion_path(operation_id: str) -> Path:
+    return _resumes_dir() / f"{operation_id}.done"
+
+
+def write_completion(operation_id: str, event: dict) -> Path:
+    """Atomically write a durable completion file. The rename(2) ensures
+    the waiter never sees a partial file."""
+    _resumes_dir().mkdir(parents=True, exist_ok=True)
+    target = completion_path(operation_id)
+    tmp = target.parent / f".{target.name}.tmp.{os.getpid()}"
+    tmp.write_text(json.dumps(event, sort_keys=True) + "\n")
+    os.rename(str(tmp), str(target))  # atomic on same filesystem
+    os.chmod(str(target), 0o666)  # cross-user bridge
+    return target
+
+
+def read_completion(operation_id: str) -> dict | None:
+    """Read a durable completion file. Returns None if absent."""
+    p = completion_path(operation_id)
+    if not p.is_file():
+        return None
+    try:
+        return json.loads(p.read_text())
+    except (OSError, ValueError):
+        return None
+
+
+def completion_exists(operation_id: str) -> bool:
+    return completion_path(operation_id).is_file()
+
+
+def _inotify_init_and_add_watch(directory: str) -> int:
+    """Create an inotify instance and add a watch on the directory.
+    Returns the inotify fd."""
+    libc = __import__("ctypes").CDLL(None, use_errno=True)
+    fd = libc.inotify_init()
+    if fd < 0:
+        raise OSError("inotify_init failed")
+    # IN_CREATE | IN_MOVED_TO: catch both direct creation and atomic rename
+    mask = IN_CREATE | IN_MOVED_TO
+    wd = libc.inotify_add_watch(fd, directory.encode(), mask)
+    if wd < 0:
+        os.close(fd)
+        raise OSError(f"inotify_add_watch failed for {directory}")
+    return fd
+
+
+def _inotify_read(fd: int, timeout_secs: float | None = None) -> list[str]:
+    """Read pending inotify events. Returns list of filenames that were
+    created/moved-to. Blocks until at least one event or timeout."""
+    poll = select.poll()
+    poll.register(fd, select.POLLIN)
+    events = []
+    buf = os.read(fd, 4096)
+    pos = 0
+    while pos < len(buf):
+        # struct: wd(i32), mask(u32), cookie(u32), name_len(u32)
+        wd, mask, cookie, name_len = struct.unpack_from("iIII", buf, pos)
+        pos += 16
+        name = buf[pos : pos + name_len].rstrip(b"\x00").decode("utf-8", "replace")
+        pos += name_len
+        if mask & (IN_CREATE | IN_MOVED_TO):
+            events.append(name)
+    return events
+
+
+def wait_for_completion(operation_id: str, *, timeout_secs: float | None = None, poll_callback=None) -> dict | None:
+    """Block until a completion file exists for this operation.
+
+    Uses inotify for zero-latency, zero-CPU wake. Falls back to
+    polling_callback (if provided) when inotify is unavailable.
+
+    The completion file is a WAKE SIGNAL. The caller must still verify
+    canonical authority after reading it.
+
+    Returns the event dict from the completion file, or None on timeout.
+    """
+    target = completion_path(operation_id)
+
+    # 1. check-before-watch (event may have arrived before we started)
+    existing = read_completion(operation_id)
+    if existing is not None:
+        return existing
+
+    _resumes_dir().mkdir(parents=True, exist_ok=True)
+
+    try:
+        fd = _inotify_init_and_add_watch(str(_resumes_dir()))
+    except OSError:
+        # inotify unavailable: fall back to polling callback
+        if poll_callback:
+            return poll_callback(operation_id, timeout_secs)
+        return None
+
+    try:
+        # 2. re-check after watch registration (closes the race between
+        #    the initial check and the watch being active)
+        existing = read_completion(operation_id)
+        if existing is not None:
+            return existing
+
+        # 3. block in kernel on inotify — zero CPU, exact latency
+        import select as _select
+
+        poll = _select.poll()
+        poll.register(fd, _select.POLLIN)
+        deadline = (time.monotonic() + timeout_secs) if timeout_secs else None
+
+        while True:
+            remaining = None
+            if deadline is not None:
+                remaining = max(0.1, deadline - time.monotonic())
+                if remaining <= 0.1:
+                    return None  # timeout
+            events = poll.poll(remaining * 1000 if remaining else None)
+            if not events:
+                continue  # poll timeout, re-check
+            names = _inotify_read(fd)
+            if target.name in names:
+                # 4. file appeared — read it
+                existing = read_completion(operation_id)
+                if existing is not None:
+                    return existing
+                # partial/corrupt: keep watching
+    finally:
+        os.close(fd)
+    return None
+
+
+def clear_completion(operation_id: str) -> None:
+    """Remove a consumed completion file (idempotent cleanup)."""
+    p = completion_path(operation_id)
+    try:
+        p.unlink()
+    except OSError:
+        pass
@@ -0,0 +1,799 @@
+#!/usr/bin/env python3
+"""daemon.py — Gitea webhook receiver + reconciliation + resume driver (v0.2).
+
+W/R-v0.2 invariants implemented here:
+  A. SINGLE-FLIGHT PER SESSION — per-session lease (waitstore.acquire_lease,
+     /proc-validated, atomic) + in-process inflight set. A second READY wait
+     for the same session stays queued until the lease releases.
+  B. BUSY SESSION PROTECTION — before spawning, /proc is scanned for a live
+     `mimo run -s <sid>`-shaped process not owned by our lease. Found → skip
+     launch (wait stays READY). No invented MiMo APIs; the lease is the
+     conservative floor against W/R-created concurrency.
+  C. ATOMIC SESSION LEASE — persisted {session_id, wait_id, pid, pgid,
+     pid_starttime, acquired_at, state}; O_EXCL create, steal only on /proc
+     evidence of a dead owner.
+  D. PROCESS TREE OWNERSHIP — workers spawn with start_new_session=True (own
+     process group). Watch timeout ⇒ SIGTERM group → bounded grace → SIGKILL
+     group → verify zero surviving group members → release lease → truthful
+     READY_TO_RETRY (never terminal-on-timeout).
+  E. TIMEOUT IS NOT SUCCESS — rc==0 maps to verdict terminal
+     (COMPLETED/FAILED); rc!=0 or unverifiable ⇒ RESUME_UNVERIFIED
+     (not schedulable, operator-requeueable); timeout ⇒ READY_TO_RETRY.
+  F. STALE RESUMING RECOVERY — on startup, every RESUMING record is
+     inspected: live lease owner ⇒ adopted and monitored; otherwise ⇒
+     RESUME_UNVERIFIED (+ stale lease released). Never auto-relaunched.
+  G. MULTIPLE WAITS, SAME SESSION — schedulable() orders oldest-first; one
+     continuation finishes before the next starts.
+  H. PIDFILE CORRECTNESS — singleton enforced by flock on daemon.pid; the
+     lock dies with the process, so stale pidfiles (incl. SIGKILL) cannot
+     admit a second daemon.
+
+The resume command is configurable (WR_RESUME_CMD, JSON list with
+{session_id}/{msg} placeholders) so the test harness can use a stub; the
+default is the real mechanism: `mimo run -s <sid> -m <model> <msg>`.
+"""
+
+from __future__ import annotations
+import fcntl, json, os, signal, subprocess, sys, threading, time
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).parent))
+import waitstore
+import verify
+import completion_file  # REAL_ASYNC_BRIDGE R3: durable completion-file delivery
+
+RESUME_MODEL = "Verboo Code/glm-5.3-flash"
+RECONCILE_SECS = int(os.environ.get("WR_RECONCILE_SECS", "60"))
+RECONCILE_EVERY_TICKS = max(1, int(os.environ.get("WR_RECONCILE_TICKS", "3")))
+WAKE_MIN_AGE_SECS = float(os.environ.get("WR_WAKE_MIN_AGE_SECS", "5"))
+WAKE_RETRIES = int(os.environ.get("WR_WAKE_RETRIES", "3"))
+WAKE_RETRY_SECS = float(os.environ.get("WR_WAKE_RETRY_SECS", "4"))
+DRIVE_LOOP_SECS = int(os.environ.get("WR_DRIVE_SECS", "20"))
+# REAL_ASYNC_BRIDGE §4: a RESUMING wait older than this grace without
+# progress is an orphaned resume flow (thread died between lease acquire
+# and release). Must exceed the longest legitimate in-flight watch
+# (WATCH_SECS); default 3x. Env-overridable for deterministic tests.
+WATCH_SECS = int(os.environ.get("WR_WATCH_SECS", "600"))
+# REAL_ASYNC_BRIDGE §4: a RESUMING wait older than this grace without
+# progress is an orphaned resume flow (thread died between lease acquire
+# and release). Must exceed the longest legitimate in-flight watch
+# (WATCH_SECS); default 3x. Env-overridable for deterministic tests.
+ORPHAN_GRACE_SECS = int(os.environ.get("WR_ORPHAN_GRACE_SECS", str(3 * WATCH_SECS)))
+TERM_GRACE_SECS = int(os.environ.get("WR_TERM_GRACE_SECS", "10"))
+MAX_ATTEMPTS = int(os.environ.get("WR_MAX_ATTEMPTS", "3"))
+_WR_ROOT = Path(os.environ.get("WR_ROOT", "") or Path(__file__).resolve().parent)
+PIDFILE = _WR_ROOT / "daemon.pid"
+LOG = _WR_ROOT / "daemon.log"
+
+DEFAULT_RESUME_CMD = ["mimo", "run", "-s", "{session_id}", "-m", RESUME_MODEL, "{msg}"]
+
+
+def log(msg: str) -> None:
+    with open(LOG, "a") as f:
+        f.write(f"{time.strftime('%H:%M:%S')} {msg}\n")
+
+
+def resume_cmd(session_id: str, msg: str, wait_id: str = "", override: list | None = None) -> list[str]:
+    """T15 §15D: per-wait override wins; placeholders {session_id}/{msg}/
+    {wait_id}. A synthetic controller session (Scaffolding stage driver)
+    is not a MiMo session — the global default cannot resume it, which is
+    exactly why fixed-sleep polling crept back into T12-T14."""
+    if override:
+        return [
+            c.replace("{session_id}", session_id).replace("{msg}", msg).replace("{wait_id}", wait_id) for c in override
+        ]
+    tmpl = os.environ.get("WR_RESUME_CMD")
+    cmd = json.loads(tmpl) if tmpl else DEFAULT_RESUME_CMD
+    return [c.replace("{session_id}", session_id).replace("{msg}", msg).replace("{wait_id}", wait_id) for c in cmd]
+
+
+# ── process-tree evidence & ownership ─────────────────────────────────────
+
+
+def pgid_members(pgid: int) -> list[int]:
+    out = []
+    for d in Path("/proc").iterdir():
+        if not d.name.isdigit():
+            continue
+        try:
+            stat = (d / "stat").read_text()
+            fields = stat.rsplit(")", 1)[1].split()
+            # zombies: killed but not yet reaped. Counting them makes the
+            # survivor check report the direct child we just KILLed before
+            # communicate() reaps it (observed live 2026-09-12).
+            if fields[0] == "Z":
+                continue
+            if int(fields[2]) == pgid:
+                out.append(int(d.name))
+        except (OSError, IndexError, ValueError):
+            continue
+    return out
+
+
+def terminate_tree(pgid: int) -> list[int]:
+    """TERM → bounded grace → KILL. Returns surviving members (must be [])."""
+    try:
+        os.killpg(pgid, signal.SIGTERM)
+    except ProcessLookupError:
+        pass
+    deadline = time.time() + TERM_GRACE_SECS
+    while time.time() < deadline:
+        if not pgid_members(pgid):
+            return []
+        time.sleep(0.2)
+    try:
+        os.killpg(pgid, signal.SIGKILL)
+    except ProcessLookupError:
+        pass
+    for _ in range(50):  # up to 10s for KILL to land
+        if not pgid_members(pgid):
+            return []
+        time.sleep(0.2)
+    return pgid_members(pgid)
+
+
+def stray_worker_pids(session_id: str, exclude_pgids: set[int] = ()) -> list[int]:
+    """Live `mimo run -s <sid>`-shaped processes not owned by our workers."""
+    out = []
+    for d in Path("/proc").iterdir():
+        if not d.name.isdigit():
+            continue
+        try:
+            parts = [p.decode(errors="replace") for p in (d / "cmdline").read_bytes().split(b"\0") if p]
+        except OSError:
+            continue
+        if not parts:
+            continue
+        joined = " ".join(parts)
+        if session_id not in joined or "-s" not in parts:
+            continue
+        if "mimo" not in joined.lower() and "@mimo-ai" not in joined.lower():
+            continue
+        try:
+            if os.getpgid(int(d.name)) in exclude_pgids:
+                continue
+        except OSError:
+            continue
+        out.append(int(d.name))
+    return out
+
+
+# ── verification → transition (unchanged authority model) ─────────────────
+
+
+def _authoritative_verify(rec: dict) -> tuple[str, dict]:
+    """Authoritative verification with a deterministic test seam.
+
+    Production (WR_VERIFY_STUB unset): real Gitea API via verify.py — the
+    ONLY source of success/failure truth. Test (WR_VERIFY_STUB=path to
+    JSON {"<run_ref-or-sha>": verdict, "default": verdict}): stubbed
+    verdicts so invariant tests never touch the network. The stub is a
+    test fixture, not an alternate authority — no production path sets it.
+    """
+    stub_path = os.environ.get("WR_VERIFY_STUB", "")
+    if stub_path:
+        try:
+            spec = json.loads(Path(stub_path).read_text())
+        except Exception:
+            return "pending", {"stub_error": "unreadable stub"}
+        key = str(rec.get("run_ref", "") or rec.get("sha", ""))
+        verdict = spec.get(key, spec.get("default", "pending"))
+        return verdict, {"stub": True, "key": key}
+    import verify
+
+    if rec.get("wait_kind") == "RELEASE_DEPLOY":
+        return verify.verify_release_deploy(rec.get("sha", ""), rec.get("repo_root", ""), rec.get("deployed_link", ""))
+    return verify.verify_run(rec.get("run_ref", ""), rec.get("sha", ""), rec.get("expected", "success"))
+
+
+def try_transition(rec: dict, event: dict | None = None) -> dict:
+    if rec["status"] != "WAITING_EXTERNAL":
+        return rec
+    if rec.get("wait_kind") == "PROCESS_EVENT":
+        # T15 §15D: the matching event IS the authority. The notifier
+        # wrapper observed the real process exit; exit_code decides the
+        # verdict. No Gitea lookup applies. A duplicate event never
+        # reaches here (match applies only to WAITING_EXTERNAL waits).
+        if not event:
+            return rec  # no event → nothing authoritative happened
+        try:
+            exit_code = int(event.get("exit_code"))
+        except (TypeError, ValueError):
+            exit_code = None
+        if exit_code is None:
+            return rec  # malformed event: stay WAITING (15E hostile case)
+        verdict = "success" if exit_code == 0 else "failure"
+        evidence = {"process_event": True, "exit_code": exit_code, "operation_id": event.get("run_ref", "")}
+        log(f"verify {rec['wait_id']}: {verdict} {json.dumps(evidence)[:120]}")
+        rec["status"] = "READY_TO_RESUME"
+        rec["verdict"] = verdict
+        if verdict == "failure":
+            rec["failure_evidence"] = evidence
+        rec["process_event_evidence"] = evidence
+        waitstore.save(rec)
+        return rec
+    verdict, evidence = _authoritative_verify(rec)
+    log(f"verify {rec['wait_id']}: {verdict} {json.dumps(evidence)[:120]}")
+    if verdict == "success":
+        rec["status"] = "READY_TO_RESUME"
+        rec["verdict"] = "success"
+    elif verdict == "failure":
+        rec["status"] = "READY_TO_RESUME"
+        rec["verdict"] = "failure"
+        rec["failure_evidence"] = verify.failed_job_evidence(rec.get("run_ref", ""))
+    else:
+        return rec  # pending/unknown → stay WAITING
+    waitstore.save(rec)
+    return rec
+
+
+def continuation_msg(rec: dict) -> str:
+    c = rec["continuation"]
+    if rec.get("verdict") == "success":
+        return (
+            f"External wait satisfied (wait {rec['wait_id']}).\n"
+            f"Original objective: {c.get('objective', '')}\n"
+            f"Waited for: {c.get('waited_for', '')}\n"
+            f"Authoritative result: workflow success verified via Gitea API.\n"
+            f"SHA/run: {rec.get('sha', '')}/{rec.get('run_ref', '')}\n"
+            f"Verify: {c.get('verify_when_woken', '')}\n"
+            f"Continue from: {c.get('on_success', '')}"
+        )
+    ev = json.dumps(rec.get("failure_evidence", {}))[:400]
+    return (
+        f"External wait finished with FAILURE (wait {rec['wait_id']}).\n"
+        f"Original objective: {c.get('objective', '')}\n"
+        f"Authoritative result: workflow FAILURE verified via Gitea API.\n"
+        f"Failed job evidence: {ev}\n"
+        f"SHA/run: {rec.get('sha', '')}/{rec.get('run_ref', '')}\n"
+        f"Do NOT treat this as success. Continue from: {c.get('on_failure', '')}"
+    )
+
+
+# ── resume driver (single-flight, tree-owned) ─────────────────────────────
+
+
+def resume(rec: dict) -> None:
+    """Run one continuation to completion. Runs on a dedicated worker thread.
+    Preconditions: caller verified no inflight worker for this session."""
+    sid, wid = rec["session_id"], rec["wait_id"]
+    # REAL_ASYNC_BRIDGE §2 invariant: a successful authority-verified
+    # operation completion must NEVER be overwritten or downgraded by a
+    # legacy session/FIFO delivery failure. Terminal operation state
+    # (COMPLETED/FAILED — stamped by the waiter's own authority
+    # verification) is owned by the OPERATION domain; this function owns
+    # only SESSION-DELIVERY. Re-read from disk: the in-memory copy may be
+    # stale relative to the waiter's terminal stamp.
+    fresh = waitstore.load(wid)
+    if fresh and fresh.get("status") in waitstore.TERMINAL:
+        log(
+            f"resume {wid}: skipped — operation already terminal "
+            f"({fresh['status']}); delivery worker never overrides "
+            f"operation truth"
+        )
+        return
+    rec = fresh or rec
+    msg = continuation_msg(rec)
+
+    # B: busy-session guard before touching the lease
+    strays = stray_worker_pids(sid)
+    if strays:
+        rec["status"] = "READY_TO_RESUME"
+        rec["resume_note"] = f"queued: target session busy (stray pids {strays})"
+        waitstore.save(rec)
+        log(f"queued {wid}: session busy {strays}")
+        return
+
+    # pre-register the lease owner as THIS daemon so acquisition is atomic
+    lease = waitstore.acquire_lease(sid, wid, os.getpid(), os.getpid())
+    if lease is None:
+        live = waitstore.read_lease(sid)
+        rec["status"] = "READY_TO_RESUME"
+        rec["resume_note"] = f"queued: lease held by wait {live and live.get('wait_id')}"
+        waitstore.save(rec)
+        log(f"queued {wid}: lease held")
+        return
+
+    rec["status"] = "RESUMING"
+    rec["resume_attempts"] = rec.get("resume_attempts", 0) + 1
+    waitstore.save(rec)
+
+    try:
+        proc = subprocess.Popen(
+            resume_cmd(sid, msg, wid, rec.get("resume_cmd")),
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            text=True,
+            start_new_session=True,
+        )
+    except Exception as e:
+        waitstore.release_lease(sid, wid)
+        rec["status"] = "READY_TO_RETRY"  # spawn failed pre-delivery → retryable
+        rec["resume_note"] = f"spawn failed: {e}"
+        waitstore.save(rec)
+        log(f"spawn {wid} failed pre-delivery: {e}")
+        return
+
+    pgid = proc.pid  # start_new_session ⇒ pgid == child pid
+    waitstore.update_lease(sid, wait_id=wid, pid=proc.pid, pgid=pgid, pid_starttime=waitstore._pid_starttime(proc.pid))
+    log(f"launched {wid}: pid={proc.pid} pgid={pgid} attempt={rec['resume_attempts']}")
+
+    timed_out = False
+    try:
+        stdout, stderr = proc.communicate(timeout=WATCH_SECS)
+        rc = proc.returncode
+    except subprocess.TimeoutExpired:
+        timed_out = True
+        survivors = terminate_tree(pgid)
+        try:
+            stdout, stderr = proc.communicate(timeout=30)
+        except Exception:
+            stdout, stderr = "", ""
+        rc = proc.returncode
+        if survivors:
+            log(f"ERROR {wid}: group {pgid} survivors after KILL: {survivors}")
+
+    waitstore.release_lease(sid, wid)
+
+    if timed_out:
+        # D+E: timeout is NOT success. Tree terminated; retry truthfully.
+        if rec["resume_attempts"] >= MAX_ATTEMPTS:
+            rec["status"] = "RESUME_UNVERIFIED"
+            rec["resume_note"] = (
+                f"watch timeout x{rec['resume_attempts']}; process group {pgid} terminated; attempts exhausted"
+            )
+        else:
+            rec["status"] = "READY_TO_RETRY"
+            rec["resume_note"] = f"watch timeout; process group {pgid} terminated (TERM→KILL, verified); will retry"
+        log(f"timeout {wid}: tree {pgid} terminated, rc={rc}")
+    elif rc == 0:
+        # COMPLETION TRUTH (constraint 4): rc==0 means the headless worker
+        # EXITED CLEANLY — the strongest terminal evidence the supported CLI
+        # provides. Timeout/cancel can never reach this branch. The evidence
+        # chain (rc, output tail, timing) is recorded on the wait record.
+        terminal = "COMPLETED" if rec.get("verdict") == "success" else "FAILED"
+        rec["status"] = terminal
+        rec["resume_evidence"] = {
+            "kind": "worker_exited_rc0",
+            "rc": 0,
+            "output_tail": (stdout or "")[-400:],
+            "finished_at": waitstore.now(),
+        }
+        log(f"resumed {wid}: rc=0 → {terminal}")
+    else:
+        # Nonzero rc: delivery/completion unproven. Honest, not schedulable.
+        rec["status"] = "RESUME_UNVERIFIED"
+        rec["resume_evidence"] = {
+            "kind": "worker_exited_nonzero",
+            "rc": rc,
+            "output_tail": ((stdout or "") + (stderr or ""))[-400:],
+            "finished_at": waitstore.now(),
+        }
+        rec["resume_note"] = f"resume rc={rc}; delivery unverified"
+        log(f"resumed {wid}: rc={rc} → RESUME_UNVERIFIED")
+    waitstore.save(rec)
+
+
+# ── startup recovery (invariant F) ────────────────────────────────────────
+
+
+def startup_recovery() -> None:
+    for rec in waitstore.all_waits():
+        if rec.get("status") != "RESUMING":
+            continue
+        sid, wid = rec["session_id"], rec["wait_id"]
+        lease = waitstore.read_lease(sid)
+        if lease and lease.get("wait_id") == wid and waitstore._owner_alive(lease):
+            # Adopt the live worker and monitor it to a bounded conclusion.
+            pgid = lease["pgid"]
+            log(f"recover {wid}: adopted live worker pgid={pgid}")
+            deadline = time.time() + WATCH_SECS
+            while time.time() < deadline and pgid_members(pgid):
+                time.sleep(2)
+            survivors = terminate_tree(pgid) if pgid_members(pgid) else []
+            waitstore.release_lease(sid, wid)
+            rec["status"] = "RESUME_UNVERIFIED"
+            rec["resume_note"] = "adopted post-restart; outcome unverifiable" + (
+                f"; survivors killed: {survivors}" if survivors else ""
+            )
+            waitstore.save(rec)
+            log(f"recover {wid}: adopted worker ended → RESUME_UNVERIFIED")
+        else:
+            if lease:
+                waitstore.release_lease(sid, wid)  # stale
+            rec["status"] = "RESUME_UNVERIFIED"
+            rec["resume_note"] = "recovered at daemon startup: no live owner; delivery unverifiable"
+            waitstore.save(rec)
+            log(f"recover {wid}: no live owner → RESUME_UNVERIFIED")
+
+
+# ── reconciler + drive loop (invariant G: deterministic queue) ────────────
+
+_inflight: dict[str, threading.Thread] = {}
+_inflight_lock = threading.Lock()
+
+
+def drive_ready() -> None:
+    for rec in waitstore.schedulable():
+        sid = rec["session_id"]
+        # CONTENTION POLICY (constraint 2): MiMo exposes NO supported
+        # busy/idle session signal (verified 2026-09-12: `mimo session list
+        # --format json` has no status field; no `session status` command;
+        # no status column in the session schema; no documented ACP/serve
+        # endpoint). Therefore automatic injection into interactive sessions
+        # is NOT YET SAFE and stays disabled. Only waits explicitly declared
+        # target_kind="headless" (disposable/owned sessions) auto-dispatch.
+        if rec.get("target_kind") != "headless":
+            if not rec.get("hold_reason"):
+                rec["hold_reason"] = (
+                    "INTERACTIVE_AUTO_RESUME = NOT YET SAFE: "
+                    "no supported MiMo busy/idle signal; "
+                    "queued until operator releases"
+                )
+                waitstore.save(rec)
+            continue
+        if rec.get("ext", {}).get("resume_mode") == "self":
+            # REAL_ASYNC_BRIDGE §3: self-resuming wait (completion-file /
+            # inotify waiter owns its continuation). The daemon has NO
+            # delivery role here — dispatching a delivery worker would
+            # stamp a phantom rc!=0 over the waiter's own truthful
+            # terminal state (the canonical-state contradiction).
+            # NB: keyed on the explicit self marker, NOT on resume_cmd
+            # absence — records without inline resume_cmd may still be
+            # dispatchable via the WR_RESUME_CMD env template.
+            continue
+        with _inflight_lock:
+            t = _inflight.get(sid)
+            if t is not None and t.is_alive():
+                continue  # single-flight: session already being driven
+            # B: never launch into a busy session
+            strays = stray_worker_pids(sid)
+            if strays:
+                log(f"drive: {rec['wait_id']} held, session busy {strays}")
+                continue
+            th = threading.Thread(target=resume, args=(rec,), daemon=True, name=f"resume-{rec['wait_id']}")
+            _inflight[sid] = th
+            # start INSIDE the lock: a just-created thread is not yet
+            # is_alive() until start() runs — starting outside the lock let
+            # the same drive cycle dispatch a second worker for the session
+            # (single-flight race, caught by the T2 invariant test).
+            th.start()
+        log(f"drive: dispatched {rec['wait_id']} (session {sid})")
+
+
+def reconcile_once() -> None:
+    for rec in waitstore.pending():
+        if rec["status"] != "WAITING_EXTERNAL":
+            continue
+        created = rec.get("created_at", "")
+        try:
+            import calendar
+
+            age = time.time() - calendar.timegm(time.strptime(created, "%Y-%m-%dT%H:%M:%SZ"))
+        except ValueError:
+            continue
+        if age > RECONCILE_SECS:
+            try_transition(rec)
+
+
+_tick = 0
+
+
+def orphan_sweep() -> int:
+    """REAL_ASYNC_BRIDGE §4: permanent mid-run orphan-lease recovery.
+
+    FAILURE DETECTION ONLY — never involved in normal completion
+    scheduling. A wait stuck in RESUMING past ORPHAN_GRACE_SECS means the
+    resume flow died between lease acquire and release (worker thread
+    lost, exception mid-watch). Reconcile exactly as startup_recovery
+    would have: release the lease, stamp RESUME_UNVERIFIED truthfully
+    (delivery unverifiable), never auto-relaunch.
+    """
+    import calendar as _cal
+
+    healed = 0
+    for rec in waitstore.all_waits():
+        if rec.get("status") != "RESUMING":
+            continue
+        try:
+            updated = _cal.timegm(time.strptime(rec.get("last_updated", ""), "%Y-%m-%dT%H:%M:%SZ"))
+        except ValueError:
+            continue
+        if time.time() - updated < ORPHAN_GRACE_SECS:
+            continue  # legitimate in-flight watch (up to WATCH_SECS)
+        wid, sid = rec["wait_id"], rec["session_id"]
+        lease = waitstore.read_lease(sid)
+        owner_dead = lease is not None and not waitstore._owner_alive(lease)
+        daemon_owned = lease is not None and lease.get("pid") == os.getpid()
+        if lease is None or owner_dead or daemon_owned:
+            if lease is not None:
+                waitstore.release_lease(sid, wid)
+            if rec.get("resume_attempts", 0) >= MAX_ATTEMPTS:
+                rec["status"] = "RESUME_UNVERIFIED"
+            else:
+                rec["status"] = "READY_TO_RETRY"
+            rec["resume_note"] = (
+                "orphaned mid-run: resume flow died between lease acquire "
+                "and delivery ("
+                + ("lease owner dead" if owner_dead else "resume thread lost" if daemon_owned else "no lease")
+                + "); orphan sweep recovered"
+            )
+            waitstore.save(rec)
+            log(f"orphan-sweep {wid}: {rec['status']} — {rec['resume_note']}")
+            healed += 1
+    return healed
+
+
+def reconciler() -> None:
+    global _tick
+    while True:
+        # v0.3: the local-runner wake is the fast path; periodic
+        # authoritative verification demotes to a slower fallback
+        # (missed-event + daemon-restart recovery), not API polling.
+        _tick += 1
+        if _tick % RECONCILE_EVERY_TICKS == 0:
+            try:
+                reconcile_once()
+            except Exception as e:
+                log(f"reconcile error: {e}")
+            try:
+                orphan_sweep()  # §4 failure detection — bounded, periodic
+            except Exception as e:
+                log(f"orphan-sweep error: {e}")
+        try:
+            drive_ready()
+        except Exception as e:
+            log(f"drive error: {e}")
+        with _inflight_lock:
+            for sid, t in list(_inflight.items()):
+                if not t.is_alive():
+                    del _inflight[sid]
+        time.sleep(DRIVE_LOOP_SECS)
+
+
+# ── webhook receiver (wake signal only) ───────────────────────────────────
+
+
+def wake_sweep(event: dict, eid: str) -> dict:
+    """Sweep WAITING_EXTERNAL waits through authoritative verification.
+
+    Runs on a background thread (the hook's HTTP response returns
+    instantly, minimising the runner's offline window). If any verified
+    wait got 'pending' — the normal case for a job that finished
+    microseconds ago: the post-task hook fires after results are reported
+    but before the runner's final task acknowledgement flips server-side
+    status — retry on a BOUNDED event-triggered schedule (default 4s x3),
+    never a polling loop. The final ack lands milliseconds after the hook
+    exits, so the first retry almost always sees the terminal state.
+    """
+    arrival = time.time()
+    log(f"local wake {eid}: {json.dumps(event)[:140]}")
+    import calendar
+
+    def sweep() -> tuple[int, int, int]:
+        verified = advanced = pending = 0
+        for rec in waitstore.pending():
+            if rec["status"] != "WAITING_EXTERNAL":
+                continue
+            try:
+                age = time.time() - calendar.timegm(time.strptime(rec.get("created_at", ""), "%Y-%m-%dT%H:%M:%SZ"))
+            except ValueError:
+                continue
+            if age < WAKE_MIN_AGE_SECS:
+                continue
+            verified += 1
+            before = rec["status"]
+            try_transition(rec)
+            if rec["status"] != before and rec["status"] == "READY_TO_RESUME":
+                advanced += 1
+            elif rec["status"] == "WAITING_EXTERNAL":
+                pending += 1
+        return verified, advanced, pending
+
+    verified, advanced, pending = sweep()
+    log(
+        f"local wake {eid}: verified={verified} advanced={advanced} "
+        f"pending={pending} sweep_ms={int((time.time() - arrival) * 1000)}"
+    )
+    for attempt in range(1, WAKE_RETRIES + 1):
+        # Retry while the sweep didn't conclusively advance anything:
+        # pending verdicts (ack-window) AND fresh-wait skips (age gate).
+        # Bounded, event-triggered — not a polling loop.
+        if not pending and verified:
+            break
+        time.sleep(WAKE_RETRY_SECS)
+        verified, advanced, pending = sweep()
+        log(
+            f"local wake {eid}: retry{attempt} verified={verified} "
+            f"advanced={advanced} pending={pending} "
+            f"total_ms={int((time.time() - arrival) * 1000)}"
+        )
+    return {"event_id": eid}
+
+
+class Handler(BaseHTTPRequestHandler):
+    def do_POST(self):
+        if self.path == "/process-event":
+            # T15 §15D: process-exit events from builder/controller
+            # notifiers. Minimal dedicated endpoint — the Gitea webhook
+            # path stays untouched. Payload: {operation_id, exit_code,
+            # completed_at} + OPTIONAL provider/event_type/sha (T16
+            # §16C/§16E: the CI adapter registers as provider="gitea",
+            # event_type="ci_completed"; defaults preserve the T15 builder
+            # identity so existing notifiers are unaffected). The event is
+            # a WAKE, never truth — resumed controllers re-read canonical
+            # authority before any verdict.
+            length = int(self.headers.get("Content-Length", 0))
+            body = self.rfile.read(length)
+            try:
+                payload = json.loads(body) if body else {}
+            except json.JSONDecodeError:
+                self.send_response(400)
+                self.end_headers()
+                return
+            event = {
+                "provider": str(payload.get("provider", "builder")),
+                "event_type": str(payload.get("event_type", "builder_exit")),
+                "run_ref": str(payload.get("operation_id", "")) or None,
+                "sha": payload.get("sha"),
+                "exit_code": payload.get("exit_code"),
+                "completed_at": payload.get("completed_at") or "",
+            }
+            event = {k: v for k, v in event.items() if v is not None}
+            eid = waitstore.record_event(event)
+            log(f"process-event {eid}: {json.dumps(event)[:140]}")
+            # REAL_ASYNC_BRIDGE (R3): durable completion-file delivery.
+            # Every event reaching the bridge is written as
+            # resumes/<operation_id>.done so inotify waiters wake on the
+            # kernel event (R4) with no FIFO/lease machinery, and the
+            # event survives daemon AND waiter restarts (R9). Atomic
+            # rename; duplicate events rewrite identical content (R10).
+            _op_id = str(payload.get("operation_id", ""))
+            if _op_id:
+                try:
+                    # Stable identity only: duplicate deliveries of the
+                    # same logical event must produce byte-identical
+                    # completion files (R10 idempotence by construction).
+                    # completed_at is volatile and lives in the durable
+                    # event log; the file is a WAKE SIGNAL, never truth.
+                    _stable = {
+                        k: event.get(k)
+                        for k in ("provider", "event_type", "run_ref", "sha", "exit_code")
+                        if event.get(k) is not None
+                    }
+                    completion_file.write_completion(_op_id, _stable)
+                except Exception as e:  # noqa: BLE001 — delivery best-effort; wait-record transition + watchdog recovery still cover the lost file
+                    log(f"process-event {eid}: completion-file write failed: {e}")
+            matched = 0
+            for rec in waitstore.pending():
+                if rec["status"] == "WAITING_EXTERNAL" and waitstore.match(rec, event):
+                    matched += 1
+                    try_transition(rec, event)
+            # matched READY controller waits are dispatched IMMEDIATELY
+            # (event-driven, no floor). Guard: only never-resumed waits —
+            # a duplicate/late event must not double-dispatch (15E);
+            # single-flight lease + this guard together make resume
+            # idempotent.
+            for rec in waitstore.pending():
+                if rec["status"] == "READY_TO_RESUME" and rec.get("resume_cmd") and rec.get("resume_attempts", 0) == 0:
+                    threading.Thread(target=resume, args=(rec,), daemon=True, name=f"resume-{rec['wait_id']}").start()
+                    matched += 1
+            try:
+                self.send_response(200)
+                self.end_headers()
+                self.wfile.write(json.dumps({"event_id": eid, "matched": matched}).encode())
+            except (BrokenPipeError, ConnectionResetError):
+                pass
+            return
+        if self.path == "/wake":
+            length = int(self.headers.get("Content-Length", 0))
+            body = self.rfile.read(length)
+            try:
+                payload = json.loads(body) if body else {}
+            except json.JSONDecodeError:
+                self.send_response(400)
+                self.end_headers()
+                return
+            event = {
+                "provider": "gitea-local-runner",
+                "task_id": str(payload.get("task_id", "")) or None,
+                "run_id": str(payload.get("run_id", "")) or None,
+                "repository": str(payload.get("repository", "")) or None,
+                "job_result": str(payload.get("job_result", "")) or None,
+            }
+            event = {k: v for k, v in event.items() if v}
+            eid = waitstore.record_event(event)
+            # Respond instantly; sweep on a worker thread. The hook is
+            # fire-and-forget (curl -m 2) — a broken pipe on write is
+            # harmless and must not propagate.
+            try:
+                threading.Thread(
+                    target=wake_sweep, args=(event, eid), daemon=True, name=f"wake-{payload.get('task_id', '?')}"
+                ).start()
+            except Exception as e:
+                log(f"wake spawn error: {e}")
+            try:
+                self.send_response(200)
+                self.end_headers()
+                self.wfile.write(json.dumps({"queued": True, "event_id": eid}).encode())
+            except (BrokenPipeError, ConnectionResetError):
+                pass
+            return
+        if self.path != "/webhook":
+            self.send_response(404)
+            self.end_headers()
+            return
+        length = int(self.headers.get("Content-Length", 0))
+        body = self.rfile.read(length)
+        try:
+            payload = json.loads(body)
+        except json.JSONDecodeError:
+            self.send_response(400)
+            self.end_headers()
+            return
+        action = payload.get("action") or payload.get("event_type") or ""
+        wf = payload.get("workflow_run") or {}
+        event = {
+            "provider": "gitea",
+            "event_type": f"workflow.{action}" if wf else action,
+            "sha": (wf.get("head_sha") or payload.get("sha") or "")[:12] or None,
+            "run_ref": str(wf.get("id") or payload.get("run_number") or "") or None,
+        }
+        event = {k: v for k, v in event.items() if v}
+        eid = waitstore.record_event(event)
+        log(f"webhook event {eid}: {json.dumps(event)[:140]}")
+        matched = 0
+        for rec in waitstore.pending():
+            if rec["status"] == "WAITING_EXTERNAL" and waitstore.match(rec, event):
+                matched += 1
+                try_transition(rec)
+        self.send_response(200)
+        self.end_headers()
+        self.wfile.write(json.dumps({"event_id": eid, "matched": matched}).encode())
+
+    def do_GET(self):
+        if self.path == "/health":
+            self.send_response(200)
+            self.end_headers()
+            self.wfile.write(b'{"ok":true}')
+            return
+        self.send_response(404)
+        self.end_headers()
+
+    def log_message(self, *a):
+        pass
+
+
+def main() -> int:
+    port = int(sys.argv[1]) if len(sys.argv) > 1 else 8791
+
+    # H (constraint 1): flock is the SINGLETON AUTHORITY. daemon.pid's
+    # content is diagnostic metadata only — it never gates startup, never
+    # implies liveness, and is overwritten after the exclusive lock is held.
+    # Stale pid content (incl. SIGKILL debris) is thus harmless by design.
+    PIDFILE.parent.mkdir(parents=True, exist_ok=True)
+    lock_fd = os.open(PIDFILE, os.O_CREAT | os.O_RDWR, 0o644)
+    try:
+        fcntl.flock(lock_fd, fcntl.LOCK_EX | fcntl.LOCK_NB)
+    except OSError:
+        log("refusing start: another daemon holds the pidfile flock")
+        print("daemon already running (pidfile flock held)", file=sys.stderr)
+        return 1
+    os.ftruncate(lock_fd, 0)
+    os.write(lock_fd, f"{os.getpid()}\n".encode())  # diagnostic, lock held for life
+
+    startup_recovery()
+    t = threading.Thread(target=reconciler, daemon=True)
+    t.start()
+    log(f"daemon starting on :{port}")
+    srv = ThreadingHTTPServer(("0.0.0.0", port), Handler)
+    try:
+        srv.serve_forever()
+    except KeyboardInterrupt:
+        pass
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,246 @@
+#!/usr/bin/env python3
+"""verify.py — authoritative Gitea verification (webhook ≠ success).
+
+Queries the Gitea API for the TRUE state of the run/job/commit the wait
+references. Returns (verdict, evidence). verdict ∈ {success, failure,
+pending, unknown}. Only this module may move a wait out of
+WAITING_EXTERNAL.
+"""
+
+from __future__ import annotations
+import json, subprocess
+from pathlib import Path
+
+HOST = "https://gitea.thingscouldgetdicey.online"
+REPO = "conductor/rune"
+
+
+def _token() -> str:
+    # Canonical ladder without importing Rune: env → token file → git store
+    import os
+
+    tok = os.environ.get("GITEA_TOKEN", "").strip()
+    if tok:
+        return tok
+    tf = os.path.expanduser("~/.config/opencode/secrets/gitea-token")
+    if os.path.isfile(tf):
+        return open(tf).read().strip()
+    r = subprocess.run(
+        ["git", "credential", "fill"],
+        input="protocol=https\nhost=gitea.thingscouldgetdicey.online\nusername=conductor\n\n",
+        capture_output=True,
+        text=True,
+        timeout=10,
+    )
+    for line in r.stdout.splitlines():
+        if line.startswith("password="):
+            return line[9:]
+    raise RuntimeError("no gitea credential available")
+
+
+def _api(path: str) -> tuple[int, object]:
+    import urllib.request
+
+    req = urllib.request.Request(f"{HOST}/api/v1/repos/{REPO}{path}", headers={"Authorization": f"token {_token()}"})
+    try:
+        with urllib.request.urlopen(req, timeout=20) as r:
+            return r.status, json.loads(r.read())
+    except urllib.error.HTTPError as e:
+        return e.code, {"error": e.read().decode()[:200]}
+
+
+def _jobs_list(run_ref: str):
+    """GET jobs for a run, unwrapping the {'jobs': [...]} envelope."""
+    s, data = _api(f"/actions/runs/{run_ref}/jobs")
+    if s != 200:
+        return s, None
+    if isinstance(data, dict):
+        data = data.get("jobs", [])
+    return s, data if isinstance(data, list) else None
+
+
+def _resolve_run_id(run_ref: str, sha: str) -> str:
+    """Task-list ids are NOT run ids. If run_ref lookup 404s, resolve the
+    real run id by head SHA via /actions/runs (T15: run 6907 vs task 15743)."""
+    if run_ref:
+        s, jobs = _jobs_list(run_ref)
+        if s == 200 and jobs:
+            return run_ref
+    if sha:
+        s, runs = _api(f"/actions/runs?limit=30")
+        items = runs.get("workflow_runs", []) if isinstance(runs, dict) else (runs if isinstance(runs, list) else [])
+        # One SHA fans out to multiple runs (lint + deploy-only). Prefer a
+        # substantive (non-skipped) run; a skipped-only run proves nothing.
+        fallback = ""
+        for r in items:
+            if str(r.get("head_sha", "")).startswith(sha):
+                rid = str(r.get("id"))
+                if r.get("conclusion") != "skipped":
+                    return rid
+                fallback = fallback or rid
+        return fallback
+    return ""
+
+
+def verify_run(run_ref: str, sha: str, expected: str) -> tuple[str, dict]:
+    """run_ref = actions run id (or task id — resolved via SHA lookup).
+
+    Defect fixed 2026-09-12 (surfaced by W/R-v0.3 live proof, run 7110):
+    the jobs query here used raw _api + isinstance(jobs, list), but this
+    Gitea wraps jobs in the {"jobs": [...]} envelope — the check always
+    failed, so EVERY EXACT_WORKFLOW verdict silently fell through to the
+    aggregate commit-status fallback, where unrelated runs on the same
+    SHA (multi-run fan-out: ci.yml + deploy-only) poison the verdict.
+    Use _jobs_list (unwraps the envelope), same as _resolve_run_id."""
+    run_ref = _resolve_run_id(run_ref, sha)
+    if run_ref:
+        s, jobs = _jobs_list(run_ref)
+        if s == 200 and isinstance(jobs, list) and jobs:
+            conclusions = {j.get("name"): j.get("conclusion") for j in jobs}
+            statuses = {j.get("status") for j in jobs}
+            if statuses <= {"completed"}:
+                verdict = "success" if set(conclusions.values()) <= {"success", "skipped"} else "failure"
+                return verdict, {"run_ref": run_ref, "jobs": conclusions}
+            return "pending", {"run_ref": run_ref, "statuses": sorted(statuses)}
+    if sha:
+        s, st = _api(f"/commits/{sha}/status")
+        if s == 200 and isinstance(st, dict):
+            state = st.get("state", "unknown")  # success|error|failure|pending|warning
+            rid = _resolve_run_id("", sha)  # attach run id so failure evidence works
+            if state == "success":
+                return "success", {"sha": sha, "state": state, "run_ref": rid}
+            if state in ("error", "failure"):
+                return "failure", {"sha": sha, "state": state, "run_ref": rid}
+            return "pending", {"sha": sha, "state": state}
+    return "unknown", {"run_ref": run_ref, "sha": sha}
+
+
+def failed_job_evidence(run_ref: str, sha: str = "") -> dict:
+    """For failure continuations: exact failing job + conclusion."""
+    run_ref = _resolve_run_id(run_ref, sha)
+    if not run_ref:
+        return {}
+    s, jobs = _jobs_list(run_ref)
+    if s != 200 or not isinstance(jobs, list):
+        return {}
+    bad = [j for j in jobs if j.get("conclusion") not in ("success", "skipped", None)]
+    return {
+        "failed_jobs": [
+            {"name": j.get("name"), "conclusion": j.get("conclusion"), "url": j.get("url", "")} for j in bad
+        ]
+    }
+
+
+# ── RELEASE_DEPLOY verification (W/R v0.1, overnight handoff Part 2) ──────
+# Aggregate commit status is NOT a valid completion authority for Rune's
+# release/deploy flow: the release successor SHA, not the source SHA, is what
+# converges (observed live: wr-1789179026 stranded on pending while 0.66.15
+# was deployed). The authority chain is:
+#   source SHA → source runs terminal → release successor commit (descendant)
+#   → successor runs success → deployed/current identity converges.
+
+
+def _git(args: list[str], repo_root: str) -> str:
+    import subprocess
+
+    r = subprocess.run(["git"] + args, capture_output=True, text=True, cwd=repo_root or None, timeout=15)
+    if r.returncode != 0:
+        raise RuntimeError(f"git {' '.join(args[:2])}: {r.stderr.strip()[:120]}")
+    return r.stdout
+
+
+def _runs_for_sha(sha: str) -> list[dict]:
+    s, runs = _api("/actions/runs?limit=50")
+    items = runs.get("workflow_runs", []) if isinstance(runs, dict) else []
+    return [r for r in items if str(r.get("head_sha", "")).startswith(sha)]
+
+
+def _sha_verdict_from_runs(runs: list[dict]) -> tuple[str, dict]:
+    """success | failure | pending from a SHA's workflow runs.
+    Terminal verdict ONLY when every run is completed — mixed states
+    (some success, some still in flight) must stay PENDING (observed live:
+    a SHA with 1 completed-success + 1 queued + 1 in-progress was judged
+    success mid-flight; wr-1789182522 2026-09-12)."""
+    if not runs:
+        return "pending", {"note": "no runs found"}
+    non_terminal = [r for r in runs if r.get("status") != "completed"]
+    if non_terminal:
+        return "pending", {"in_flight": len(non_terminal)}
+    conclusions = {r.get("name"): r.get("conclusion") for r in runs}
+    bad = {n: c for n, c in conclusions.items() if c == "failure"}
+    if bad:
+        return "failure", {"failed": bad}
+    return "success", {"runs": conclusions}
+
+
+def verify_release_deploy(source_sha: str, repo_root: str = "", deployed_link: str = "") -> tuple[str, dict]:
+    """RELEASE_DEPLOY authority chain. Returns (verdict, evidence)."""
+    source_sha = (source_sha or "")[:12]
+    ev: dict = {"source_sha": source_sha, "kind": "RELEASE_DEPLOY"}
+
+    # 1. Source SHA's own runs must be terminal and healthy.
+    src_runs = _runs_for_sha(source_sha)
+    src_verdict, src_ev = _sha_verdict_from_runs(src_runs)
+    ev["source"] = src_ev
+    if src_verdict != "success":
+        return src_verdict, ev
+
+    # 2. Release successor: first chore(release) commit descending from source.
+    if not repo_root:
+        return "pending", {**ev, "note": "no repo_root — cannot check lineage"}
+    try:
+        out = _git(
+            [
+                "log",
+                "--format=%H %s",
+                "--reverse",
+                f"{source_sha}..master",
+                "--grep=^chore\\(release\\)",
+                "--extended-regexp",
+            ],
+            repo_root,
+        )
+    except RuntimeError as e:
+        return "pending", {**ev, "note": str(e)}
+    successors = [l for l in out.splitlines() if l.strip()]
+    if not successors:
+        return "pending", {**ev, "note": "no release successor yet"}
+    succ_line = successors[0]
+    succ_sha_full, _, succ_subject = succ_line.partition(" ")
+    succ_sha = succ_sha_full[:12]
+    ev["successor"] = {"sha": succ_sha, "subject": succ_subject[:80]}
+
+    # Ancestry is guaranteed by the range, but verify explicitly (deterministic
+    # lineage over title guessing):
+    try:
+        _git(["merge-base", "--is-ancestor", source_sha, succ_sha_full], repo_root)
+    except RuntimeError:
+        return "failure", {**ev, "note": "successor is not a descendant of source"}
+
+    # 3. Successor runs must be terminal and healthy.
+    succ_runs = _runs_for_sha(succ_sha)
+    succ_verdict, succ_ev = _sha_verdict_from_runs(succ_runs)
+    ev["successor"]["runs"] = succ_ev
+    if succ_verdict != "success":
+        return succ_verdict, ev
+
+    # 4. Deployed identity convergence. A mismatch is FAILURE only when the
+    # successor's runs are ALL completed — if any is still running/queued the
+    # deploy may simply not have propagated yet → PENDING (observed live:
+    # reconciler checked mid-deploy, verdict failure, deployment converged
+    # ~2 min later; wr-1789182522 2026-09-12).
+    if deployed_link:
+        link = Path(deployed_link)
+        if link.is_symlink() or link.is_dir():
+            deployed = str(link.resolve())
+            deployed_id = Path(deployed).name
+            ev["deployed"] = deployed_id
+            if not deployed_id.startswith(succ_sha[:12]):
+                still = [r for r in _runs_for_sha(succ_sha) if r.get("status") != "completed"]
+                if still:
+                    return "pending", {**ev, "note": "deploy propagating (runs in flight)"}
+                return "failure", {**ev, "note": f"deployed {deployed_id} != successor {succ_sha} (all runs terminal)"}
+        else:
+            return "failure", {**ev, "note": f"deployed link missing: {deployed_link}"}
+
+    return "success", ev
@@ -0,0 +1,337 @@
+#!/usr/bin/env python3
+"""waitstore.py — wait/resume state machine v0.2 (developer tooling, NOT Rune).
+
+States:
+  RUNNABLE → WAITING_EXTERNAL → READY_TO_RESUME → RESUMING → COMPLETED | FAILED
+                                       ↑               │
+                                       │               ├─(watch timeout, tree killed)
+                                       │               └─(rc!=0 / unverifiable)
+                                  READY_TO_RETRY ←─────┘
+  RESUME_UNVERIFIED : delivery after crash/timeout cannot be proven.
+                      TERMINAL-ish — never auto-scheduled; explicit
+                      `requeue()` required to make it READY_TO_RESUME.
+  TIMED_OUT         : external event never arrived within the wait budget.
+
+Waits persist as waits/<wait_id>.json (atomic tmp+rename).
+Leases persist as leases/<session_id>.lease.json (atomic tmp+rename,
+validated against /proc so a dead or PID-reused owner is stealable).
+"""
+
+from __future__ import annotations
+import fcntl, json, os, tempfile, time, uuid
+from pathlib import Path
+
+ROOT = Path(os.environ.get("WR_ROOT", "") or Path(__file__).resolve().parent)
+WAITS = ROOT / "waits"
+EVENTS = ROOT / "events"
+LEASES = ROOT / "leases"
+
+SCHEDULABLE = {"READY_TO_RESUME", "READY_TO_RETRY"}
+VALID_STATES = {
+    "RUNNABLE",
+    "WAITING_EXTERNAL",
+    "READY_TO_RESUME",
+    "RESUMING",
+    "COMPLETED",
+    "FAILED",
+    "READY_TO_RETRY",
+    "RESUME_UNVERIFIED",
+    "TIMED_OUT",
+}
+TERMINAL = {"COMPLETED", "FAILED", "RESUME_UNVERIFIED", "TIMED_OUT"}
+
+
+def _atomic_write(path: Path, data: dict) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    fd, tmp = tempfile.mkstemp(dir=str(path.parent), prefix=".tmp-")
+    try:
+        with os.fdopen(fd, "w") as f:
+            json.dump(data, f, indent=1, sort_keys=True)
+        os.replace(tmp, path)
+        # T17 17C: the bridge root is SHARED across users (deploy waiter
+        # runs as the runner user; daemon as nick). mkstemp creates 0600 —
+        # daemon re-saves locked the runner user out of records it must
+        # read (waitstore.load in await_resume_event; find_matching_event
+        # replay scan silently skipped them). 0644: owner writes, all read.
+        os.chmod(path, 0o644)
+    except BaseException:
+        if os.path.exists(tmp):
+            os.unlink(tmp)
+        raise
+
+
+def now() -> str:
+    return time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+
+
+def new_wait(
+    *,
+    session_id: str,
+    provider: str,
+    event_type: str,
+    repo: str,
+    sha: str = "",
+    run_ref: str = "",
+    expected: str = "success",
+    wait_kind: str = "EXACT_WORKFLOW",
+    target_kind: str = "interactive",
+    repo_root: str = "",
+    deployed_link: str = "",
+    objective: str = "",
+    waited_for: str = "",
+    verify_when_woken: str = "",
+    on_success: str = "",
+    on_failure: str = "",
+    resume_cmd: list | None = None,
+) -> dict:
+    wid = f"wr-{int(time.time())}-{uuid.uuid4().hex[:6]}"
+    rec = {
+        "wait_id": wid,
+        "session_id": session_id,
+        "created_at": now(),
+        "provider": provider,
+        "event_type": event_type,
+        "repo": repo,
+        "sha": sha,
+        "run_ref": run_ref,
+        "expected": expected,
+        # T15 §15D: PROCESS_EVENT — a wait whose authoritative verdict is
+        # the matching process-exit event itself (exit_code carried on the
+        # event), e.g. a builder/controller process completing. The event
+        # IS the authority here (the notifier wrapper observed the real
+        # process exit); no Gitea lookup applies.
+        "wait_kind": wait_kind
+        if wait_kind in ("EXACT_WORKFLOW", "RELEASE_DEPLOY", "PROCESS_EVENT")
+        else "EXACT_WORKFLOW",
+        # CONTENTION POLICY: "interactive" (default, conservative) never
+        # auto-dispatches — MiMo exposes NO supported busy/idle session
+        # signal, so injecting into a possibly-busy interactive session is
+        # NOT YET SAFE. Only explicitly declared headless/disposable
+        # targets ("headless") are auto-resumed.
+        "target_kind": target_kind if target_kind in ("headless", "interactive") else "interactive",
+        "repo_root": repo_root,
+        "deployed_link": deployed_link,
+        # T15 §15D: per-wait resume command override (placeholder template,
+        # same substitution as WR_RESUME_CMD: {session_id}/{msg}/{wait_id}).
+        # A synthetic controller session (e.g. a Scaffolding stage driver)
+        # is NOT a MiMo session — the global mimo-shaped default cannot
+        # resume it, which is exactly why fixed-sleep polling crept back
+        # in T12-T14. Per-wait override makes ANY registered continuation
+        # target resumable through the same bridge.
+        "resume_cmd": list(resume_cmd) if resume_cmd else None,
+        "status": "WAITING_EXTERNAL",
+        "last_verified": "",
+        "resume_attempts": 0,
+        "continuation": {
+            "objective": objective,
+            "waited_for": waited_for,
+            "verify_when_woken": verify_when_woken,
+            "on_success": on_success,
+            "on_failure": on_failure,
+        },
+    }
+    _atomic_write(WAITS / f"{wid}.json", rec)
+    return rec
+
+
+def load(wid: str) -> dict | None:
+    p = WAITS / f"{wid}.json"
+    return json.loads(p.read_text()) if p.is_file() else None
+
+
+def save(rec: dict) -> None:
+    assert rec.get("status") in VALID_STATES, rec.get("status")
+    rec["last_updated"] = now()
+    _atomic_write(WAITS / f"{rec['wait_id']}.json", rec)
+
+
+def all_waits() -> list[dict]:
+    out = []
+    for p in WAITS.glob("*.json"):
+        try:
+            out.append(json.loads(p.read_text()))
+        except json.JSONDecodeError:
+            continue
+    return out
+
+
+def pending() -> list[dict]:
+    return [r for r in all_waits() if r.get("status") in {"WAITING_EXTERNAL"} | SCHEDULABLE]
+
+
+def schedulable() -> list[dict]:
+    """READY waits, oldest first — deterministic queue order."""
+    rs = [r for r in all_waits() if r.get("status") in SCHEDULABLE]
+    rs.sort(key=lambda r: (r.get("created_at", ""), r.get("wait_id", "")))
+    return rs
+
+
+def requeue(wid: str, note: str) -> dict | None:
+    """Operator-explicit requeue of a RESUME_UNVERIFIED/RESUME_FAILED wait."""
+    rec = load(wid)
+    if not rec:
+        return None
+    rec["status"] = "READY_TO_RESUME"
+    rec["resume_note"] = note
+    save(rec)
+    return rec
+
+
+# ── events (wake signals, idempotent) ─────────────────────────────────────
+
+
+def record_event(event: dict) -> str:
+    import hashlib
+
+    body = json.dumps(event, sort_keys=True)
+    eid = hashlib.sha256(body.encode()).hexdigest()[:16]
+    f = EVENTS / f"{eid}.json"
+    if not f.is_file():
+        _atomic_write(f, {"event_id": eid, "received_at": now(), "event": event})
+    return eid
+
+
+def match(rec: dict, event: dict) -> bool:
+    if rec.get("provider") != event.get("provider"):
+        return False
+    if rec.get("event_type") and event.get("event_type") and rec["event_type"] != event["event_type"]:
+        return False
+    if rec.get("sha") and event.get("sha") and rec["sha"] != event["sha"]:
+        return False
+    if rec.get("run_ref") and event.get("run_ref") and rec["run_ref"] != event["run_ref"]:
+        return False
+    return True
+
+
+# ── leases (per-session single-flight, /proc-validated) ───────────────────
+#
+# Invariant 3 (PID reuse): liveness is NEVER "PID exists". The lease records
+# pid + pgid + process starttime (/proc/<pid>/stat field 22); all three must
+# match or the owner is a different process wearing a recycled PID.
+#
+# Invariant 1 (flock as authority): the lease file's flock serializes
+# read-modify-write across processes and dies with its holder, so a crashed
+# daemon can never wedge the lock. The lease JSON + /proc decide who OWNS the
+# session; the flock only prevents racing acquirers from double-stealing.
+
+_lease_fds: dict[str, int] = {}  # session_id → open fd holding the flock
+
+
+def _pid_starttime(pid: int) -> str | None:
+    """Field 22 of /proc/<pid>/stat — distinguishes PID reuse."""
+    try:
+        stat = Path(f"/proc/{pid}/stat").read_text()
+        return stat.rsplit(")", 1)[1].split()[19]
+    except (OSError, IndexError):
+        return None
+
+
+def _owner_alive(lease: dict) -> bool:
+    pid = lease.get("pid")
+    if not pid or not Path(f"/proc/{pid}").is_dir():
+        return False
+    if _pid_starttime(pid) != lease.get("pid_starttime"):
+        return False  # PID reused by an unrelated process
+    try:
+        if os.getpgid(pid) != lease.get("pgid"):
+            return False
+    except OSError:
+        return False
+    return True
+
+
+def lease_path(session_id: str) -> Path:
+    # T16 §16C finding (observed live in daemon.log): session ids like
+    # "controller-<op>-Verboo Code/glm-5.3-flash" contain "/" — the raw
+    # embed made lease creation raise FileNotFoundError, so resume
+    # delivery for those waits could never succeed. Sanitize to a flat
+    # filesystem-safe name; the hash suffix prevents collisions between
+    # ids that sanitize identically.
+    import hashlib as _hashlib
+    import re as _re
+
+    safe = _re.sub(r"[^A-Za-z0-9._-]+", "_", session_id)[:80]
+    digest = _hashlib.sha256(session_id.encode()).hexdigest()[:10]
+    return LEASES / f"{safe}-{digest}.lease.json"
+
+
+def read_lease(session_id: str) -> dict | None:
+    p = lease_path(session_id)
+    if not p.is_file():
+        return None
+    try:
+        return json.loads(p.read_text())
+    except json.JSONDecodeError:
+        return None
+
+
+def _locked_lease_fd(session_id: str) -> int | None:
+    """Open the lease file and take its flock (non-blocking). The fd stays
+    open (lock held) until release_lease closes it."""
+    if session_id in _lease_fds:
+        return _lease_fds[session_id]  # already hold it
+    LEASES.mkdir(parents=True, exist_ok=True)
+    fd = os.open(lease_path(session_id), os.O_CREAT | os.O_RDWR, 0o600)
+    try:
+        fcntl.flock(fd, fcntl.LOCK_EX | fcntl.LOCK_NB)
+    except OSError:
+        os.close(fd)
+        return None  # another process holds the lock — mid-acquire
+    _lease_fds[session_id] = fd
+    return fd
+
+
+def _write_lease_locked(session_id: str, lease: dict) -> None:
+    fd = _lease_fds[session_id]
+    os.ftruncate(fd, 0)
+    os.lseek(fd, 0, os.SEEK_SET)
+    os.write(fd, json.dumps(lease, indent=1, sort_keys=True).encode())
+    os.fsync(fd)
+
+
+def acquire_lease(session_id: str, wait_id: str, pid: int, pgid: int) -> dict | None:
+    """Atomically create or steal the session lease. Returns the lease dict,
+    or None if a live owner holds it (or another process is mid-steal).
+    Staleness is decided by /proc identity evidence, never by age alone."""
+    fd = _locked_lease_fd(session_id)
+    if fd is None:
+        return None
+    existing = read_lease(session_id)
+    if existing and _owner_alive(existing):
+        _lease_fds.pop(session_id, None)
+        os.close(fd)
+        return None  # live worker owns this session
+    lease = {
+        "session_id": session_id,
+        "wait_id": wait_id,
+        "pid": pid,
+        "pgid": pgid,
+        "pid_starttime": _pid_starttime(pid),
+        "acquired_at": now(),
+        "state": "RESUMING",
+    }
+    _write_lease_locked(session_id, lease)
+    return lease
+
+
+def update_lease(session_id: str, **fields) -> None:
+    lease = read_lease(session_id) or {}
+    lease.update(fields)
+    _write_lease_locked(session_id, lease)
+
+
+def release_lease(session_id: str, wait_id: str) -> bool:
+    """Release only if the current lease belongs to this wait (no stealing
+    a successor's lease during a slow shutdown path). Closes the flock fd."""
+    lease = read_lease(session_id)
+    if not (lease and lease.get("wait_id") == wait_id):
+        return False
+    fd = _lease_fds.pop(session_id, None)
+    if fd is not None:
+        try:
+            fcntl.flock(fd, fcntl.LOCK_UN)
+        except OSError:
+            pass
+        os.close(fd)
+    lease_path(session_id).unlink(missing_ok=True)
+    return True
@@ -0,0 +1,56 @@
+import type { Hooks } from "@mimo-ai/plugin"
+
+// event-wait-guard — RETIRED (T21.5 §6/§8, 2026-09-17).
+//
+// HISTORY: this hook was the T21 §21B-3 runtime interception layer that
+// blocked model-generated sleep→poll continuation (T20 regression), and
+// later carried the T21.1 §21.1C cwd guard. Both were DOWNSTREAM patches:
+// they corrected the model's output instead of fixing the generator.
+//
+// ROOT-CAUSE FIX (T21.5 §2-§5): the generator was ci_event_wait's
+// wait loop returning nonterminal control to the model — caused by a
+// single hard-coded master-deploy authority that was blind to branch/PR
+// operations. T21.5 fixed this upstream:
+//   - OPERATION_SPECS: per-operation-type completion authority
+//     (branch_ci/pull_request_checks -> own Actions runs;
+//     master_ci/deploy/release -> deploy qualification);
+//   - SINGLE LOGICAL AWAIT: all nonterminal conditions re-arm inside
+//     ci_event_wait.run(); the model sees only terminal results;
+//   - structured PendingOperation envelopes (op_type, authority,
+//     internal_rearm_count) propagate to the controller.
+//
+// With the generator fixed, runtime interception is retired per the
+// tranche rule FIXED UPSTREAM BEHAVIOUR > DOWNSTREAM GUARD CORRECTION.
+// Regression prevention lives in the test suite:
+//   tests/test_single_logical_await.py  (no model re-arm contract)
+//   tests/test_ci_event_wait.py         (watchdog=recovery, event=wake)
+//   tests/test_wait_guard_corpus.py     (canonical wait-pattern corpus)
+// NOT as a permanent runtime block. The hook file stays installed as a
+// documented no-op so canonical-source/install parity (test_hook_drift,
+// test_hook_parity) continues to prove the installed runtime matches
+// this repository's canonical state.
+
+export const EventWaitGuard: Hooks = {
+  "tool.execute.before": async (_input, _output) => {
+    // RETIRED — no interception. See history above.
+    return
+  },
+}
+
+export const CwdGuard: Hooks = {
+  "tool.execute.before": async (_input, _output) => {
+    // RETIRED (T21.5 §8) — the harness/construction execution boundary
+    // now resolves project context upstream (execution-context binding).
+    // No runtime refusal. See tests/test_auto_execution_context.py.
+    return
+  },
+}
+
+export const CombinedGuards: Hooks = {
+  "tool.execute.before": async (input, output) => {
+    // Kept as the default export shape; both sub-guards are retired.
+    return
+  },
+}
+
+export default CombinedGuards
@@ -0,0 +1,92 @@
+import { tool } from "@mimo-ai/plugin"
+import { execSync } from "child_process"
+
+export default tool({
+    description: "Wait for an external operation (CI, deploy, release) to complete. Call this when you cannot make progress until an external process finishes. You do NOT need to know the sha, op_type, or duration — Rune's controller resolves those from the bound PendingOperation. Just call this when you want to wait.",
+    args: {
+        operation_id: tool.schema.string().optional().describe("Optional: the operation to wait for. If omitted, the controller resolves the current pending operation automatically."),
+        task_id: tool.schema.string().optional().describe("Optional: the active task id (e.g. T4). When provided, resolution is scoped to operations bound to this task — stale global operations are never consulted."),
+        sha: tool.schema.string().optional().describe("OPTIONAL HINT — non-authoritative. The controller uses this only to help resolve which operation you mean."),
+        op_type: tool.schema.string().optional().describe("OPTIONAL HINT — non-authoritative. The controller derives the canonical operation type from repository/Gitea state."),
+        deadline_secs: tool.schema.number().optional().describe("OPTIONAL HINT — non-authoritative. The controller applies governed policy timing. Real user deadlines must be stated in conversation."),
+    },
+    async execute(args) {
+        // INTENT NORMALIZATION — model output is intent; the controller
+        // owns execution semantics. All model args are hints recorded for
+        // observability; canonical truth comes from the bound operation.
+        const parts = []
+        if (args.operation_id) parts.push(`--operation-id ${args.operation_id}`)
+        if (args.task_id) parts.push(`--task-id ${args.task_id}`)
+        if (args.sha) parts.push(`--sha ${args.sha}`)
+        if (args.op_type) parts.push(`--model-op-type ${args.op_type}`)
+        if (args.deadline_secs) parts.push(`--model-deadline ${args.deadline_secs}`)
+        const normCmd = `python3 /mnt/workdrive/issue161-harness/scripts/await_intent_cli.py normalize ${parts.join(" ")}`
+        let norm: any
+        try {
+            const normOut = execSync(normCmd, {
+                cwd: "/mnt/workdrive/issue161-harness",
+                encoding: "utf-8" as const,
+                timeout: 120000,
+                env: { ...process.env, RUNE_WR_ROOT: "/mnt/workdrive/wr-bridge" },
+            })
+            norm = JSON.parse(normOut.trim().split("\n").find((l: string) => l.startsWith("{")) || "{}")
+        } catch (e: any) {
+            return JSON.stringify({ ok: false, stage: "normalization", error: (e.stderr || e.message || "?").slice(0, 300) })
+        }
+        if (norm.normalized === "AMBIGUOUS") {
+            return JSON.stringify({ ok: false, stage: "normalization", ...norm })
+        }
+        if (!norm.normalized) {
+            return JSON.stringify({ ok: false, stage: "normalization", ...norm })
+        }
+        // canonical continuation: stored/derived op_type + POLICY deadline
+        // (the sha for the wait comes from the normalization resolution,
+        // NOT from the model's args)
+        const waitSha = norm.sha_for_wait || args.sha || ""
+        if (!waitSha) {
+            return JSON.stringify({ ok: false, stage: "resolution", error: "no sha resolved for the pending operation", norm })
+        }
+        const cmd = `python3 /mnt/workdrive/issue161-harness/scripts/ci_event_wait.py --sha ${waitSha} --deadline ${norm.deadline_secs} --op-type ${norm.op_type}`
+        try {
+            const result = execSync(cmd, {
+                cwd: "/mnt/workdrive/issue161-harness",
+                timeout: (norm.deadline_secs + 60) * 1000,
+                encoding: "utf-8" as const,
+                env: { ...process.env, DEPLOY_SHA: waitSha, RUNE_WR_ROOT: "/mnt/workdrive/wr-bridge", RUNE_WR_CODE: "/home/nick/.local/share/mimo-wait-resume" },
+            })
+            const jl = result.trim().split("\n").find((l: string) => l.startsWith("{"))
+            const envelope = jl ? JSON.parse(jl) : { ok: false, error: "no output" }
+            return JSON.stringify({
+                ...envelope,
+                normalization: {
+                    model_deadline_ignored: norm.model_deadline_ignored,
+                    model_op_type_ignored: norm.model_op_type_ignored,
+                    model_sha_ignored: !!args.sha && args.sha !== waitSha,
+                    canonical_op_type: norm.op_type,
+                    canonical_authority: norm.authority,
+                    deadline_provenance: norm.deadline_provenance,
+                    matched_by: norm.matched_by || norm.evidence || "auto-resolved",
+                },
+            })
+        } catch (e: any) {
+            const lines = (e.stdout || "").trim().split("\n")
+            const jl = lines.find((l: string) => l.startsWith("{"))
+            if (jl) {
+                const envelope = JSON.parse(jl)
+                return JSON.stringify({
+                    ...envelope,
+                    normalization: {
+                        model_deadline_ignored: norm.model_deadline_ignored,
+                        model_op_type_ignored: norm.model_op_type_ignored,
+                        model_sha_ignored: !!args.sha && args.sha !== waitSha,
+                        canonical_op_type: norm.op_type,
+                        canonical_authority: norm.authority,
+                        deadline_provenance: norm.deadline_provenance,
+                        matched_by: norm.matched_by || norm.evidence || "auto-resolved",
+                    },
+                })
+            }
+            return JSON.stringify({ ok: false, stage: "wait", error: (e.stderr || e.message || "?").slice(0, 300) })
+        }
+    },
+})
@@ -0,0 +1,167 @@
+
+// bash — executes a shell command. PROCESS-OWNERSHIP CONTRACT (v2):
+//
+//   MODE A (tool-owned): the child is launched in its own process group;
+//   stdout/stderr are CAPTURED pipes (never the TUI tty). On timeout or
+//   failure the ENTIRE process group is signalled (SIGTERM, grace,
+//   SIGKILL) so grandchildren cannot survive the invocation, then the
+//   result is finalized. After finalization NO descendant retains a
+//   writer into MiMo's streams.
+//
+//   MODE B (explicitly detached): a command suffixed with " &" that the
+//   user/model intends to outlive the tool (nohup/setsid/disown-style)
+//   is executed via the detached helper: stdin=/dev/null, stdout/stderr
+//   -> durable log file, own session (setsid), no MiMo PTY inheritance.
+//   The tool returns the log path as the durable handle.
+//
+// Wait-intent compilation (sleep/check/poll aimed at a pending operation)
+// is preserved: interception happens at the SEMANTIC level via
+// await_intent route-bash, and a NO-OP for anything that is not a genuine
+// continuation of a live pending operation. Arbitrary shell syntax
+// (quotes, pipes, semicolons) passes through untouched — the router
+// response is only trusted when it carries normalized=true AND
+// ContinuePendingOperation with a concrete operation bound.
+import { tool } from "@mimo-ai/plugin"
+import { spawn, execFile } from "child_process"
+
+const REPO = "/mnt/workdrive/issue161-harness"
+const WR_ROOT = "/mnt/workdrive/wr-bridge"
+
+// ── MODE A: process-group-owned capture execution ──────────────────────
+interface OwnedResult { stdout: string; stderr: string; status: number; timedOut: boolean }
+
+function runOwned(command: string, timeoutMs: number): Promise<OwnedResult> {
+    return new Promise((resolve) => {
+        // /bin/sh -c "command" in its OWN process group (detached:true makes
+        // the child a session/group leader we can signal as a WHOLE).
+        const child = spawn("/bin/sh", ["-c", command], {
+            cwd: REPO,
+            stdio: ["ignore", "pipe", "pipe"],   // NEVER inherit the TUI tty
+            env: process.env,
+            detached: true,                      // own process group
+        })
+        let stdout = "", stderr = ""
+        let timedOut = false
+        let settled = false
+        const outChunks: Buffer[] = []
+        const errChunks: Buffer[] = []
+        child.stdout!.on("data", (d) => outChunks.push(d))
+        child.stderr!.on("data", (d) => errChunks.push(d))
+
+        const killGroup = (sig: NodeJS.Signals) => {
+            try { process.kill(-child.pid!, sig) } catch { /* already gone */ }
+        }
+        const finalize = (status: number) => {
+            if (settled) return
+            settled = true
+            clearTimeout(timer)
+            // detach listeners, then resolve — no late listeners remain
+            child.stdout!.removeAllListeners()
+            child.stderr!.removeAllListeners()
+            resolve({ stdout: Buffer.concat(outChunks).toString(), stderr: Buffer.concat(errChunks).toString(), status, timedOut })
+        }
+        const timer = setTimeout(() => {
+            timedOut = true
+            killGroup("SIGTERM")
+            // bounded grace for clean shutdown, then SIGKILL the group
+            setTimeout(() => killGroup("SIGKILL"), 2000)
+            // the exit/close event settles; a hard stop-guard settles anyway
+            setTimeout(() => finalize(124), 4000)
+        }, timeoutMs)
+
+        child.on("error", (e) => finalize(-1))
+        child.on("close", (code) => finalize(code ?? -1))
+    })
+}
+
+// ── MODE B: explicit detachment for "cmd &" invocations ────────────────
+function runDetached(command: string): Promise<string> {
+    return new Promise((resolve) => {
+        const logPath = `/tmp/mimo-detached-${Date.now()}-${Math.floor(Math.random() * 1e6)}.log`
+        const out = `exec >>"${logPath}" 2>&1; exec 0</dev/null; ${command}`
+        const child = spawn("/bin/sh", ["-c", `setsid /bin/sh -c '${out.replace(/'/g, `'\\''`)}'`], {
+            cwd: REPO,
+            stdio: "ignore",     // child rebinds its own fds — no PTY inheritance
+            env: process.env,
+            detached: true,
+            // unref: MiMo's event loop does not wait for it
+        })
+        child.unref()
+        resolve(JSON.stringify({
+            detached: true,
+            log: logPath,
+            note: "process detached (setsid, own fds); output streams to the log file only; fd 0/1/2 do NOT reference the MiMo tty",
+        }))
+    })
+}
+
+// ── wait-intent routing (semantic level, preserved from v1) ────────────
+async function tryCompileWaitIntent(command: string): Promise<string | null> {
+    let routeOut: string
+    try {
+        routeOut = await new Promise<string>((resolve, reject) => {
+            execFile("python3", [`${REPO}/scripts/await_intent_cli.py`, "route-bash", "--command", command], {
+                cwd: REPO, timeout: 60000, encoding: "utf-8",
+                env: { ...process.env, RUNE_WR_ROOT: WR_ROOT },
+                maxBuffer: 8 * 1024 * 1024,
+            }, (err, stdout) => { if (err && !stdout) reject(err); else resolve(stdout || "") })
+        })
+    } catch {
+        return null // router unavailable: normal execution (never block on routing)
+    }
+    const jl = routeOut.trim().split("\n").find((l) => l.startsWith("{"))
+    if (!jl) return null
+    let norm: any
+    try { norm = JSON.parse(jl) } catch { return null }
+    // STRICT gate: compile ONLY a normalized, concrete continuation.
+    if (!(norm.normalized === true &&
+          norm.action === "ContinuePendingOperation" &&
+          norm.operation_id && norm.sha_for_wait)) return null
+    const waitCmd = `python3 ${REPO}/scripts/ci_event_wait.py --sha ${norm.sha_for_wait} --deadline ${norm.deadline_secs} --op-type ${norm.op_type}`
+    const r = await runOwned(waitCmd, (norm.deadline_secs + 60) * 1000)
+    const jline = r.stdout.trim().split("\n").find((l) => l.startsWith("{"))
+    if (jline) {
+        return JSON.stringify({
+            ...JSON.parse(jline),
+            intent_compilation: {
+                raw_command_kind: norm.raw_command_kind || "sleep_poll",
+                literal_sleep_executed: false,
+                operation_id: norm.operation_id,
+                canonical_op_type: norm.op_type,
+                authority: norm.authority,
+            },
+        })
+    }
+    return JSON.stringify({ ok: false, stage: "compiled-wait", error: (r.stderr || "no output").slice(0, 300) })
+}
+
+export default tool({
+    description: "Run a shell command. If the command expresses external wait intent (sleep/check/poll aimed at CI, deploy, or release operations with pending operations), it is automatically compiled to Rune's event-driven continuation — you do not need to manage wait mechanics. Plain local commands execute normally. Commands ending in ' &' run detached with output logged to a file.",
+    args: {
+        command: tool.schema.string().describe("Shell command to run"),
+    },
+    async execute(args) {
+        // 1. wait-intent compilation (semantic, strict-gated)
+        try {
+            const compiled = await tryCompileWaitIntent(args.command)
+            if (compiled !== null) return compiled
+        } catch { /* fall through to normal execution */ }
+
+        // 2. MODE B: explicit detachment for backgrounded invocations
+        const cmd = args.command
+        if (/\b(nohup|setsid)\b/.test(cmd) || /&\s*$/.test(cmd)) {
+            return await runDetached(cmd)
+        }
+
+        // 3. MODE A: tool-owned, group-killed, captured execution
+        const r = await runOwned(cmd, 120000)
+        const text = r.stdout + (r.stderr ? `\n[stderr]\n${r.stderr}` : "")
+        if (r.timedOut) {
+            return `exit timed-out (process group terminated): ${text.slice(0, 500)}`
+        }
+        if (r.status !== 0) {
+            return `exit ${r.status}: ${text.slice(0, 500) || "?"}`
+        }
+        return text || "(no output)"
+    },
+})
@@ -1,7 +1,7 @@
 {
 "_comment": "GENERATED by rune.spine \u2014 compatibility view for legacy umbrella consumers. DO NOT EDIT: structural truth lives in store/spine/current.json; edits here are flagged by spine drift.",
 "_generated_by_spine": true,
- "generated_at": "2026-09-12T16:02:18Z",
+ "generated_at": "2026-09-16T18:23:38Z",
 "umbrellas": {
  "capability": {
   "description": "Tool selection and execution authority",
@@ -12,11 +12,13 @@
    "capability_broker",
    "connection_pool",
    "evidence_probe",
+    "forge",
    "forge_client",
    "intent_continuity",
    "local_tool_runtime",
    "message_queue",
    "openai_compat_transport",
+    "provider_context",
    "recall_uplift",
    "route_risk_flags",
    "tool_executor",
@@ -43,7 +45,8 @@
    "drift_core",
    "drift_ledger",
    "drift_observability",
-    "failure_auto_filer"
+    "failure_auto_filer",
+    "programme_gates"
   ],
   "depends_on": [],
   "mirrors": [],
@@ -215,12 +218,45 @@
    "attention_capsule",
    "auto_context",
    "auto_tasks",
+    "behaviour_ledger",
+    "build_basis",
+    "capability_fit",
+    "capability_ledger",
    "cognitive_budgeter",
-    "complexity_scorer",
+    "construction_activation",
+    "construction_commands",
+    "construction_decision",
+    "construction_depth",
+    "construction_intent",
+    "construction_interface",
+    "construction_mode",
+    "construction_pattern",
+    "construction_patterns_builtin",
+    "construction_projection",
    "context_broker",
    "context_pressure",
+    "drift_resume",
+    "event_wait",
+    "failure_recovery",
+    "graph_projection",
+    "impact_preview",
+    "iteration",
+    "layout_contract",
+    "programme_continuation",
+    "project_registry",
+    "release_notes",
    "salience_gate",
+    "scaffold",
    "spine",
+    "stage_guards",
+    "structural_migration",
+    "structural_model",
+    "structural_store",
+    "testing_mode",
+    "toolchain",
+    "tui_backlog",
+    "wait_guard",
+    "wait_guard_corpus",
    "witness"
   ],
   "depends_on": [
@@ -104,15 +104,15 @@
        },
        {
          "path": "src/rune/attention_capsule.py",
-          "role": "AttentionCapsule — the typed ledger of admitted blocks the model receives (to_messages re-derives the runtime payload)",
+          "role": "AttentionCapsule \u2014 the typed ledger of admitted blocks the model receives (to_messages re-derives the runtime payload)",
          "disposition": "retain-authority",
-          "evidence": "P1-02 (#25): the capsule IS the product the model receives — the broker renders admitted blocks into it, the safety override mutates its user block in place (re-measured), runtime.run consumes capsule.to_messages(), and the trace v4 capsule_messages/capsule_content_hashes prove the logged capsule EXACTLY equals the sent payload. Legacy create_capsule/to_prompt (string-trim fake budgeting) deprecated with no live callers."
+          "evidence": "P1-02 (#25): the capsule IS the product the model receives \u2014 the broker renders admitted blocks into it, the safety override mutates its user block in place (re-measured), runtime.run consumes capsule.to_messages(), and the trace v4 capsule_messages/capsule_content_hashes prove the logged capsule EXACTLY equals the sent payload. Legacy create_capsule/to_prompt (string-trim fake budgeting) deprecated with no live callers."
        },
        {
          "path": "src/rune/inference_trace.py",
          "role": "Durable per-run inference trace (tokens served/estimated, output, latency, per-block visibility)",
          "disposition": "migrate-adapter",
-          "evidence": "P0-07 (#23): persists what ContextBroker (#19) assembled and what the transport served — builds FROM the broker trace, never re-measures; store state/inference_traces.jsonl."
+          "evidence": "P0-07 (#23): persists what ContextBroker (#19) assembled and what the transport served \u2014 builds FROM the broker trace, never re-measures; store state/inference_traces.jsonl."
        }
      ]
    },
@@ -160,13 +160,13 @@
          "path": "src/rune/capability_broker.py",
          "role": "Authoritative capability selector (select_for_job + verified floor + find_capability escape hatch)",
          "disposition": "retain-authority",
-          "evidence": "P1-01 (#24): select_for_job is the single selection authority — its ordered decision feeds LocalToolRuntime(tools=) at both construction sites and the _contributor_tools prompt block; degrade-not-crash falls back to full TOOLS with a tier-1 record_failure."
+          "evidence": "P1-01 (#24): select_for_job is the single selection authority \u2014 its ordered decision feeds LocalToolRuntime(tools=) at both construction sites and the _contributor_tools prompt block; degrade-not-crash falls back to full TOOLS with a tier-1 record_failure."
        },
        {
          "path": "memory/index/capability_cards.json",
          "role": "Capability-card index (rebuilt on empty/stale/corrupt; native cards synthesized from TOOLS metadata)",
          "disposition": "migrate-adapter",
-          "evidence": "P1-01 (#24): load_cards validates and rebuilds — the packaged empty array no longer passes as an index; cards carry risk/preconditions/side_effects copied from tool_executor.TOOLS."
+          "evidence": "P1-01 (#24): load_cards validates and rebuilds \u2014 the packaged empty array no longer passes as an index; cards carry risk/preconditions/side_effects copied from tool_executor.TOOLS."
        },
        {
          "path": "src/rune/local_tool_runtime.py",
@@ -486,12 +486,6 @@
          "disposition": "retain-authority",
          "evidence": "Defines the owner-approved canary-first Gitea sequence."
        },
-        {
-          "path": "src/rune/plugins/build_mode/__init__.py",
-          "role": "Checkpoint/ledger state-machine primitive used by current governed builds",
-          "disposition": "migrate-adapter",
-          "evidence": "Persists checkpoints but encodes an older phase order."
-        },
        {
          "path": "src/rune/build_mode.py",
          "role": "Separate BuildManager and worktree engine",
@@ -504,6 +498,9 @@
          "disposition": "archive-candidate",
          "evidence": "Reports fixed success strings for phases without executing their claimed effects."
        }
+      ],
+      "notes": [
+        "T17 17U: plugin BuildStateMachine deleted (zero importers; second parallel orchestrator)"
      ]
    },
    {
@@ -540,4 +537,4 @@
      ]
    }
  ]
-}
+}
@@ -0,0 +1,53 @@
+# Rune Semantic Release Policy (T20 §20B)
+
+Pre-1.0 convention. Scaffolding "V1" does NOT mean Rune 1.0.
+
+## Bump kinds
+
+**PATCH** (0.66.80 → 0.66.81):
+- bug fix, correction, implementation/internal maintenance
+- no meaningful new user/system capability
+
+**MINOR** (0.66.80 → 0.67.0):
+- meaningful new Rune capability
+- new subsystem functionality
+- materially expanded system behaviour
+- significant new supported workflow
+
+**MAJOR**: reserved for Rune 1.0+ compatibility/public-contract semantics.
+
+## How the machinery expresses this (no manual version edits)
+
+The existing machinery already maps commit shape → bump kind:
+- `semver_bump.next_version` (PR titles): feat→minor, else patch
+- `version_convergence.highest_warranted_bump` (push ranges):
+  feat/breaking→minor, fix/perf→patch, else None
+
+**Tranche-level rule (20B)**: a feature tranche lands its feature
+commits (`feat(...)`), and the convergence machinery bumps MINOR at
+the tranche's release boundary. Patch-only follow-ups (fix/perf)
+warrant only PATCH — the machinery cannot accidentally re-minor
+because `highest_warranted_bump` takes the max over the *post-
+boundary* range, and a fix-only range yields "patch".
+
+## Why T20 → 0.67.0
+
+T20 is a feature-level tranche (Rune-wide retrofit: structural
+taxonomy, capability/behaviour ledgers, rune graph, /testing backlog,
+Spine modernization, semantic release). The release is classified
+MINOR under this policy, earned through the governed convergence
+machinery — never a manual version edit.
+
+## T20 0.67.0 release record
+T20 classified MINOR under this policy: Rune-wide retrofit (structural
+taxonomy, capability/behaviour ledgers, rune graph, /testing backlog,
+Spine modernization, semantic release) = new subsystem functionality +
+materially expanded system behaviour. Earned through the governed
+convergence machinery (feat-bearing range → minor), never a manual edit.
+
+## Root cause record (T20, 2026-09-16)
+semver_bump re-derived the bump kind from the SYNTHESIZED title text —
+"minor: converged release" matched nothing → patch. Feature tranches
+T18/T19/T20 downgraded to 0.66.77-83. Fixed: MINOR_PREFIX_RE carries the
+convergence kind through; first push after the fix with a feat-bearing
+range lands 0.67.0 through the normal machinery.
@@ -68,3 +68,208 @@ for a 131072-window model (observed live in every PL1 Task A/B run).
  acquisition / provisional-trust lifecycle — FOUNDATIONAL ONLY (architecture
  does not block them; the projection source-list is the extension surface).
 - Deferred with reasons recorded: none blocked by Tranche 1.
+
+---
+
+# Scaffolding V1 — Tranche 2: Construction Substrate (2026-09-13)
+
+_Tranche 2 (contract SCAFFOLD-TRANCHE-001) built the construction substrate
+itself: a shared structural-store primitive, the Project Scaffold layer over
+it, and durable Construction Contracts — Rune's first build tracked by the
+system it builds._
+
+## Architecture
+
+    src/rune/structural_store.py   shared generic primitive — NO Rune constants
+            ↓                               ↓
+    src/rune/spine.py            src/rune/scaffold.py
+    (Rune Spine — canonical      (Project Scaffold — canonical
+     structural self-model)       construction/structural truth
+                                  about EXTERNAL projects)
+
+Authority map (no duplicates):
+
+| fact | authority | others |
+|---|---|---|
+| Rune's structure | `spine.py` + `store/spine/current.json` | scaffold only references |
+| external project structure | `scaffold.py` store | Spine holds bridge facts only |
+| content history / rollback | Git | both reference commits |
+| coordination (issues/PRs/CI) | Gitea | scaffold stores observations |
+| why work exists | Task/Intent | contracts reference task ids |
+
+Construction Contract = ONE durable object (`contract:<id>`, kind
+`construction_contract`) with plain `phase`/`status` fields governed by a
+small legal-transition map (draft → approved → building → verifying →
+migrating → superseding → closed; failed/abandoned terminal-or-retry). No
+workflow engine. Resumability = store durability, proven across real
+process boundaries.
+
+## Mechanics
+
+- **Repo binding** (`bind_repo`): forge/owner/name/branch/root/placement as
+  scaffold DATA. No secrets — schema rejects secret-named fields AND every
+  `save()` recursively redacts credential-named keys (defense in depth in
+  the shared core, so Spine inherits it too).
+- **Scope check** (`check_scope`): changed files vs approved scope paths →
+  green / red / **unknown** (an unscoped contract is never silently green).
+  Out-of-scope change = RED until explicitly amended; amendments recorded,
+  original RED preserved in history.
+- **Predicted vs observed**: frontier snapshot at approval; observed impact
+  captured at closure from INDEPENDENT evidence (git, tests) — never from
+  re-walking the prediction's own graph. Missed entities become CANDIDATE
+  corrections awaiting evidence.
+- **Drift**: authority-owned facts that fail re-observation go STALE —
+  never silently fresh, never force-reverted. Reality wins.
+- **Bridge** (`spine_bridge_entity`): identity + repo surface only. Project
+  internals NEVER leak into Rune Spine (test-enforced). Contract evidence
+  may cite `spine://` URIs — citations are inert on the referenced
+  authority.
+
+## Real proof (this tranche on itself)
+
+SCAFFOLD-TRANCHE-001 migrated from its pre-scaffold JSON into the new
+store, advanced through phases with evidence checkpoints, closed with
+predicted-vs-observed from git/test evidence, scope check caught the
+store's own runtime artifacts (RED → explicit amendment → green, history
+kept), and construction provenance attached into Rune Spine without
+duplicating authority. A fresh interpreter answers what/why/state/
+evidence/changed/next from the store alone — no transcripts.
+
+## Maturity (honest)
+
+| component | state |
+|---|---|
+| structural_store core | EXERCISED (unit-proven + real use) |
+| Rune Spine after refactor | LIVE_PROVEN behavior-neutral (68 tests + identical rebuild) |
+| Project Scaffold | EXERCISED |
+| Construction Contract | EXERCISED (resumability proven cross-process) |
+| repo binding (identity model) | UNIT-PROVEN (local git fixture) |
+| external-repo live proof | BLOCKED ON USER REPO SELECTION |
+| Scaffolding overall | substrate only — capability-fit discovery, gap decisions, toolchain selection, patterns, maintenance TUI all NOT BUILT (data seams exist) |
+
+---
+
+# Scaffolding Tranche 4 — Live Need → Decision → Construction Contract (2026-09-13)
+
+_The adoption tranche: a real Need now flows automatically through the
+Tranche 2/3 mechanisms and produces a governed Construction Contract. The
+model no longer has to remember the procedure — it is durable Scaffolding._
+
+## Live flow
+
+    NEED (reference to Task/Intent — pointer, never a copy)
+      ↓ normalization (gaps recorded, never invented)
+      ↓ structural discovery + fit + resolution (Spine + Project Scaffold)
+      ↓ toolchain decision (existing project toolchain wins)
+      ↓ layout obligations (USER-LOCKED authority inherited)
+      ↓ user-approval gate (material changes require approval; approved
+        bounded work never re-asks)
+      ↓ DecisionTrace (durable, model-independent)
+      ↓ generated Construction Contract → existing build/verify pipeline
+
+## Authority map (unchanged — the path CONNECTS, never duplicates)
+
+| authority | owns |
+|---|---|
+| Task/Intent | why work exists |
+| Rune Spine | Rune structural truth |
+| Project Scaffold | external-project structural/construction truth |
+| Layout Contract | desired interface constraints (USER-LOCKED = user authority) |
+| Construction Contract | governed build obligations/history |
+| Git / Gitea | history / execution |
+
+## Boundary honesty
+
+- **ADAPT/research**: the live path records a research ROUTE
+  (TARGETED_EVIDENCE_PROBE / FULL_RESEARCH_PIPELINE) when evidence is
+  insufficient — a seam, NOT an executed pipeline call. No fake
+  integration claimed.
+- **BUILD_NEW** is never a fallback: insufficient evidence blocks it;
+  normalization gaps block it; merge precedence across authorities
+  requires every consulted authority to agree.
+- **No second task engine**: one decision function + one trace object.
+
+## Contract-quality repair (Tranche 3 defects)
+
+Closure now distinguishes:
+
+    predicted_existing_affected  planned changes to existing entities
+    expected_new                 entities the build PLANNED to create
+    correctly_predicted          predicted AND independently observed
+    predicted_but_unaffected     predicted, NOT observed
+    expected_new_observed        planned creation happened
+    expected_new_missing         planned creation did NOT happen
+    genuinely_unexpected_new     observed, planned by neither list
+
+Prediction cannot self-certify observation: the observed set comes ONLY
+from caller evidence (git/tests/build/deploy). Legacy keys (missed/
+needless/observed) preserved — old contracts readable without migration.
+
+## Layout-obligation inheritance
+
+A change touching a UI component/region/implementation inherits its Layout
+Contract as a build obligation. USER-LOCKED obligations state that changing
+placement requires explicit user authorization. Unrelated backend changes
+inherit nothing. USER-LOCKED drift is never reconciled by editing the
+contract — code adapts.
+
+## Maturity (honest)
+
+- Need intake: EXERCISED (real live proof)
+- decision orchestration: LIVE-PROVEN for the bounded live flow
+- capability discovery adoption: EXERCISED
+- fit/gap adoption: EXERCISED
+- toolchain adoption: EXERCISED
+- contract generation: LIVE-PROVEN (one real generated contract)
+- contract-quality prediction: EXERCISED (repaired semantics, one live closure)
+- layout-obligation inheritance: UNIT-PROVEN (fixtures) + real TUI binding
+- Scaffolding overall: adoption path live for one bounded flow — NOT
+  general autonomous construction, NOT universal discovery/toolchain
+  intelligence, NOT Scaffolding complete.
+
+
+---
+
+# Scaffolding Tranche 9 — Release History + Evidence Correction (2026-09-13)
+
+_Gitea Releases are now an evidence-derived human-readable history of
+Rune's evolution. Plus the Tranche 8 builder-evidence correction._
+
+## Release semantics (audited)
+
+| surface | meaning |
+|---|---|
+| VERSION | semver in pyproject/init (e.g. 0.66.44) |
+| TAG | v<VERSION>, created by auto-version at the BUMP commit |
+| release commit | the auto-bump SHA — NOT necessarily final state |
+| converged deployed identity | last commit deployed under the same VERSION |
+
+Same-version convergence commits (closure/accounting/lock fixes landing and
+deploying after the bump) are normal; the notes record BOTH the tag commit
+and the final deployed identity, distinction exposed honestly.
+
+## Mechanism
+
+`release_notes.py`: deterministic evidence extraction from governed records
+(closed contracts, resolved maintenance, contract-quality records) →
+structured ReleaseNote → Markdown → Gitea API (POST create / PATCH update).
+Idempotent: publishing twice for a tag UPDATES, never duplicates. Notes are
+historical documentation — NOT Memory, NOT structural authority.
+
+## T8 evidence correction (§1)
+
+The T8 report claimed a close-from-verifying refusal + reactive superseding
+transition. Durable checkpoints show NO superseding transition and the
+phase map permits verifying→closed; live output showed first-attempt close
+success. The claim was RETRACTED with full provenance (original claim,
+correction, reason, evidence preserved in the contract record) and
+recorded as process-defect:unsupported-causal-claim. MiMo adoption
+maturity survives on the SUPPORTED evidence: scope constrained file edits,
+pattern obligations forced preservation proof, interface-reported
+observations, fresh resume.
+
+## Maturity
+
+Release-note generator: LIVE-PROVEN (real Gitea, 10 releases). Backfill:
+0.66.35→0.66.44 carrying historical-backfill provenance. Pre-0.66.35
+history: skipped (weak evidence — honest limitation).
@@ -0,0 +1,79 @@
+# SCAFFOLDING IMPLEMENTATION AUDIT + V1 GAP MATRIX (T18 §18B)
+
+_Every claim classified against evidence. Scale: LIVE_PROVEN (production
+evidence) / TEST_PROVEN (suite green, no production exercise) /
+IMPLEMENTED_UNPROVEN / PARTIAL / PLANNED / ABSENT / LEGACY / DUPLICATE._
+
+## Kernel claims
+
+| Claim | Status | Evidence |
+|---|---|---|
+| Contract lifecycle + legal transitions | LIVE_PROVEN | 28 contracts driven through real flow; refusal envelopes observed |
+| classify_mode (one classifier) | LIVE_PROVEN | gates TUI /build (17G), operator hook, tests 62+ |
+| Conversational intent layer | TEST_PROVEN | 22+ held-outs incl. unseen paraphrases; 17R live exercise (one real conversational proof) |
+| Programme continuation | LIVE_PROVEN | 17E 4-stage proof on real bridge; restart/dup/failure held-outs |
+| Event-driven waits (builder) | LIVE_PROVEN | T15 15Q real model build; T17E async stage |
+| Event-driven waits (CI) | LIVE_PROVEN | T16 16D deadline path ×2 + T17 natural wake (mode=event, 112.5s) |
+| Event replay (before-registration) | TEST_PROVEN + production-realistic exercise | 17C 4.6s vs real completed CI; natural-cycle confirmation: the 17Z wake arrived mid-wait, replay path itself not yet hit naturally |
+| Cross-user bridge delivery | LIVE_PROVEN | 17C repro as runner user (2.05s) + production deploys |
+| Provider boundary | TEST_PROVEN (structural guard) + observed loud-failure | 16F/G suites; no silent fallback observed in production runs |
+| Builder cannot self-certify | LIVE_PROVEN | T14 closure refusals ×2; phase-gated close |
+| Completion truth evidence-driven | LIVE_PROVEN | contract checkpoints; verify_test_truth refusals |
+| Interruption/resume | LIVE_PROVEN | T14E SIGKILL; 17E restart held-outs |
+| Impact preview (visible, pre-build) | ABSENT → T18 | nothing renders structured impact before build today |
+| Structural project model (first-class) | PARTIAL | stores are spine-shaped (entities/rels) but projects model CONTRACTS, not architecture; no domain/subsystem/interface nesting for arbitrary projects |
+| Structural queries/impact closure for projects | PARTIAL | spine.impact_frontier exists for Rune; project stores lack query API |
+| Research→Build Basis (durable, versioned) | ABSENT → T18 | research output today stays in transient context/artifacts |
+| Research-depth decision at construction entry | PARTIAL | ADAPT routes task classes; no explicit construction-stage research gate |
+| OUTWARD prior-art-first pass | PARTIAL | stage exists in Research Core; not strengthened to precedent-first ordering |
+| Schema versioning + consent-gated migration | PARTIAL | stores carry schema_version=1; no migration machinery/consent flow |
+| Environment/runtime model | PARTIAL | deployment_surface exists in Spine; project scaffolds don't model envs |
+| Workspace lifecycle (worktree per contract) | PARTIAL | BuildManager worktrees exist (legacy adapter); not wired to contracts |
+| Forge abstraction | ABSENT (Gitea-coupled scripts) | deploy/qualify call Gitea API directly |
+| Structural drift for project scaffolds | PARTIAL | drift_resume (16P) covers checkpoint-vs-reality; no file/entity-level drift detection |
+| Graph projection for visualisation | ABSENT → T18 | Spine has snapshots; no stable read-only graph projection |
+| Memory/Substrate | ABSENT (BY DESIGN — out of scope) | rule L |
+
+## V1 North Star gap matrix (this tranche's targets)
+
+| North Star capability | Gap | T18 action |
+|---|---|---|
+| Research-led construction (early research, Build Basis) | ABSENT Build Basis | 18C: build_basis.py + ADAPT depth hook |
+| OUTWARD precedent-first | PARTIAL | 18D: strengthen stage contract + real research run |
+| First-class structural Scaffold | PARTIAL (contract-shaped) | 18E: structural_model.py over structural_store core |
+| Structure→scope→governance | PARTIAL | 18F: queries + impact closure feeding contracts |
+| Visible impact preview | ABSENT | 18G: impact_preview projection |
+| Structural drift | PARTIAL | 18H: extend drift_resume with entity-level checks |
+| Schema versioning + consent | PARTIAL | 18I: migration + consent flow + tests |
+| Environment/runtime model | PARTIAL | 18J: env entities + verification distinction |
+| Workspace lifecycle | PARTIAL (legacy adapter) | 18K: workspace contract on the BuildManager adapter |
+| Forge boundary | ABSENT | 18L: forge.py adapter interface (Gitea impl) |
+| Full route convergence | PARTIAL (classify shared; backends differ) | 18M: audit + smallest fix |
+| Generalisation (not phrase list) | TEST_PROVEN | 18N: adversarial paraphrase families |
+| Self-scaffold | PARTIAL (contracts only) | 18O: structural model of Scaffolding itself |
+| Upgrade-via-Scaffold proof | ABSENT | 18P: one real upgrade driven by the scaffold |
+| Graph projection | ABSENT | 18Q: versioned read-only projection |
+
+## T19 additions (post-T18 classifications)
+
+| Capability | Status | Evidence |
+|---|---|---|
+| Project/Scaffold Registry | TEST_PROVEN | registry tests; boundaries proven (no truth in index, no Spine leak) |
+| /build governance fail-closed | LIVE_PROVEN | 19A: T18 degradation paths removed; failure-path held-outs |
+| Capability Ledgers | TEST_PROVEN | lifecycle+history+packet tests |
+| Surgical context packets | LIVE_PROVEN | 19R: packet bound exactly 1 file for a real change |
+| Behaviour/Test Specs + hardening lifecycle | TEST_PROVEN | ledger suites incl. generation history |
+| Selective invalidation | TEST_PROVEN | 19S-2/5/9/10 held-outs |
+| Predictive vs descriptive impact | TEST_PROVEN | delta persistence tests |
+| Post-change obligations | TEST_PROVEN | machine-derived from structural truth |
+| /testing backend + aggregate | TEST_PROVEN + LIVE (19R) | testing_mode suites; command wired in real TUI |
+| Observation→closure (user-confirm gate) | TEST_PROVEN | closure requires retest AND user confirm |
+| Forge failure never rewrites truth | TEST_PROVEN | 19S-14 held-out |
+| Self-hosted T19 change | LIVE_PROVEN | contract 19r-testing-command, 11-step scaffold-driven flow |
+
+## Kernel exceptions required (rule B record)
+
+None yet. Every T18 plan so far EXTENDS (new modules consuming
+structural_store) rather than modifying frozen kernel behavior. Any
+exception discovered during implementation will be recorded here with
+evidence that the V1 requirement could not be met otherwise.
@@ -0,0 +1,153 @@
+# SCAFFOLDING V1 DEFINITION (T18 §18B)
+
+_Plane-English + technical definition of Scaffolding as it ACTUALLY
+exists at tranche T18 start (master 29570f4). Companion artifact:
+SCAFFOLDING_IMPLEMENTATION_AUDIT (evidence classification per claim)._
+
+## What Scaffolding is (plain English)
+
+Scaffolding is Rune's **construction authority**: the governed process
+by which software gets built or changed. It takes an authorized need,
+classifies what kind of construction it is, proportionally governs the
+work through a durable contract (with legal lifecycle transitions),
+derives scope from structural reality, drives builders, waits on events
+(not sleeps), refuses self-certified completion, verifies from
+canonical reality, and records evidence on durable objects so any
+future session can resume without a transcript.
+
+It is NOT a scheduler, NOT a memory system, NOT a provider selector,
+NOT a research engine, NOT a release system. It is the layer that
+decides *whether, how, and with what obligations* construction happens
+— and whether it actually finished.
+
+## What it owns
+
+- **Construction classification** (`construction_mode.classify_mode` —
+  the ONE classifier: NOT_CONSTRUCTION / LOCAL_CHANGE /
+  EXISTING_PROJECT_CHANGE / PROJECT_BUILD / UNKNOWN; never silently
+  chooses; unknown surfaces probe questions)
+- **Conversational interpretation** (`construction_intent.py` —
+  meaning/referents/authorization; DISCUSSION→PROPOSAL→USER_APPROVED;
+  feeds the ONE classifier, never deepens)
+- **Project Scaffold + Contracts** (`scaffold.py` — legal phase map
+  draft→approved→building→verifying→…→closed/failed/abandoned; scope;
+  obligations; checkpoints; predicted-vs-observed at close)
+- **Programme continuation** (`programme_continuation.py` —
+  run_construction_programme: stages 1:1 to contracts, persist→canary→
+  next; "task done ≠ programme done")
+- **Impact/scope machinery** (construction_decision, completeness
+  gates, affected-surface propagation)
+- **Stage guards** (`stage_guards.py` — readiness blocks, liveness
+  proves progress, exit canary proves effect; verify_test_truth refuses
+  false greens)
+- **Failure taxonomy** (`failure_recovery.py` — 7 classes incl.
+  PROVIDER_UNAVAILABLE; retry budgets; PERMANENT/APPROVAL never retry)
+- **Builder interface** (`construction_interface.py` — ONE builder
+  boundary; JSON envelopes; builder identity recorded-but-inert;
+  discover_active_contract; capabilities())
+
+## What it does NOT own
+
+- **Provider/model selection** (provider layer owns; ambient context
+  via `provider_context.require_model`; structural guard test)
+- **Research competence** (Research Core owns; Scaffolding invokes and
+  consumes — see 18C integration)
+- **Asynchronous wait/resume machinery** (W/R event bridge owns;
+  `event_wait.py` is the repo-side adapter: event drives continuation,
+  watchdog detects failure, durable record is truth)
+- **Release/deployment** (deploy_stage/promotion/qualify + Gitea
+  workflows own it; Scaffolding records evidence)
+- **Rune's structural self-model** (Spine owns — `spine.py` over
+  `structural_store.py`; Scaffolding stores use the SAME generic core)
+- **Long-horizon relevance/memory** (future Substrate — not built;
+  seam exists: construction_projection + construction_commands)
+- **The TUI** (client; sends text, renders projections)
+
+## How the flows work
+
+**New build (PROJECT_BUILD)**: conversational/explicit intent →
+classify_mode (+ project-existence evidence) → activate() bootstraps a
+provisional Project Scaffold (identity before repository, UNKNOWN
+fields explicit) → contract draft→approved (user provenance) →
+building (builder under event wait) → verifying (exit canary;
+independent evidence) → closed with predicted-vs-observed.
+
+**Existing-project change**: intent → classify (scoped/file/target
+evidence; proportional) → attach to existing authority → light or full
+contract by depth → same lifecycle, proportionate ceremony.
+
+**Research-led construction (V1 target — CORRECTED 2026-09-15, Nick):**
+
+> THE RESEARCH PLAN ITSELF IS CONTEXTUALLY GENERATED. There is NO
+> permanent research-topic checklist. For every request:
+> REQUEST/INTENT → understand target/current reality → identify material
+> unknowns/decisions/risks → generate a context-specific research plan →
+> perform proportional INWARD/OUTWARD/deeper research. OUTWARD asks what
+> already exists relevant to THIS specific intent (adopt/adapt/combine/
+> reject/gaps/innovation). The categories researched are selected from
+> context; deeper technical research is driven by unresolved questions
+> and consequence level, never performed mechanically.
+
+`build_basis.generate_research_plan()` implements this: topics are
+synthesized from THIS intent's material unknowns/decisions/risks;
+the held-out suite proves unrelated intents produce materially
+different plans and that no canned category list exists.
+
+**Future upgrades (V1 target)**: authorized need → research-depth
+decision (ADAPT) → Research Core pass → durable Build Basis →
+structural Scaffold consult/construct → impact closure + visible
+preview → contract with structurally-derived scope → isolated
+workspace → governed build/verify → merge/release → structural
+Scaffold updated. (Parts of this chain are new in T18 — see audit.)
+
+**Interruption/resume**: SIGKILL-safe. Durable contract + programme
+records + wait records are the truth; fresh process resumes from
+identity alone; duplicate signals idempotent (live-proven T14E/T17E).
+
+**Deployment/release participation**: Scaffolding does not run the
+pipeline; it consumes qualified CI/deploy evidence, records deployed
+identity on contracts, and refuses closure without it where contracted.
+
+## T19 layer model (the general engineering system)
+
+> "The Scaffold should let the LLM operate with surgical precision
+>  locally, while preserving global awareness of what else must be
+>  checked or changed."
+>
+> "Lower active cognitive load without losing the whole idea."
+
+- **STRUCTURAL MODEL** tells Rune what the project contains / how it connects.
+- **CAPABILITY LEDGER** provides compact local re-entry context (stable IDs
+  across moves; history preserved through supersede/retire).
+- **BEHAVIOUR VALIDATION LEDGER** tells Rune what intended behaviour is
+  actually confirmed — generation-aware, selective invalidation, automated
+  green never implies USER_CONFIRMED.
+- **PROJECT REGISTRY** tells Rune which Scaffolds/projects exist
+  (discovery/index only — not structural truth, not Spine).
+- **ITERATION ENGINE** (predictive→descriptive impact, impact-delta
+  records, machine-derived obligations) makes quicksand protection
+  STANDARD behaviour of every construction route.
+- **/TESTING** is a dependency-aware hardening programme: dynamic queue,
+  fix-for-B-may-reopen-A, closure requires retest AND user confirmation.
+- **FORGE** provides workflow/history projection (failure never rewrites
+  canonical validation truth).
+- **FUTURE SUBSTRATE** will coordinate sequencing/current focus/pause-
+  resume but does NOT own structural or validation truth.
+
+## How Spine relates
+
+Spine = Rune's canonical structural self-model (structural_store
+generic core, maturity ladder, impact_frontier). Scaffolding's project
+stores share the SAME generic core (zero Rune constants) — one graph
+discipline, two scopes: Spine models Rune; project Scaffolds model
+their projects. No duplicate truth: a project's structural model lives
+in its own store; Rune's lives in Spine. Later retrofit will modernise
+Spine using the discipline V1 generalises (18K relationship).
+
+## How Substrate will interact (without owning internals)
+
+Read: versioned outbound projection (construction_projection,
+SCHEMA_VERSION=1). Write: bounded validated commands
+(construction_commands — 6 commands, lifecycle-validated, idempotent,
+no direct storage mutation). Substrate asks; Scaffolding decides.
+Memory graph explicitly out of scope (T18 rule L).
@@ -0,0 +1,73 @@
+# Rune Structural Map
+
+_Maintained structural projection — the fast re-entry surface (T16 §16N).
+Companion to the Spine store (`store/spine/current.json`, structural
+authority) and `store/evidence/t16/16M_inventory.md` (full classification
+record). Update this map when authority or canonical routes change; do
+not copy the repo here._
+
+## What is Rune?
+
+A self-hosting coding agent: TUI client over an operator core, with a
+structural self-model (Spine), a construction authority (Scaffolding),
+an event-driven wait/resume bridge (W/R), a research pipeline, and a
+fail-closed release/deployment pipeline that builds, tests, releases,
+and deploys Rune itself to `/mnt/workdrive/rune-code/current`.
+
+## Where things happen (canonical routes only)
+
+| Concern | Authority | State |
+|---|---|---|
+| Structural self-model | `src/rune/spine.py` | `store/spine/current.json` |
+| Construction (build lifecycle) | `src/rune/scaffold.py` + `construction_*.py` | scaffold store (`<dir>/.rune/scaffold`) |
+| Builder sessions | `scripts/t15_builder_session.py` (event-driven successor; T14 is legacy) | contract + wait records |
+| Provider/model selection | `rune_operator` final decider; ambient context via `provider_context.require_model` | `state/model_routing_profiles.json` |
+| Wait/resume (any long op) | `src/rune/event_wait.py` + W/R daemon (out-of-repo snapshot at `~/.local/share/mimo-wait-resume`, data root `/mnt/workdrive/wr-bridge` — PIN `RUNE_WR_ROOT`: unpinned consumers get a loud warning, never a silent split-brain) | waits/events/leases/resumes JSON |
+| Programme continuation | `src/rune/programme_continuation.py` (WIRED T17: `run_construction_programme` binds stages to Scaffolding contracts) | scaffold store `programmes/` |
+| Conversational construction | `src/rune/construction_intent.py` (meaning/referents/authorization) → `construction_mode.classify_mode` (the ONE classifier) | intents are transient; contracts durable |
+| CI qualification verdict | `scripts/deploy_qualify.py` (two legs, fail-closed) | Gitea commit status + runs |
+| Release/deploy | `scripts/deploy_stage.py` → `deploy_promotion.py` (monotonic, flock) | deploy root + `current`/`last-good` symlinks |
+| Research | `scripts/research_task_tree.py` (authority; labels are projection) | `research/issue-<N>/` |
+| Task completion truth | `verification_layer.py` + `orchestration/conversation_state.py` | durable event store + checkpoints |
+| Routing (task class) | `src/rune/adaptive_router.py` | `routes/route.jsonl` |
+| Canary registry | `src/rune/canary_native.MODULES` | observed by Spine |
+| Structural health | `henxels_checks.py` + `henxels.yaml` | repo root |
+
+## Known non-canonical / duplicate routes (do not extend; reconcile via 16Q)
+
+- `build_mode.BuildManager` (legacy /build) — superseded by builder_bridge
+  + event-driven sessions; still wired.
+- `scripts/t14_builder_session.py` — superseded by t15 event-driven driver.
+- Provider selection entry points ×5 (`model_routing.rank`,
+  `model_connect route`, `model_registry get`, `provider_registry`,
+  `work_manager`) — operator is the de-facto final decider.
+- Task-ish state ×4 (task-tool DB, `task_persist` persisted.json,
+  `memory_store.working`, orchestration event store).
+- Maintenance ledgers ×3 (scaffold maintenance_item, tui_backlog, drift_ledger).
+
+## Unfinished / blocked / deprecated
+
+- programme_continuation.py: WIRED (T17 17D — run_construction_programme
+  is the canonical multi-stage driver; 17E live-proven).
+- maintenance.yml canary-suite job: loops `shared/canary_*.py` — wrong
+  dir (real canaries in `src/rune/`), job likely no-ops (LEGACY).
+- daemon out-of-repo: the W/R daemon is deployed from
+  `~/.local/share/mimo-wait-resume` with its own git snapshot —
+  packaging debt, no CI coverage of the deployed copy.
+- Substrate: NOT BUILT. Scaffolding exposes the seam
+  (`construction_projection.py`, `construction_commands.py`); mock
+  client proves compatibility (`tests/test_substrate_seam.py`).
+
+## Fresh-session quick answers
+
+- Is a claim true? → Spine store + evidence_refs, never commit messages.
+- Is construction allowed? → scaffold contract phase + LEGAL_TRANSITIONS.
+- How does a long wait work? → register on the W/R bridge, yield on the
+  FIFO event; watchdog is failure detection, never scheduling; the event
+  is a WAKE — authority is re-read after resume.
+- Who picks the provider? → the user layer, via ambient context;
+  construction never does (`test_provider_ownership.py` enforces).
+- What runs where? → CI on `rune-ci-desktop` runner; deploy stages to
+  `/mnt/workdrive/rune-code/<sha>`; data home pinned at
+  `/mnt/workdrive/rune` (RUNE_HOME); implementation worktree
+  `/mnt/workdrive/issue161-harness`.
@@ -9,6 +9,7 @@ new surface._
 |---|---|---|
 | `rune_tui.py` | The main Textual application — chat pane, work pane, status bar (version label + refresh), footer, dialogs | The known giant (file-size debt tracked in henxels); composition root for all UI |
 | `tui_validator.py` | Import-allowlist validation for TUI code | Drift defect tracked in #139 (allowlist vs `__future__`/relative imports) |
+| `tui_backlog.py` | Not a surface — the durable TUI Upgrade Backlog ledger (outstanding projection/control work per backend feature) | SCAFFOLD-TRANCHE-007; ledger only, never structural authority |

 ## Adding a TUI surface

@@ -1,4 +1,4 @@
 {
-  "BLE001": 352,
-  "S110": 80
-}
+  "BLE001": 347,
+  "S110": 75
+}
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"

 [project]
 name = "rune"
-version = "0.66.31"
+version = "0.86.0"
 description = "Rune — a stateful coding assistant with persistent memory, orchestration, and TUI"
 requires-python = ">=3.10"

@@ -1,5 +1,5 @@
 {
- "errorCount": 124,
- "warningCount": 429,
- "informationCount": 0
+  "errorCount": 117,
+  "warningCount": 454,
+  "informationCount": 0
 }
@@ -1,29 +1,34 @@
 {
-    "typeCheckingMode": "standard",
-    "pythonVersion": "3.14",
-    "reportMissingImports": "warning",
-    "reportMissingModuleSource": "none",
-    "reportMissingTypeStubs": "none",
-    "reportUnknownArgumentType": "none",
-    "reportUnknownParameterType": "none",
-    "reportUnknownVariableType": "none",
-    "reportUnknownMemberType": "none",
-    "reportMissingTypeArgument": "none",
-    "reportUntypedFunctionDecorator": "none",
-    "reportUntypedClassDecorator": "none",
-    "reportUntypedBaseClass": "none",
-    "reportPrivateUsage": "none",
-    "reportOptionalMemberAccess": "warning",
-    "reportUnnecessaryIsInstance": "warning",
-    "reportUnnecessaryComparison": "warning",
-    "reportUnnecessaryTypeIgnoreComment": "warning",
-    "reportPossiblyUnboundVariable": "error",
-    "reportUndefinedVariable": "error",
-    "reportAttributeAccessIssue": "warning",
-    "include": ["src/rune"],
-    "exclude": [
-        "src/rune/__pycache__",
-        "**/node_modules",
-        "**/__pycache__"
-    ]
-}
+  "typeCheckingMode": "standard",
+  "pythonVersion": "3.14",
+  "reportMissingImports": "warning",
+  "reportMissingModuleSource": "none",
+  "reportMissingTypeStubs": "none",
+  "reportUnknownArgumentType": "none",
+  "reportUnknownParameterType": "none",
+  "reportUnknownVariableType": "none",
+  "reportUnknownMemberType": "none",
+  "reportMissingTypeArgument": "none",
+  "reportUntypedFunctionDecorator": "none",
+  "reportUntypedClassDecorator": "none",
+  "reportUntypedBaseClass": "none",
+  "reportPrivateUsage": "none",
+  "reportOptionalMemberAccess": "warning",
+  "reportUnnecessaryIsInstance": "warning",
+  "reportUnnecessaryComparison": "warning",
+  "reportUnnecessaryTypeIgnoreComment": "warning",
+  "reportPossiblyUnboundVariable": "error",
+  "reportUndefinedVariable": "error",
+  "reportAttributeAccessIssue": "warning",
+  "include": [
+    "src/rune"
+  ],
+  "exclude": [
+    "src/rune/__pycache__",
+    "**/node_modules",
+    "**/__pycache__"
+  ],
+  "extraPaths": [
+    "scripts"
+  ]
+}
@@ -1,5 +1,5 @@
 # This file was autogenerated by uv via the following command:
-#    uv pip compile pyproject.toml -o requirements-ci.txt --extra dev --extra memory --python-version 3.11
+#    uv pip compile pyproject.toml -o /tmp/req_new.txt --extra dev --extra memory --python-version 3.11
 aiohappyeyeballs==2.7.1
    # via aiohttp
 aiohttp==3.14.3
@@ -18,7 +18,7 @@ certifi==2026.7.22
    #   httpx
 chardet==7.6.0
    # via diff-cover
-coverage==7.16.0
+coverage==7.16.1
    # via rune (pyproject.toml)
 diff-cover==10.5.1
    # via rune (pyproject.toml)
@@ -34,7 +34,7 @@ httpcore==1.0.9
    # via httpx
 httpx==0.28.1
    # via rune (pyproject.toml)
-idna==3.19
+idna==3.20
    # via
    #   anyio
    #   httpx
@@ -56,7 +56,7 @@ mdit-py-plugins==0.6.1
    # via textual
 mdurl==0.1.2
    # via markdown-it-py
-multidict==6.8.0
+multidict==6.9.0
    # via
    #   aiohttp
    #   yarl
@@ -64,13 +64,13 @@ nodeenv==1.10.0
    # via pyright
 packaging==26.3
    # via pytest
-platformdirs==4.11.8
+platformdirs==4.11.11
    # via textual
 pluggy==1.6.0
    # via
    #   diff-cover
    #   pytest
-propcache==0.5.2
+propcache==0.5.4
    # via
    #   aiohttp
    #   yarl
@@ -118,5 +118,5 @@ typing-extensions==4.16.0
    #   textual-autocomplete
 websockets==17.1
    # via rune (pyproject.toml)
-yarl==1.24.5
+yarl==1.25.1
    # via aiohttp
@@ -0,0 +1,480 @@
+"""await_intent.py — ASYNC CORRECTIVE PATCH: intent normalization at the
+execution boundary.
+
+PRINCIPLE: MODEL OUTPUT = INTENT. RUNE CONTROLLER = EXECUTION SEMANTICS.
+
+The model may express continuation intent in any quality — `await_operation
+(deadline_secs=1200)`, `await-operation [op_type=master_ci, sha=X]`,
+"sleep 300 then check CI", "poll until done". When a matching
+PendingOperation exists (or the deterministic resolver can derive the
+operation truth from git/Gitea state), the controller executes the
+CANONICAL continuation and discards model-supplied scheduling policy.
+
+Authority provenance for timing:
+  USER_EXPLICIT / EXTERNAL_CONTRACT / TEST_HARNESS /
+  GOVERNED_OPERATION_POLICY  -> authoritative, retained
+  MODEL_INVENTED             -> non-authoritative, ignored
+
+This module owns NO storage: it reads the existing W/R wait store
+(rune.event_wait) and git/Gitea state. The canonical continuation remains
+scripts/ci_event_wait.py run() (the T21.5 single-logical-await).
+"""
+from __future__ import annotations
+
+import json
+import os
+import re
+import subprocess
+import sys
+from pathlib import Path
+from typing import Any
+
+_SCRIPT_DIR = Path(__file__).resolve().parent
+if str(_SCRIPT_DIR) not in sys.path:
+    sys.path.insert(0, str(_SCRIPT_DIR))
+_REPO_SRC = str(_SCRIPT_DIR.parent / "src")
+if _REPO_SRC not in sys.path:
+    sys.path.insert(0, _REPO_SRC)
+
+WR_BRIDGE = Path(os.environ.get("RUNE_WR_ROOT", "/mnt/workdrive/wr-bridge"))
+REPO = Path("/mnt/workdrive/issue161-harness")
+GIT_TIMEOUT = 10
+
+# Default policy deadline per operation type — GOVERNED_OPERATION_POLICY.
+# The model's numbers never become policy; these are the controller's.
+POLICY_DEADLINES = {
+    "branch_ci": 1800,
+    "pull_request_checks": 1800,
+    "master_ci": 1800,
+    "deploy": 900,
+    "release": 900,
+}
+
+VALID_OP_TYPES = ("branch_ci", "pull_request_checks", "master_ci",
+                  "deploy", "release")
+
+
+# ── timing provenance ───────────────────────────────────────────────────────
+
+def classify_timing_provenance(model_deadline: Any = None,
+                               user_deadline: Any = None,
+                               contract_deadline: Any = None) -> tuple[int, str]:
+    """Resolve the effective deadline + its provenance.
+
+    Authority order: USER_EXPLICIT > EXTERNAL_CONTRACT > TEST_HARNESS >
+    GOVERNED_OPERATION_POLICY. MODEL_INVENTED values are recorded but
+    NEVER used as policy.
+    """
+    if model_deadline is not None:
+        try:
+            model_deadline = int(model_deadline)
+        except (TypeError, ValueError):
+            model_deadline = None
+    if user_deadline is not None:
+        return int(user_deadline), "USER_EXPLICIT"
+    if contract_deadline is not None:
+        return int(contract_deadline), "EXTERNAL_CONTRACT"
+    return (model_deadline or 0), "MODEL_INVENTED"  # caller discards this
+
+
+# ── PendingOperation lookup over the existing W/R wait store ────────────────
+
+def _load_wait_records() -> list[dict]:
+    """All durable wait records from the bridge store (read-only)."""
+    out = []
+    waits = WR_BRIDGE / "waits"
+    if not waits.is_dir():
+        return out
+    for f in sorted(waits.glob("*.json")):
+        try:
+            rec = json.loads(f.read_text())
+            rec["_file"] = f.name
+            out.append(rec)
+        except (OSError, ValueError):
+            continue
+    return out
+
+
+def _rec_operation_id(rec: dict) -> str:
+    return str((rec.get("ext") or {}).get("operation_id")
+               or rec.get("operation_id") or "")
+
+
+def _rec_sha(rec: dict) -> str:
+    return str(rec.get("sha") or "")
+
+
+def pending_operations(sha: str = "", op_type: str = "") -> list[dict]:
+    """Pending (non-terminal) operations, optionally filtered."""
+    terminal = {"COMPLETED", "FAILED", "CLOSED", "CANCELLED", "EXPIRED"}
+    out = []
+    for rec in _load_wait_records():
+        status = str(rec.get("status", "")).upper()
+        if status in terminal:
+            continue
+        if sha and _rec_sha(rec) != sha:
+            continue
+        out.append(rec)
+    return out
+
+
+def find_op_type_for_sha(sha: str) -> str | None:
+    """The op_type a prior canonical wait already recorded for this SHA
+    (objective text embeds the type: 'event-driven {op_type} qualification')."""
+    for rec in _load_wait_records():
+        if _rec_sha(rec) != sha:
+            continue
+        obj = str(rec.get("continuation", {}).get("objective", ""))
+        m = re.search(r"event-driven (\w+) qualification", obj)
+        if m and m.group(1) in VALID_OP_TYPES:
+            return m.group(1)
+    return None
+
+
+# ── deterministic op_type derivation from repo/Gitea truth ─────────────────
+
+def _git(*args: str) -> str:
+    try:
+        r = subprocess.run(["git", "-C", str(REPO), *args],
+                           capture_output=True, text=True, timeout=GIT_TIMEOUT)
+        return r.stdout.strip()
+    except (OSError, subprocess.TimeoutExpired):
+        return ""
+
+
+def sha_branch(sha: str) -> str:
+    """Branch(es) containing this commit in the local repo."""
+    return _git("branch", "--all", "--contains", sha)
+
+
+def sha_on_master(sha: str) -> bool:
+    out = _git("branch", "--contains", sha)
+    return any(l.strip().endswith("master") or l.strip() == "master"
+               for l in out.splitlines())
+
+
+def open_pr_for_sha(sha: str) -> int | None:
+    """Is there an OPEN PR whose head SHA is this one? (Gitea API, read-only.)"""
+    try:
+        sys.path.insert(0, str(_SCRIPT_DIR))
+        import gitea_auth  # noqa: E402
+        st, prs = gitea_auth.api("/pulls?state=open&limit=50")
+        if st != 200:
+            return None
+        for p in prs:
+            hs = str(p.get("head") or {}).get("sha") if isinstance(
+                p.get("head"), dict) else None
+            if hs and str(hs).startswith(sha[:12]):
+                return int(p.get("number"))
+    except Exception:  # noqa: BLE001 — resolver must never crash the tool
+        return None
+    return None
+
+
+def derive_op_type(sha: str) -> tuple[str, str]:
+    """Deterministically derive the canonical op_type for a SHA.
+
+    Returns (op_type, evidence). Order (A-patch §2 amendment):
+      1. POST-PATCH normalized record (ext.intent_normalized=true) — true
+         stored operation truth, wins outright;
+      2. open PR for the sha -> pull_request_checks;
+      3. sha ancestor of master -> master_ci;
+      4. sha on a feature branch -> branch_ci;
+      5. legacy (pre-patch) records only CONFIRM, never override — they
+         were created from model-supplied defaults and cannot be trusted.
+    Never trusts model input.
+    """
+    for rec in _load_wait_records():
+        if _rec_sha(rec) != sha:
+            continue
+        if (rec.get("ext") or {}).get("intent_normalized"):
+            obj = str(rec.get("continuation", {}).get("objective", ""))
+            m = re.search(r"event-driven (\w+) qualification", obj)
+            if m and m.group(1) in VALID_OP_TYPES:
+                return m.group(1), "normalized_record:" + m.group(1)
+    pr = open_pr_for_sha(sha)
+    if pr:
+        return "pull_request_checks", f"open_pr:#{pr}"
+    if sha_on_master(sha):
+        return "master_ci", "commit_ancestor_of_master"
+    return "branch_ci", "commit_on_feature_branch"
+
+
+# ── intent normalization ────────────────────────────────────────────────────
+
+def normalize_action(action: dict, *,
+                     model_deadline: Any = None,
+                     user_deadline: Any = None,
+                     contract_deadline: Any = None,
+                     task_id: str = "") -> dict:
+    """Normalize a model continuation action into the canonical continuation.
+
+    action: {kind: await_operation|sleep_poll|check_status|unknown,
+             sha?: str, operation_id?: str, op_type?: str, raw?: str}
+    task_id: REAL_ASYNC_BRIDGE §9 — when the caller knows its task
+             binding, pending-operation resolution is scoped to operations
+             registered with that task_id; stale global operations are
+             never consulted when a binding exists.
+    Returns:
+      {normalized: bool, action: "ContinuePendingOperation", operation_id,
+       op_type, authority, deadline_secs, deadline_provenance,
+       model_deadline_ignored, model_op_type_ignored, evidence}
+    or {normalized: False, reason: ...} / {normalized: "AMBIGUOUS", candidates}
+    """
+    sha = str(action.get("sha") or "")
+    op_id = str(action.get("operation_id") or "")
+    model_op_type = action.get("op_type")
+    model_deadline_ignored = model_deadline is not None
+    model_op_type_ignored = model_op_type is not None
+
+    # 0. task binding scope (§9): task-bound operations beat global state.
+    #    Stored operation wins over ANY model hint (sha/op_type/deadline).
+    if task_id and not op_id:
+        bound = [r for r in pending_operations()
+                 if str(r.get("ext", {}).get("task_id", "")) == task_id]
+        if len(bound) == 1:
+            rec = bound[0]
+            norm = _canonical_from_record(rec, model_deadline_ignored,
+                                          model_op_type_ignored,
+                                          "task_bound",
+                                          user_deadline, contract_deadline)
+            norm["sha_for_wait"] = _rec_sha(rec)
+            norm["task_id"] = task_id
+            return norm
+        if len(bound) > 1:
+            return {"normalized": "AMBIGUOUS",
+                    "candidates": [_rec_operation_id(c) for c in bound],
+                    "reason": f"multiple pending operations bound to task {task_id}"}
+        return {"normalized": False,
+                "reason": f"no pending operation bound to task {task_id}"}
+
+    # 1. explicit operation_id wins
+    if op_id:
+        from rune import event_wait as ew  # noqa: E402
+        rec = ew.operation_wait(op_id)
+        if rec is None:
+            # REAL_ASYNC_BRIDGE §10: a FRESH operation has no wait record
+            # yet — registration happens at the park step (ci_event_wait).
+            # For an explicit, well-formed, EXISTING commit id, return the
+            # provisional canonical continuation (policy deadline, own-runs
+            # authority) so the model-facing tool can park on first call.
+            # A typo'd/ghost sha is refused here (A-patch lesson): the
+            # commit must exist in the repository.
+            import re as _re
+            m = _re.fullmatch(r"ci-([0-9a-f]{40})", op_id)
+            if m:
+                sha40 = m.group(1)
+                probe = subprocess.run(
+                    ["git", "-C", str(REPO), "cat-file", "-e",
+                     f"{sha40}^{{commit}}"],
+                    capture_output=True, timeout=15)
+                if probe.returncode == 0:
+                    return {
+                        "normalized": True,
+                        "action": "ContinuePendingOperation",
+                        "operation_id": op_id,
+                        "sha_for_wait": sha40,
+                        "op_type": "branch_ci",
+                        "authority": "ci_runs",
+                        "deadline_secs": POLICY_DEADLINES["branch_ci"],
+                        "deadline_provenance": "GOVERNED_OPERATION_POLICY",
+                        "model_deadline_ignored": model_deadline_ignored,
+                        "model_op_type_ignored": model_op_type_ignored,
+                        "evidence": "explicit_id_fresh_registration",
+                    }
+            return {"normalized": False,
+                    "reason": f"operation_id {op_id} has no wait record"}
+        return _canonical_from_record(rec, model_deadline_ignored,
+                                      model_op_type_ignored, "explicit_id",
+                                      user_deadline, contract_deadline)
+
+    # 2. resolve by sha / active operation.
+    # H-PATCH H2/H3: the model sha is a HINT. If it doesn't resolve but
+    # unambiguous pending operations exist, the STORED target wins — a
+    # mistyped model sha must never create a ghost operation (the A-patch
+    # live failure: a typo'd sha derived branch_ci for a nonexistent commit
+    # and parked a wait on a ghost).
+    # 2. no explicit operation_id: resolve by sha / pending operations.
+    # H-PATCH: when the model provides NO sha (operation-id-first mode),
+    # resolve from unambiguous pending operations — the model should not
+    # need to reconstruct a 40-char sha to express "wait for it".
+    if not sha:
+        all_pending = pending_operations()
+        if len(all_pending) == 1:
+            rec = all_pending[0]
+            norm = _canonical_from_record(rec, model_deadline_ignored,
+                                          model_op_type_ignored,
+                                          "auto_resolved_no_sha",
+                                          user_deadline, contract_deadline)
+            norm["sha_for_wait"] = _rec_sha(rec)
+            return norm
+        if len(all_pending) > 1:
+            return {"normalized": "AMBIGUOUS",
+                    "candidates": [_rec_operation_id(c) for c in all_pending],
+                    "reason": "multiple pending operations, no sha hint"}
+        return {"normalized": False,
+                "reason": "no pending operations and no sha provided"}
+
+    candidates = pending_operations(sha=sha)
+    if len(candidates) > 1 and not sha:
+        return {"normalized": "AMBIGUOUS",
+                "candidates": [_rec_operation_id(c) for c in candidates],
+                "reason": "multiple pending operations, no target"}
+    if len(candidates) == 1:
+        rec = candidates[0]
+        stored_sha = _rec_sha(rec)
+        norm = _canonical_from_record(rec, model_deadline_ignored,
+                                      model_op_type_ignored, "pending_match",
+                                      user_deadline, contract_deadline)
+        if sha and stored_sha and sha != stored_sha:
+            norm["model_sha_mismatch"] = {
+                "model_sha": sha, "stored_sha": stored_sha,
+                "resolution": "stored_target_used"}
+        return norm
+    # model sha unresolved: any unambiguous pending op at all? then the
+    # model probably mistyped the sha — use the stored target, log it.
+    all_pending = pending_operations()
+    if len(all_pending) == 1 and sha:
+        rec = all_pending[0]
+        stored_sha = _rec_sha(rec)
+        norm = _canonical_from_record(rec, model_deadline_ignored,
+                                      model_op_type_ignored,
+                                      "unresolved_sha_single_pending",
+                                      user_deadline, contract_deadline)
+        norm["model_sha_mismatch"] = {
+            "model_sha": sha, "stored_sha": stored_sha,
+            "resolution": "stored_target_used"}
+        return norm
+
+    # 3. no stored record: derive truth deterministically
+    if not sha:
+        return {"normalized": False,
+                "reason": "no pending operation and no sha to resolve"}
+    op_type, evidence = derive_op_type(sha)
+    return {
+        "normalized": True,
+        "action": "ContinuePendingOperation",
+        "operation_id": f"ci-{sha}",
+        "op_type": op_type,
+        "authority": "ci_runs" if op_type in ("branch_ci",
+                                              "pull_request_checks")
+                     else "deploy_qualify",
+        "deadline_secs": POLICY_DEADLINES[op_type],
+        "deadline_provenance": "GOVERNED_OPERATION_POLICY",
+        "model_deadline_ignored": model_deadline_ignored,
+        "model_op_type_ignored": model_op_type_ignored,
+        "evidence": evidence,
+    }
+
+
+def _canonical_from_record(rec: dict, model_deadline_ignored: bool,
+                           model_op_type_ignored: bool, how: str,
+                           user_deadline: Any, contract_deadline: Any) -> dict:
+    op_id = _rec_operation_id(rec)
+    obj = str(rec.get("continuation", {}).get("objective", ""))
+    m = re.search(r"event-driven (\w+) qualification", obj)
+    recorded = m.group(1) if (m and m.group(1) in VALID_OP_TYPES) else None
+    rec_sha = _rec_sha(rec)
+    if (rec.get("ext") or {}).get("intent_normalized") or how == "explicit_id":
+        # normalized record or explicit id: stored truth wins
+        op_type = recorded or "master_ci"
+    else:
+        # legacy record: git/Gitea derivation outranks the model-era default
+        op_type, _ev = derive_op_type(rec_sha) if rec_sha else \
+            (recorded or "master_ci", "no_sha")
+    deadline, prov = classify_timing_provenance(
+        user_deadline=user_deadline, contract_deadline=contract_deadline)
+    if prov == "MODEL_INVENTED" or not deadline:
+        deadline = POLICY_DEADLINES.get(op_type, 1800)
+        prov = "GOVERNED_OPERATION_POLICY"
+    return {
+        "normalized": True,
+        "action": "ContinuePendingOperation",
+        "operation_id": op_id,
+        "op_type": op_type,
+        "authority": "ci_runs" if op_type in ("branch_ci",
+                                              "pull_request_checks")
+                     else "deploy_qualify",
+        "deadline_secs": deadline,
+        "deadline_provenance": prov,
+        "model_deadline_ignored": model_deadline_ignored,
+        "model_op_type_ignored": model_op_type_ignored,
+        "matched_by": how,
+    }
+
+
+def normalize_bash(command: str, *, sha_hint: str = "") -> dict | None:
+    """Recognize sleep/check/poll intent aimed at a known pending operation.
+
+    Returns a normalization record if the command's semantic intent is
+    continuation of a known external operation; None if the command is
+    legitimate local timing (short sync, user delay, backoff) with no
+    pending operation to continue. NEVER returns an executed sleep.
+    """
+    if not command:
+        return None
+    cmd = command.strip()
+    # long sleep + (status-check language or an explicit sha hint)
+    sleep_m = re.search(r"\bsleep\s+(\d{2,5})\b", cmd)
+    if not sleep_m:
+        return None
+    secs = int(sleep_m.group(1))
+    check_intent = bool(re.search(
+        r"\b(ci|deploy|release|status|check|poll|pipeline|run)\b", cmd,
+        re.IGNORECASE))
+    sha = sha_hint or ""
+    m = re.search(r"\b([0-9a-f]{40}|[0-9a-f]{12})\b", cmd)
+    if m:
+        sha = m.group(1)
+    if not check_intent and not sha:
+        return None  # plain sleep, no continuation intent
+    cands = pending_operations(sha=sha) if sha else pending_operations()
+    if not cands and not sha:
+        return None
+    if len(cands) > 1:
+        return {"normalized": "AMBIGUOUS",
+                "candidates": [_rec_operation_id(c) for c in cands],
+                "raw_command_kind": "sleep_poll"}
+    rec = cands[0] if cands else None
+    if rec is not None:
+        norm = _canonical_from_record(rec, True, False, "sleep_poll_intent",
+                                      None, None)
+        norm["sha_for_wait"] = _rec_sha(rec)
+    elif sha:
+        op_type, evidence = derive_op_type(sha)
+        norm = {
+            "normalized": True, "action": "ContinuePendingOperation",
+            "operation_id": f"ci-{sha}", "op_type": op_type,
+            "authority": "ci_runs" if op_type in ("branch_ci",
+                                                  "pull_request_checks")
+                         else "deploy_qualify",
+            "deadline_secs": POLICY_DEADLINES[op_type],
+            "deadline_provenance": "GOVERNED_OPERATION_POLICY",
+            "model_deadline_ignored": True, "model_op_type_ignored": False,
+            "evidence": evidence, "sha_for_wait": sha,
+        }
+    else:
+        return None
+    norm["raw_command_kind"] = "sleep_poll"
+    norm["raw_sleep_seconds_suppressed"] = secs
+    return norm
+
+
+def emit_normalized_record(record: dict) -> None:
+    """Compact structured observability — one JSONL line per normalization.
+    No payloads, no secrets."""
+    try:
+        log = _SCRIPT_DIR.parent / "store" / "await_intent_log.jsonl"
+        log.parent.mkdir(parents=True, exist_ok=True)
+        slim = {k: record.get(k) for k in (
+            "normalized", "action", "operation_id", "op_type", "authority",
+            "deadline_secs", "deadline_provenance", "model_deadline_ignored",
+            "model_op_type_ignored", "matched_by", "evidence",
+            "raw_command_kind", "raw_sleep_seconds_suppressed", "reason",
+            "candidates") if k in record}
+        slim["at"] = __import__("time").strftime(
+            "%Y-%m-%dT%H:%M:%SZ", __import__("time").gmtime())
+        with open(log, "a") as f:
+            f.write(json.dumps(slim, sort_keys=True) + "\n")
+    except OSError:
+        pass
@@ -0,0 +1,74 @@
+#!/usr/bin/env python3
+"""await_intent_cli.py — CLI surface for the intent normalization layer.
+
+Used by the await-operation harness tool so the MODEL's scheduling hints
+never reach the wait engine. Prints one JSON line.
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+import await_intent  # noqa: E402
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser(description=__doc__)
+    sub = ap.add_subparsers(dest="cmd", required=True)
+    n = sub.add_parser("normalize", help="normalize a continuation intent")
+    n.add_argument("--sha", default="")
+    n.add_argument("--operation-id", default="")
+    n.add_argument("--task-id", default="",
+                   help="§9 task binding: scope pending resolution to "
+                        "operations registered for this task")
+    n.add_argument("--model-op-type", default=None)
+    n.add_argument("--model-deadline", default=None)
+    n.add_argument("--user-deadline", default=None,
+                   help="USER_EXPLICIT deadline (authoritative)")
+    n.add_argument("--contract-deadline", default=None,
+                   help="EXTERNAL_CONTRACT deadline (authoritative)")
+    rb = sub.add_parser("route-bash",
+                        help="route a Bash command expressing wait intent")
+    rb.add_argument("--command", required=True)
+    rb.add_argument("--sha-hint", default="")
+    args = ap.parse_args()
+
+    if args.cmd == "route-bash":
+        record = await_intent.normalize_bash(args.command,
+                                             sha_hint=args.sha_hint)
+        if record is None:
+            print(json.dumps({"normalized": False,
+                              "reason": "no external continuation intent"}))
+            return 0
+        if record.get("normalized") is True:
+            # the compiled wait needs the SHA: derive from operation_id
+            oid = str(record.get("operation_id", ""))
+            if oid.startswith("ci-"):
+                record["sha_for_wait"] = oid[3:]
+        await_intent.emit_normalized_record(record)
+        print(json.dumps(record, sort_keys=True))
+        return 0
+
+    action = {"kind": "await_operation", "sha": args.sha,
+              "operation_id": args.operation_id,
+              "op_type": args.model_op_type}
+    model_deadline = (int(args.model_deadline)
+                      if args.model_deadline else None)
+    user_deadline = (int(args.user_deadline)
+                     if args.user_deadline else None)
+    contract_deadline = (int(args.contract_deadline)
+                         if args.contract_deadline else None)
+    record = await_intent.normalize_action(
+        action, model_deadline=model_deadline,
+        user_deadline=user_deadline, contract_deadline=contract_deadline,
+        task_id=args.task_id)
+    await_intent.emit_normalized_record(record)
+    print(json.dumps(record, sort_keys=True))
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,588 @@
+#!/usr/bin/env python3
+"""ci_event_wait.py — T16 §16C/§16D: event-driven CI qualification wait.
+
+Replaces the deploy.yml sleep-30 → query → repeat polling loop with the
+EXISTING W/R event bridge (src/rune/event_wait.py — the SAME primitive the
+builder adapter uses; §16E: one bridge, adapters differ only in identity
+and authority probe).
+
+Flow (mandate §16C):
+
+    push / CI workflow starts
+        ↓
+    register correlation  (provider="gitea", event_type="ci_completed",
+                           run_ref="ci-<sha>", sha=<sha>)
+        ↓
+    WAITING_FOR_EVENT     (await_resume_event: select() on FIFO, no floor)
+        ↓
+    ci.yml final job POSTs /process-event  (the WAKE)
+        ↓
+    daemon matches → resume_cmd fires → FIFO token → controller resumes
+        ↓
+    canonical Gitea state INDEPENDENTLY re-read (deploy_qualify.py verdict)
+        ↓
+    verdict
+
+The event is a WAKE, never truth. The final verdict always comes from
+deploy_qualify.py (required-checks context + exact-SHA ci.yml run, both
+legs). Watchdog deadline is a FAILURE DETECTOR, never the scheduler:
+
+  - event arrives → authority re-read; pending → bounded re-arm until the
+    overall deadline (ack-window semantics, same as the daemon's wake sweep).
+  - event lost/suppressed → watchdog expires → ONE-SHOT authority query:
+    terminal verdict → recovered truthfully (no invented wait);
+    CI still running (non-terminal Gitea runs for the SHA) → re-arm;
+    nothing alive and no verdict → truthful failure (deployment refused).
+
+Exit codes: 0 = qualified (success), 1 = refused (failure/timeout).
+Output: one JSON line with the §16D latency-proof timestamps.
+"""
+from __future__ import annotations
+
+import json
+import os
+import subprocess
+import sys
+import time
+from pathlib import Path
+
+_SCRIPT_DIR = Path(__file__).resolve().parent
+sys.path.insert(0, str(_SCRIPT_DIR))
+sys.path.insert(0, str(_SCRIPT_DIR.parent / "src"))
+
+sys.path.insert(0, str(_SCRIPT_DIR))
+from rune import event_wait as ew  # noqa: E402
+import completion_file as cf  # noqa: E402
+
+REPO_ROOT = _SCRIPT_DIR.parent
+QUALIFY_SCRIPT = _SCRIPT_DIR / "deploy_qualify.py"
+
+
+def authority_verdict(sha: str = "") -> str:
+    """ONE-SHOT canonical Gitea verdict via deploy_qualify.py — the only
+    completion truth. Never parses the event payload.
+
+    H-PATCH H4: the target sha is passed EXPLICITLY (as the subprocess's
+    --sha argument) — ambient DEPLOY_SHA can no longer change which
+    operation is verified. The env var remains only as a CLI backward
+    compat for deploy_qualify.py invoked standalone."""
+    cmd = [sys.executable, str(QUALIFY_SCRIPT)]
+    if sha:
+        cmd += ["--sha", sha]
+    proc = subprocess.run(
+        cmd,
+        capture_output=True, text=True, timeout=120,
+        env={k: v for k, v in os.environ.items() if k != "DEPLOY_SHA"}
+        if sha else dict(os.environ, PYTHONPATH=""),
+    )
+    out = (proc.stdout or "").strip().splitlines()
+    return (out[-1] if out else "pending").strip()
+
+
+# ── T21.5 §4: operation-specific completion authority ──────────────────────
+#
+# T21.4 root cause: the wait loop had ONE hard-coded authority
+# (deploy qualification = master-ref CI) while operations waited on
+# branch CI, PR checks, or merges. Wrong authority => nonterminal
+# returns => the MODEL had to re-arm (the generator of the sleeps).
+# The wait verifier MUST match the operation being waited for.
+
+OPERATION_SPECS = {
+    # op_type: {
+    #   workflow_path: Actions run path filter (None = any),
+    #   head_branches: head-branch filter (None = any),
+    #   authority: which canonical verifier decides completion,
+    #   event_types: expected wake event types,
+    #   terminal: terminal verdicts, failure: failure verdicts,
+    # }
+    "branch_ci": {
+        "workflow_path": "ci.yml",
+        "head_branches": None,
+        "authority": "ci_runs",          # any completed run for this SHA
+        "event_types": ("ci_completed",),
+        "terminal": ("success",),
+        "failure": ("failure", "cancelled"),
+    },
+    "pull_request_checks": {
+        "workflow_path": None,           # all PR-triggered runs for SHA
+        "head_branches": None,
+        "authority": "ci_runs",
+        "event_types": ("ci_completed",),
+        "terminal": ("success",),
+        "failure": ("failure", "cancelled"),
+    },
+    "master_ci": {
+        "workflow_path": "ci.yml",
+        "head_branches": ("master",),
+        "authority": "deploy_qualify",   # full two-leg qualification
+        "event_types": ("ci_completed",),
+        "terminal": ("success",),
+        "failure": ("failure", "cancelled"),
+    },
+    "deploy": {
+        "workflow_path": "deploy.yml",
+        "head_branches": ("master",),
+        "authority": "deploy_qualify",
+        "event_types": ("ci_completed", "deploy_completed"),
+        "terminal": ("success",),
+        "failure": ("failure", "cancelled"),
+    },
+    "release": {
+        "workflow_path": None,
+        "head_branches": ("master",),
+        "authority": "deploy_qualify",
+        "event_types": ("ci_completed", "release_completed"),
+        "terminal": ("success",),
+        "failure": ("failure", "cancelled"),
+    },
+}
+
+DEFAULT_OP_TYPE = "master_ci"
+
+
+def normalize_verdict(verdict: str) -> str:
+    """T21.5 §18 audit fix: pending-class verdicts are ALL nonterminal.
+    deploy_qualify emits "pending (probe failed)" for a transient probe
+    failure — that is pending, never a terminal answer. The wait loop
+    must not return control to the model over a string variant."""
+    v = (verdict or "").strip().lower()
+    if v.startswith("pending"):
+        return "pending"
+    return (verdict or "").strip()
+
+
+def op_spec(op_type: str) -> dict:
+    spec = OPERATION_SPECS.get(op_type)
+    if spec is None:
+        raise ValueError(
+            f"unknown op_type {op_type!r} — must be one of "
+            f"{sorted(OPERATION_SPECS)} (the wait verifier must match "
+            "the operation being waited for)")
+    return spec
+
+
+def op_authority_verdict(op_type: str, sha: str) -> str:
+    """Canonical completion verdict FOR THIS OPERATION TYPE.
+
+    ci_runs        — the op's own Actions runs for this exact SHA: every
+                     required workflow run completed; terminal iff every
+                     run of the op's workflow set concluded success
+                     (branch/PR ops are their own truth; deploy
+                     qualification is NOT their authority).
+    deploy_qualify — the two-leg deploy qualification (master CI +
+                     required-checks) via deploy_qualify.py.
+    """
+    spec = op_spec(op_type)
+    authority = spec["authority"]
+    if authority == "deploy_qualify":
+        v = normalize_verdict(authority_verdict(sha))
+        if v != "pending":
+            return v
+        # E-PATCH resilience: Gitea's combined commit-status endpoint can
+        # 404 for the newest commit (observed live on auto-version bump
+        # commits — the status object materializes late/never). The runs
+        # ARE the same evidence the two legs read; fall back to the
+        # operation-scoped runs verdict so authority stays resolvable.
+        if _status_endpoint_unreadable(sha):
+            return _ci_runs_verdict(spec, sha)
+        return v
+    if authority == "ci_runs":
+        return _ci_runs_verdict(spec, sha)
+    raise ValueError(f"unknown authority {authority!r} for {op_type!r}")
+
+
+def _status_endpoint_unreadable(sha: str) -> bool:
+    """True when the combined commit-status endpoint 404s for the SHA
+    (index lag / never-materialized status object) — the documented
+    fallback trigger, verified against the API once per call."""
+    try:
+        sys.path.insert(0, str(_SCRIPT_DIR))
+        import gitea_auth  # noqa: E402
+        st, _body = gitea_auth.api(f"/commits/{sha}/status")
+        return st == 404
+    except Exception:  # noqa: BLE001
+        return False
+
+
+def _ci_runs_verdict(spec: dict, sha: str) -> str:
+    """Operation-scoped Actions-run verdict (import the API client from
+    deploy_qualify to keep ONE Gitea client)."""
+    sys.path.insert(0, str(_SCRIPT_DIR))
+    import deploy_qualify as dq  # noqa: E402
+    import gitea_auth  # noqa: E402
+    server = os.environ.get("GITEA_SERVER",
+                            getattr(gitea_auth, "DEFAULT_SERVER", "")).rstrip("/")
+    repo = os.environ.get("GITEA_REPO",
+                          getattr(gitea_auth, "DEFAULT_REPO", ""))
+    sha_env = sha  # H-PATCH: explicit parameter is the only authority
+    token = gitea_auth.get_token() if gitea_auth.has_token() else ""
+    if not (server and repo and sha_env and token):
+        return "pending"  # cannot verify -> not verified (fail closed)
+    runs = dq._fetch_runs(server, repo, sha_env, token)
+    if runs is None:
+        return "pending"  # probe failure -> keep waiting
+    wfp = spec["workflow_path"]
+    if wfp is not None:
+        runs = [r for r in runs
+                if str(r.get("path", "")).startswith(wfp + "@")]
+    branches = spec["head_branches"]
+    if branches is not None:
+        runs = [r for r in runs
+                if str(r.get("head_branch", "")) in branches]
+    if not runs:
+        return "pending"  # no runs visible yet -> not terminal
+    for r in runs:
+        if r.get("status") != "completed":
+            return "pending"  # still running
+        if r.get("conclusion") not in ("success", "skipped"):
+            return "failure"  # any non-success conclusion
+    return "success"
+
+
+def op_ci_alive(op_type: str, sha: str) -> bool:
+    """Liveness probe FOR THIS OPERATION TYPE: any non-terminal run in
+    the op's run set (failure DETECTOR, never the scheduler)."""
+    sys.path.insert(0, str(_SCRIPT_DIR))
+    import deploy_qualify as dq  # noqa: E402
+    import gitea_auth  # noqa: E402
+    spec = op_spec(op_type)
+    server = os.environ.get("GITEA_SERVER",
+                            getattr(gitea_auth, "DEFAULT_SERVER", "")).rstrip("/")
+    repo = os.environ.get("GITEA_REPO",
+                          getattr(gitea_auth, "DEFAULT_REPO", ""))
+    sha_env = sha  # H-PATCH: explicit parameter is the only authority
+    token = gitea_auth.get_token() if gitea_auth.has_token() else ""
+    if not (server and repo and sha_env and token):
+        return False
+    runs = dq._fetch_runs(server, repo, sha_env, token)
+    if not runs:
+        return False
+    wfp = spec["workflow_path"]
+    if wfp is not None:
+        runs = [r for r in runs
+                if str(r.get("path", "")).startswith(wfp + "@")]
+    branches = spec["head_branches"]
+    if branches is not None:
+        runs = [r for r in runs
+                if str(r.get("head_branch", "")) in branches]
+    terminal = {"success", "failure", "cancelled", "skipped"}
+    return any(str(r.get("status", "")).lower() not in terminal
+               for r in runs)
+
+
+def authority_after_wake(op_type: str = DEFAULT_OP_TYPE, sha: str = "",
+                         max_tries: int = 3, gap_secs: int = 4) -> str:
+    """(sha is EXPLICIT — ambient env never selects the target.)"""
+    """Post-wake authority read. The daemon's own wake sweep uses a
+    bounded event-triggered retry (4s ×3) for the job-finished-microseconds-
+    ago ack window; a verdict read here can hit the same window. This is a
+    bounded schedule, not a polling loop — 3 reads max, then back to the
+    event wait / watchdog."""
+    verdict = "pending"
+    for attempt in range(max_tries):
+        verdict = op_authority_verdict(op_type, sha)
+        if verdict != "pending":
+            return verdict
+        if attempt < max_tries - 1:
+            time.sleep(gap_secs)
+    return verdict
+
+
+def ci_alive() -> bool:
+    """LEGACY liveness probe (master-ref deploy runs) — kept for callers
+    predating op specs; run() uses op_ci_alive(op_type, sha)."""
+    return op_ci_alive(DEFAULT_OP_TYPE, os.environ.get("DEPLOY_SHA", ""))
+
+
+def _set_cognition_state(op: str, state: str) -> None:
+    """E-PATCH §12: stamp cognitive quiescence on the durable record so a
+    fresh process knows cognition is parked (WAITING_EXTERNAL) vs resumable
+    (RESUMABLE). Best-effort: observability never breaks the wait."""
+    try:
+        rec = ew.operation_wait(op)
+        if rec is None:
+            return
+        wait_file = ew.wr_root() / "waits" / f"{rec.get('wait_id')}.json"
+        if not wait_file.is_file():
+            return
+        data = json.loads(wait_file.read_text())
+        data.setdefault("ext", {})["continuation_state"] = state
+        wait_file.write_text(json.dumps(data, indent=1, sort_keys=True))
+    except (OSError, ValueError):
+        pass
+
+
+def run(*, sha: str, deadline_secs: int,
+        watchdog_secs: int = 300,
+        op_type: str = DEFAULT_OP_TYPE,
+        task_id: str = "",
+        daemon_url: str = ew.DEFAULT_DAEMON_URL) -> int:
+    """T21.5 §3: ONE LOGICAL AWAIT.
+
+    One logical external wait is one invocation of this function. All
+    nonterminal conditions (event-seen-authority-pending, watchdog expiry
+    with live runs, bounded authority-retry exhausted) re-arm INTERNALLY —
+    control returns to the caller ONLY on a terminal verdict or on real
+    deadline expiry. Watchdogs are failure detectors inside this
+    abstraction, never the scheduler and never completion truth.
+    """
+    spec = op_spec(op_type)          # unknown op_type refused upfront
+    op = f"ci-{sha}"
+    # A-PATCH: the authority subprocess reads DEPLOY_SHA from the env;
+    # python-level run() callers (tool normalization flow) must export it.
+    if sha and not os.environ.get("DEPLOY_SHA"):
+        os.environ["DEPLOY_SHA"] = sha
+    # H-PATCH: resolve short SHAs to full 40-char — _fetch_runs and
+    # deploy_qualify match on the FULL sha; a 12-char prefix would
+    # silently fail equality and park the waiter until deadline
+    # (the 962e351 root cause).
+    if sha and len(sha) < 40:
+        try:
+            full = subprocess.run(
+                ["git", "-C", str(_SCRIPT_DIR.parent), "rev-parse", sha],
+                capture_output=True, text=True, timeout=10)
+            if full.returncode == 0:
+                sha = full.stdout.strip()
+                os.environ["DEPLOY_SHA"] = sha
+        except (OSError, subprocess.TimeoutExpired):
+            pass
+    t0 = time.time()
+    # REAL_ASYNC_BRIDGE: FIFO is replaced by completion-file inotify.
+    # The completion file write is the wake signal; inotify is the
+    # kernel-level zero-latency, zero-CPU wake primitive.
+    # Event replay (T16 §16C live finding): on a serialized runner the
+    # completion event usually arrives BEFORE the waiter registers.
+    # A matching durable event means the wake already happened — skip
+    # the rendezvous entirely and go straight to the authority read.
+    prior = ew.find_matching_event("gitea", "ci_completed", op)
+    if prior is not None and prior.get("exit_code") is not None:
+        rec = ew.register_process_wait(
+            op, project="rune",
+            objective=f"event-driven {op_type} qualification for {sha}",
+            watched_for="Gitea CI completion event (replayed from the "
+                        "durable event log)",
+            provider="gitea", event_type="ci_completed", sha=sha,
+            watchdog_secs=min(watchdog_secs, deadline_secs),
+            continuation_state="RESUMABLE",  # event already arrived
+            op_type=op_type,
+            authority=spec["authority"])
+        registered_at = rec["created_at"]
+        event_at = time.strftime(
+            "%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+        resumed_at = event_at
+        # the event is a WAKE, never truth — canonical authority decides
+        verdict = normalize_verdict(authority_after_wake(op_type, sha))
+        if verdict == "pending" and not op_ci_alive(op_type, sha):
+            verdict = normalize_verdict(op_authority_verdict(op_type, sha))
+        mode = ("event_replayed" if verdict != "pending"
+                else "authority_unresolved")
+        verdict_at = time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+        out = {
+            "sha": sha,
+            "op_type": op_type,
+            "authority": spec["authority"],
+            "registered_at": registered_at,
+            "event_received_at": event_at,
+            "controller_resumed_at": resumed_at,
+            "verdict_at": verdict_at,
+            "mode": mode,
+            "rearm_reason": "completion event replayed from durable log "
+                            "(arrived before registration)",
+            "verdict": verdict,
+            "wall_secs": round(time.time() - t0, 3),
+        }
+        print(json.dumps(out))
+        return 0 if verdict == "success" else 1
+    rec = ew.register_process_wait(
+        op, project="rune",
+        objective=f"event-driven {op_type} qualification for {sha}",
+        watched_for="Gitea CI completion event (wake; verdict re-read "
+                    f"from canonical {spec['authority']} authority)",
+        provider="gitea", event_type="ci_completed", sha=sha,
+        watchdog_secs=min(watchdog_secs, deadline_secs),
+        continuation_state="WAITING_EXTERNAL",
+        op_type=op_type,
+        authority=spec["authority"],
+        task_id=task_id,  # §9: bind the operation to the calling task
+        resume_mode="self")  # §3: waiter owns its continuation via
+                             # completion-file/inotify; the daemon has NO
+                             # session-delivery role for this wait
+    registered_at = rec["created_at"]
+    _set_cognition_state(op, "WAITING_EXTERNAL")  # cognition parked
+    deadline = time.monotonic() + deadline_secs
+    verdict = "pending"
+    mode = "event"
+    event_at = ""
+    resumed_at = ""
+    rearm_reason = ""
+    rearm_count = 0
+    while time.monotonic() < deadline:
+            remaining = max(1, int(deadline - time.monotonic()))
+
+            # REAL_ASYNC_BRIDGE: block on inotify for the completion file.
+            # Zero CPU, zero arbitrary timer. The inotify timeout is a
+            # FAILURE DETECTOR — if no event arrives within the watchdog
+            # window, we check the authority directly (lost-event recovery).
+            # The deadline is containment only.
+            watch_timeout = min(remaining, watchdog_secs)
+            result = cf.wait_for_completion(
+                op, timeout_secs=watch_timeout)
+
+            # REAL_ASYNC_BRIDGE: check durable event log AND wait record
+            # status on EVERY cycle — the daemon may have matched and
+            # transitioned the record via /process-event.
+            wr = ew.operation_wait(op)
+            if wr and wr.get("status") in ("COMPLETED", "FAILED",
+                                           "READY_TO_RESUME"):
+                event_at = event_at or time.strftime(
+                    "%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+                resumed_at = time.strftime("%Y-%m-%dT%H:%M:%SZ",
+                                           time.gmtime())
+                verdict = normalize_verdict(
+                    authority_after_wake(op_type, sha))
+                if verdict != "pending":
+                    mode = "event"
+                    break
+            prior = ew.find_matching_event("gitea", "ci_completed", op)
+            if prior is not None and prior.get("exit_code") is not None:
+                event_at = event_at or time.strftime(
+                    "%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+                resumed_at = time.strftime("%Y-%m-%dT%H:%M:%SZ",
+                                           time.gmtime())
+                verdict = normalize_verdict(
+                    authority_after_wake(op_type, sha))
+                if verdict != "pending":
+                    mode = "event"
+                    break
+            if result is not None:
+                # completion file exists: event arrived
+                event_at = event_at or time.strftime(
+                    "%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+                resumed_at = time.strftime("%Y-%m-%dT%H:%M:%SZ",
+                                           time.gmtime())
+                verdict = normalize_verdict(
+                    authority_after_wake(op_type, sha))
+                if verdict != "pending":
+                    mode = "event"
+                    break
+                rearm_count += 1
+                rearm_reason = ("completion file seen; authority "
+                                "pending — re-subscribing")
+                continue
+
+            # inotify timeout: no event within watchdog window
+            # FAILURE DETECTOR: check authority directly
+            verdict = normalize_verdict(
+                op_authority_verdict(op_type, sha))
+            if verdict != "pending":
+                mode = "watchdog_authority_recovered"
+                break
+            rearm_count += 1
+            rearm_reason = ("watchdog expiry; op pending (not started "
+                            "or lagging) — re-subscribing inotify")
+            continue
+
+    # REAL_ASYNC_BRIDGE: no FIFO to close. The completion file is durable
+    # and survives restarts. Lease release is not needed (no lease system
+    # for file-based delivery).
+
+
+    if verdict == "pending":
+        if mode == "event":
+            mode = "deadline_pending"
+        verdict = normalize_verdict(op_authority_verdict(op_type, sha))  # final one-shot truth
+    _set_cognition_state(op, "RESUMABLE" if verdict != "pending"
+                         else "RECOVERY_REQUIRED")
+    verdict_at = time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+    # §2 STATE OWNERSHIP: the WAITER owns the OPERATION domain. On a
+    # terminal authority verdict, stamp the durable record terminal with
+    # the evidence — the daemon's session-delivery machinery must never
+    # (and can no longer) overwrite this truth. Also updates the parked
+    # continuation_state so fresh-process recovery sees reality.
+    try:
+        rec2 = ew.operation_wait(op) or rec
+        if verdict != "pending":
+            rec2["status"] = "COMPLETED" if verdict == "success" else "FAILED"
+            rec2["verdict"] = verdict
+            rec2["resume_evidence"] = {
+                "kind": "authority_verified_terminal",
+                "verdict": verdict,
+                "mode": mode,
+                "authority": spec["authority"],
+                "sha": sha,
+                "op_type": op_type,
+                "verdict_at": verdict_at,
+            }
+        rec2["ext"] = {**rec2.get("ext", {}),
+                       "continuation_state": ("RESUMABLE" if verdict != "pending"
+                                              else "RECOVERY_REQUIRED")}
+        ew._load_waitstore().save(rec2)
+    except Exception as _e:  # noqa: BLE001 — record stamping must never corrupt the envelope (the envelope + authority remain the truth)
+        print(f"ci_event_wait: terminal record stamp failed: {_e}",
+              file=sys.stderr)
+    out = {
+        "sha": sha,
+        "op_type": op_type,
+        "authority": spec["authority"],
+        "registered_at": registered_at,
+        "event_received_at": event_at,
+        "controller_resumed_at": resumed_at,
+        "verdict_at": verdict_at,
+        "mode": mode,
+        "rearm_reason": rearm_reason,
+        "internal_rearm_count": rearm_count,
+        "verdict": verdict,
+        "wall_secs": round(time.time() - t0, 3),
+    }
+    print(json.dumps(out))
+    return 0 if verdict == "success" else 1
+
+KNOWN_FLAGS = ("--sha", "--deadline", "--op-type", "--watchdog", "--task-id")
+
+
+def main() -> int:
+    """§5 CLI CONTRACT: no silent argument loss. Every flag is either
+    explicitly supported (and takes effect) or loudly rejected — an
+    acceptance envelope must reflect the configuration actually used."""
+    sha = os.environ.get("DEPLOY_SHA", "")
+    args = sys.argv[1:]
+    deadline = 900
+    op_type = DEFAULT_OP_TYPE
+    watchdog = 300
+    task_id = ""
+    i = 0
+    while i < len(args):
+        a = args[i]
+        if a == "--sha":
+            sha = args[i + 1]; i += 2
+        elif a == "--deadline":
+            deadline = int(args[i + 1]); i += 2
+        elif a == "--op-type":
+            op_type = args[i + 1]; i += 2
+        elif a == "--watchdog":
+            watchdog = int(args[i + 1]); i += 2
+        elif a == "--task-id":
+            task_id = args[i + 1]; i += 2
+        elif a.startswith("--"):
+            print(json.dumps({"error": f"unknown flag {a!r}",
+                              "known_flags": list(KNOWN_FLAGS),
+                              "contract": "no silent argument loss"}))
+            return 2
+        else:
+            print(json.dumps({"error": f"unexpected positional {a!r}",
+                              "known_flags": list(KNOWN_FLAGS)}))
+            return 2
+    if not sha:
+        print(json.dumps({"error": "no sha (DEPLOY_SHA or --sha required)"}))
+        return 1
+    # A-PATCH fix: run()'s authority subprocess (deploy_qualify.py) reads
+    # DEPLOY_SHA from the environment — a CLI `--sha` consumer previously
+    # got a silent "pending" verdict forever because the var was never
+    # exported. Export it so the canonical authority can resolve.
+    os.environ["DEPLOY_SHA"] = sha
+    return run(sha=sha, deadline_secs=deadline, op_type=op_type,
+               watchdog_secs=watchdog, task_id=task_id)
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,195 @@
+"""completion_file.py — REAL_ASYNC_BRIDGE: durable completion-file delivery
+with inotify wake. Replaces FIFO+lease for process-wait delivery.
+
+DESIGN:
+  Producer (bridge/daemon/notify):
+    1. Validates operation identity
+    2. Atomically writes a completion file:
+         resumes/<operation_id>.done
+       containing the event/verdict JSON (write to tmp, rename for atomicity)
+
+  Waiter (ci_event_wait / any continuation consumer):
+    1. Checks if the completion file already exists (event-before-waiter)
+    2. If not: registers inotify watch on the resumes/ directory
+    3. Re-checks after watch registration (closes the registration race)
+    4. Blocks in kernel on inotify — zero CPU, zero timer
+    5. On file-created event: reads the file, verifies authority, returns
+
+  The event is a WAKE SIGNAL. Authority still decides terminal truth.
+  The completion file is durable — survives daemon and waiter restarts.
+
+WHY NOT FIFO:
+  FIFO is one-shot (token consumed), requires lease management, and
+  fails under session-busy guards. A completion file is atomic, durable,
+  and works across restarts. inotify wakes the process with zero CPU
+  cost and zero latency between file creation and wake.
+
+WHY NOT POLLING:
+  Polling introduces an arbitrary interval. inotify is a kernel callback
+  that fires exactly when the filesystem changes — the latency tracks
+  the real completion event, not a configured interval.
+"""
+from __future__ import annotations
+
+import json
+import os
+import select
+import struct
+import time
+from pathlib import Path
+
+def _wr_bridge() -> Path:
+    # WR_ROOT = daemon/infra convention; RUNE_WR_ROOT = repo/scripts
+    # convention. Both honored (WR_ROOT wins) so the daemon, the waiter,
+    # and any isolated test root always resolve the SAME directory.
+    root = (os.environ.get("WR_ROOT")
+            or os.environ.get("RUNE_WR_ROOT")
+            or "/mnt/workdrive/wr-bridge")
+    return Path(root)
+
+def _resumes_dir() -> Path:
+    return _wr_bridge() / "resumes"
+
+# inotify constants (Linux)
+IN_CREATE = 0x100
+IN_CLOSE_WRITE = 0x8
+IN_MOVED_TO = 0x80
+
+
+def completion_path(operation_id: str) -> Path:
+    return _resumes_dir() / f"{operation_id}.done"
+
+
+def write_completion(operation_id: str, event: dict) -> Path:
+    """Atomically write a durable completion file. The rename(2) ensures
+    the waiter never sees a partial file."""
+    _resumes_dir().mkdir(parents=True, exist_ok=True)
+    target = completion_path(operation_id)
+    tmp = target.parent / f".{target.name}.tmp.{os.getpid()}"
+    tmp.write_text(json.dumps(event, sort_keys=True) + "\n")
+    os.rename(str(tmp), str(target))  # atomic on same filesystem
+    os.chmod(str(target), 0o666)  # cross-user bridge
+    return target
+
+
+def read_completion(operation_id: str) -> dict | None:
+    """Read a durable completion file. Returns None if absent."""
+    p = completion_path(operation_id)
+    if not p.is_file():
+        return None
+    try:
+        return json.loads(p.read_text())
+    except (OSError, ValueError):
+        return None
+
+
+def completion_exists(operation_id: str) -> bool:
+    return completion_path(operation_id).is_file()
+
+
+def _inotify_init_and_add_watch(directory: str) -> int:
+    """Create an inotify instance and add a watch on the directory.
+    Returns the inotify fd."""
+    libc = __import__("ctypes").CDLL(None, use_errno=True)
+    fd = libc.inotify_init()
+    if fd < 0:
+        raise OSError("inotify_init failed")
+    # IN_CREATE | IN_MOVED_TO: catch both direct creation and atomic rename
+    mask = IN_CREATE | IN_MOVED_TO
+    wd = libc.inotify_add_watch(fd, directory.encode(), mask)
+    if wd < 0:
+        os.close(fd)
+        raise OSError(f"inotify_add_watch failed for {directory}")
+    return fd
+
+
+def _inotify_read(fd: int, timeout_secs: float | None = None) -> list[str]:
+    """Read pending inotify events. Returns list of filenames that were
+    created/moved-to. Blocks until at least one event or timeout."""
+    poll = select.poll()
+    poll.register(fd, select.POLLIN)
+    events = []
+    buf = os.read(fd, 4096)
+    pos = 0
+    while pos < len(buf):
+        # struct: wd(i32), mask(u32), cookie(u32), name_len(u32)
+        wd, mask, cookie, name_len = struct.unpack_from("iIII", buf, pos)
+        pos += 16
+        name = buf[pos:pos + name_len].rstrip(b"\x00").decode("utf-8", "replace")
+        pos += name_len
+        if mask & (IN_CREATE | IN_MOVED_TO):
+            events.append(name)
+    return events
+
+
+def wait_for_completion(operation_id: str, *,
+                        timeout_secs: float | None = None,
+                        poll_callback=None) -> dict | None:
+    """Block until a completion file exists for this operation.
+
+    Uses inotify for zero-latency, zero-CPU wake. Falls back to
+    polling_callback (if provided) when inotify is unavailable.
+
+    The completion file is a WAKE SIGNAL. The caller must still verify
+    canonical authority after reading it.
+
+    Returns the event dict from the completion file, or None on timeout.
+    """
+    target = completion_path(operation_id)
+
+    # 1. check-before-watch (event may have arrived before we started)
+    existing = read_completion(operation_id)
+    if existing is not None:
+        return existing
+
+    _resumes_dir().mkdir(parents=True, exist_ok=True)
+
+    try:
+        fd = _inotify_init_and_add_watch(str(_resumes_dir()))
+    except OSError:
+        # inotify unavailable: fall back to polling callback
+        if poll_callback:
+            return poll_callback(operation_id, timeout_secs)
+        return None
+
+    try:
+        # 2. re-check after watch registration (closes the race between
+        #    the initial check and the watch being active)
+        existing = read_completion(operation_id)
+        if existing is not None:
+            return existing
+
+        # 3. block in kernel on inotify — zero CPU, exact latency
+        import select as _select
+        poll = _select.poll()
+        poll.register(fd, _select.POLLIN)
+        deadline = (time.monotonic() + timeout_secs) if timeout_secs else None
+
+        while True:
+            remaining = None
+            if deadline is not None:
+                remaining = max(0.1, deadline - time.monotonic())
+                if remaining <= 0.1:
+                    return None  # timeout
+            events = poll.poll(remaining * 1000 if remaining else None)
+            if not events:
+                continue  # poll timeout, re-check
+            names = _inotify_read(fd)
+            if target.name in names:
+                # 4. file appeared — read it
+                existing = read_completion(operation_id)
+                if existing is not None:
+                    return existing
+                # partial/corrupt: keep watching
+    finally:
+        os.close(fd)
+    return None
+
+
+def clear_completion(operation_id: str) -> None:
+    """Remove a consumed completion file (idempotent cleanup)."""
+    p = completion_path(operation_id)
+    try:
+        p.unlink()
+    except OSError:
+        pass
@@ -0,0 +1,123 @@
+"""construction_guards.py — deterministic guards for observed process
+defects (SCAFFOLD-TRANCHE-006 §21-22).
+
+Two real defect classes from recent sessions:
+
+A. DURABLE-STORE MUTATION WITHOUT PERSISTENCE (Tranche 5: closure script
+   mutated the contract in memory, never called save(); the next read saw
+   stale state). Guard: a durable-boundary operation verifies by
+   RELOAD/READBACK, not by successful in-memory mutation.
+
+B. BLIND SOURCE-EDIT NO-OP (multiple sessions: python str.replace() on
+   ruff-reformatted files silently matched nothing; the script reported
+   success while changing nothing). Guard: an automated edit verifies its
+   POSTCONDITION — expected text present / old text absent — and a
+   zero-match replacement is an explicit, truthful no-op, never success.
+
+Not a generic editor framework. Two small, deterministic postcondition
+checks at the boundaries where the defects actually occurred.
+"""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+__all__ = ["verify_persisted", "assert_persisted", "apply_source_edit", "EditNoOp", "EditPostconditionFailed"]
+
+
+class EditNoOp(Exception):
+    """The edit matched zero targets — an explicit, truthful no-op."""
+
+
+class EditPostconditionFailed(Exception):
+    """The edit claimed success but the expected postcondition is absent."""
+
+
+# ── Guard A: durable mutation verified by readback ──────────────────────────
+
+
+def verify_persisted(path: Path | str, selector, expect) -> bool:
+    """Reload the durable store from disk and check the expectation against
+    the RELOADED state. `selector(reloaded_doc) -> value`; pass == value
+    equals `expect` (or expect(value) is truthy when callable).
+
+    The reloaded document is parsed as JSON when the file is JSON-shaped;
+    otherwise the raw text is passed to the selector (plain-text stores).
+
+    This is the Tranche 5 defect guard: an in-memory mutation that was
+    never save()d fails here, because the disk state is what gets read.
+    """
+    raw = Path(path).read_text(encoding="utf-8")
+    try:
+        doc = json.loads(raw)
+    except ValueError:
+        doc = raw
+    value = selector(doc)
+    if callable(expect):
+        return bool(expect(value))
+    return value == expect
+
+
+def assert_persisted(path: Path | str, selector, expect, *, what: str = "durable state") -> None:
+    """verify_persisted that raises on mismatch — for use right after a
+    save() at durable boundaries."""
+    if not verify_persisted(path, selector, expect):
+        raise AssertionError(
+            f"persistence verification FAILED for {what}: reloaded state "
+            f"does not match expectation — the mutation may not have been "
+            f"saved (Tranche 5 defect class A)"
+        )
+
+
+# ── Guard B: source edit with postcondition ─────────────────────────────────
+
+
+def apply_source_edit(
+    text: str,
+    old: str,
+    new: str,
+    *,
+    expected_count: int | None = None,
+    postcondition: str | None = None,
+    allow_noop: bool = False,
+) -> str:
+    """Apply a source edit WITH postcondition verification (Guard B).
+
+    - zero matches: raises EditNoOp (truthful no-op) unless allow_noop=True
+    - expected_count given: match count must equal it
+    - postcondition given: the RESULT must contain it, else
+      EditPostconditionFailed
+    Returns the new text. Never claims success for a change that did not
+    happen."""
+    count = text.count(old)
+    if count == 0:
+        if allow_noop:
+            return text  # truthful, explicit no-op — caller records it
+        raise EditNoOp(
+            f"edit matched 0 targets — nothing changed (Tranche 5 defect "
+            f"class B: str.replace silently no-ops on reformatted files)"
+        )
+    if expected_count is not None and count != expected_count:
+        raise EditPostconditionFailed(f"edit matched {count} targets, expected {expected_count}")
+    result = text.replace(old, new)
+    if postcondition and postcondition not in result:
+        raise EditPostconditionFailed(
+            "postcondition text absent from result — edit did not achieve its expected effect"
+        )
+    return result
+
+
+def apply_source_edit_file(
+    path: Path | str, old: str, new: str, *, expected_count: int | None = None, postcondition: str | None = None
+) -> int:
+    """File-level convenience: read, apply_source_edit, write back, then
+    VERIFY BY READBACK (both guards composed). Returns match count."""
+    p = Path(path)
+    text = p.read_text(encoding="utf-8")
+    result = apply_source_edit(text, old, new, expected_count=expected_count, postcondition=postcondition)
+    p.write_text(result, encoding="utf-8")
+    # Guard A composed: the written file must contain the postcondition
+    if postcondition:
+        assert_persisted(p, lambda t: t, lambda t: postcondition in t, what=f"edit of {p.name}")
+    return text.count(old)
@@ -0,0 +1,224 @@
+#!/usr/bin/env python3
+"""deploy_promotion.py — monotonic promotion authority for Rune deploys.
+
+SCAFFOLD contract DEPLOY-MONOTONIC-001. Root cause of the 2026-09-13
+deploy-order race (deploys 7355/7359): deploy_stage.py promoted
+unconditionally in job-finish order — no comparison against the currently
+deployed commit, no lock, no truthful outcome for a stale candidate.
+
+This module owns the PROMOTION decision. The invariant:
+
+    current never regresses along the canonical release branch.
+
+Ordering authority: GIT ANCESTRY on the canonical release branch
+(refs/heads/master). Rune's release flow is linear history — every deploy
+candidate is a master push (tranche commit or its auto-bump descendant).
+    candidate == deployed commit      -> NOOP_ALREADY_CURRENT (idempotent)
+    candidate is DESCENDANT of it     -> allowed (normal forward deploy)
+    candidate is ANCESTOR of it       -> STALE_SUPERSEDED (refuse; truthful)
+    no ancestry relationship          -> REFUSED_DIVERGENT (fail closed)
+    deployed commit unknown/unreadable-> REFUSED_UNKNOWN_CURRENT (fail closed)
+
+Mechanism (smallest that provides the invariant, per brief §4-5):
+    1. promotion-time stale check — ancestry comparison against the commit
+       current ACTUALLY points to at promotion time (not at job start);
+    2. an flock'd critical section around the read-compare-write of
+       current/last-good so two concurrent promotions cannot interleave
+       between the ancestry check and the symlink swap.
+Workflow-level concurrency is NOT relied on (brief §5): the fence lives at
+the promotion operation itself.
+
+Truthful outcomes (brief §8) — build/stage success != promotion success:
+    PROMOTED                candidate became current (verified below)
+    NOOP_ALREADY_CURRENT    same commit already deployed; nothing mutated
+    STALE_SUPERSEDED        a newer commit is already current; nothing mutated
+    REFUSED_DIVERGENT       ordering unprovable; fail closed
+    REFUSED_UNKNOWN_CURRENT deployed commit unreadable; fail closed
+    FAILED_*                verification/staging failures (in deploy_stage)
+
+A stale or refused outcome is a NON-ERROR terminal state for the release
+pipeline (the newer build simply won) but MUST NOT print "PROMOTED" and
+MUST NOT mutate current/last-good. deploy_stage's exit code and the
+workflow's post-deploy runtime verification enforce this: the workflow
+asserts current == candidate after a PROMOTED outcome, so a stale outcome
+makes the job report honestly rather than masquerade.
+"""
+
+from __future__ import annotations
+
+import fcntl
+import json
+import os
+import subprocess
+from pathlib import Path
+
+DEPLOY_ROOT = Path("/mnt/workdrive/rune-code")
+RELEASE_BRANCH = "refs/heads/master"
+
+OUTCOME_PROMOTED = "PROMOTED"
+OUTCOME_NOOP = "NOOP_ALREADY_CURRENT"
+OUTCOME_STALE = "STALE_SUPERSEDED"
+OUTCOME_DIVERGENT = "REFUSED_DIVERGENT"
+OUTCOME_UNKNOWN_CURRENT = "REFUSED_UNKNOWN_CURRENT"
+
+
+# ── pure decision logic (unit-tested; no I/O) ──────────────────────────────
+
+
+def decide_promotion(candidate_sha: str, deployed_sha: str, ancestry: str) -> str:
+    """Pure promotion decision from a precomputed ancestry relationship.
+    ancestry is one of: 'same' | 'descendant' | 'ancestor' | 'divergent' |
+    'unknown'. Fail-closed: anything not provably allowed is refused."""
+    if candidate_sha == deployed_sha or ancestry == "same":
+        return OUTCOME_NOOP
+    if ancestry == "descendant":
+        return OUTCOME_PROMOTED
+    if ancestry == "ancestor":
+        return OUTCOME_STALE
+    if ancestry == "divergent":
+        return OUTCOME_DIVERGENT
+    return OUTCOME_UNKNOWN_CURRENT
+
+
+def ancestry_of(candidate_sha: str, deployed_sha: str, repo: Path, fetch_on_divergent: bool = True) -> str:
+    """Classify candidate vs deployed using git merge-base --is-ancestor on
+    the canonical release branch's repository. Returns 'unknown' on any git
+    failure (fail-closed callers treat that as refusal).
+
+    Observed live 2026-09-13 (deploy 7449, SHA 535c2909): a DELAYED older
+    candidate's checkout cannot contain commits newer than itself — its
+    local object store lacks the deployed commit entirely, so ancestry
+    classifies 'divergent' and fail-closed refuses. Correct refusal, wrong
+    reason: the candidate IS an ancestor. When the local classification is
+    divergent, fetch origin's canonical branch first (best-effort; the
+    deploy workflow's checkout has credentials configured) and re-classify.
+    If still divergent afterwards, the refusal stands with better
+    evidence."""
+
+    def _is_ancestor(a: str, b: str) -> bool:
+        r = subprocess.run(
+            ["git", "-C", str(repo), "-c", f"safe.directory={repo}", "merge-base", "--is-ancestor", a, b],
+            capture_output=True,
+            text=True,
+        )
+        return r.returncode == 0
+
+    if candidate_sha == deployed_sha:
+        return "same"
+    if _is_ancestor(candidate_sha, deployed_sha):
+        return "ancestor"  # candidate is OLDER than what's deployed
+    if _is_ancestor(deployed_sha, candidate_sha):
+        return "descendant"  # candidate is NEWER (normal forward deploy)
+    if fetch_on_divergent:
+        try:
+            subprocess.run(
+                ["git", "-C", str(repo), "-c", f"safe.directory={repo}", "fetch", "origin", "master", "--quiet"],
+                capture_output=True,
+                text=True,
+                timeout=60,
+            )
+        except (OSError, subprocess.TimeoutExpired):
+            pass  # best-effort: classification below still fails closed
+        if _is_ancestor(candidate_sha, deployed_sha):
+            return "ancestor"
+        if _is_ancestor(deployed_sha, candidate_sha):
+            return "descendant"
+    return "divergent"
+
+
+def read_deployed_commit(deploy_root: Path = DEPLOY_ROOT) -> str:
+    """Read the commit the `current` symlink ACTUALLY points to right now —
+    from the deployed build identity, the deployment authority (never from
+    a stale job-start artifact). Empty string if unreadable."""
+    current = deploy_root / "current"
+    try:
+        target = Path(os.readlink(current))
+    except OSError:
+        return ""
+    identity = target / "src" / "rune" / "_build_identity.json"
+    try:
+        data = json.loads(identity.read_text(encoding="utf-8"))
+        return str(data.get("commit", ""))
+    except (OSError, ValueError):
+        return ""
+
+
+# ── the fenced promotion operation ──────────────────────────────────────────
+
+
+def promote(
+    candidate_sha: str,
+    stage: Path,
+    tmp_stage: Path,
+    repo: Path,
+    deploy_root: Path = DEPLOY_ROOT,
+    on_promoted=None,
+) -> tuple[str, str]:
+    """Atomically decide + promote under an exclusive lock.
+
+    Returns (outcome, detail). ONLY the PROMOTED outcome mutates
+    current/last-good, and it does so inside the same critical section as
+    the decision — no interleaving window between check and swap.
+
+    Crash-safety ordering (brief §16): the stage rename happens BEFORE the
+    lock (a staged-but-unpromoted tree is inert), the symlink swap is a
+    single rename(2) via a temporary link, so current is never absent and
+    never half-updated. last-good is refreshed BEFORE the swap only from a
+    value that was genuinely current a moment ago.
+    """
+    deploy_root.mkdir(parents=True, exist_ok=True)
+    lock_path = deploy_root / ".promotion.lock"
+    with open(lock_path, "w") as lock_fd:
+        fcntl.flock(lock_fd, fcntl.LOCK_EX)
+        try:
+            deployed = read_deployed_commit(deploy_root)
+            if not deployed:
+                return (OUTCOME_UNKNOWN_CURRENT, "deployed commit unreadable — fail closed")
+            rel = ancestry_of(candidate_sha, deployed, repo)
+            outcome = decide_promotion(candidate_sha, deployed, rel)
+            if outcome != OUTCOME_PROMOTED:
+                return (outcome, f"deployed={deployed[:12]} ancestry={rel}")
+
+            # --- mutate (still inside the lock) ---
+            current = deploy_root / "current"
+            last_good = deploy_root / "last-good"
+            prev_target = os.readlink(current) if current.is_symlink() else ""
+            if prev_target and (deploy_root / "last-good").is_symlink():
+                (deploy_root / "last-good").unlink()
+            if prev_target:
+                _atomic_symlink(deploy_root, "last-good", prev_target)
+            if stage.exists():
+                # re-deploying an already-staged SHA: restage over it
+                import shutil
+
+                shutil.rmtree(stage)
+            os.rename(tmp_stage, stage)
+            _atomic_symlink_target(deploy_root, "current", stage)
+            detail = f"current -> {stage.name}"
+            if on_promoted is not None:
+                # §12C: race-safe convergence — the WINNER refreshes release
+                # history atomically with its promotion, inside the lock.
+                # Failure degrades to STALE (software valid, history lags;
+                # retry by any later deploy converges idempotently).
+                try:
+                    on_promoted(candidate_sha)
+                    detail += " (+ release-history converged)"
+                except Exception as exc:  # noqa: BLE001
+                    detail += f" (release-history refresh FAILED: {exc!s:.120} — STALE, retry converges)"
+            return (OUTCOME_PROMOTED, detail)
+        finally:
+            fcntl.flock(lock_fd, fcntl.LOCK_UN)
+
+
+def _atomic_symlink(deploy_root: Path, name: str, target: str) -> None:
+    """Symlink swap that never leaves `name` absent: write temp link, rename
+    over the final name (rename(2) is atomic on the same filesystem)."""
+    _atomic_symlink_target(deploy_root, name, Path(target))
+
+
+def _atomic_symlink_target(deploy_root: Path, name: str, target: Path) -> None:
+    tmp = deploy_root / f".{name}.swap-{os.getpid()}"
+    if tmp.is_symlink() or tmp.exists():
+        tmp.unlink()
+    tmp.symlink_to(target)
+    os.replace(tmp, deploy_root / name)
@@ -6,6 +6,16 @@ has no embedded multi-line scripts — a YAML block-scalar indentation hazard).

 Prints one of: success | failure | pending
 Exit code 0 always (the shell decides what to do with the verdict).
+
+RELEASE-SAFETY HARDENING (2026-09-13, after red-CI promotions 0.66.28/29):
+the verdict is the AND of two independent legs —
+  1. required-checks context (newest observation for this exact SHA);
+  2. the latest COMPLETED ci.yml@refs/heads/master run for this exact SHA,
+     every job concluded success (branch+tag fan-out resolved by path;
+     jobs envelope unwrapped — no aggregate-commit-status fallback, the
+     exact bug class verify.py hit in W/R-v0.3).
+Any disagreement, pending, or missing evidence → pending/failure.
+FAIL-CLOSED: not verified → not deployable.
 """

 from __future__ import annotations
@@ -20,34 +30,146 @@ if _SCRIPT_DIR not in _sys.path:
    _sys.path.insert(0, _SCRIPT_DIR)
 import gitea_auth

+REQUIRED_CONTEXT = "CI Pipeline / required-checks"
+CI_WORKFLOW_PATH = "ci.yml@refs/heads/master"
+
+
+# ── pure verdict logic (unit-tested; fixtures from the real incidents) ────
+
+def required_check_verdict(statuses: list[dict]) -> str:
+    """Newest non-skipped required-checks observation for the SHA.
+    Missing → pending. Newest wins (stale success never overrides a newer
+    failure, and vice versa)."""
+    candidates = [
+        s for s in statuses
+        if str(s.get("context", "")).startswith(REQUIRED_CONTEXT)
+        and s.get("status") != "skipped"
+    ]
+    if not candidates:
+        return "pending"
+    newest = max(candidates, key=lambda s: str(s.get("created_at", "")))
+    return str(newest.get("status", "pending"))
+
+
+def independent_jobs_verdict(runs: list[dict], jobs_fetch, sha: str = "") -> str:
+    """Latest COMPLETED ci.yml@refs/heads/master run for the exact SHA:
+    every job must conclude success. Fan-out-safe (tag runs and sibling
+    workflows are excluded by path); the run is resolved by head_sha, never
+    by 'latest branch state'. Missing run → pending. Any non-success
+    conclusion → failure."""
+    mine = [r for r in runs if r.get("path") == CI_WORKFLOW_PATH]
+    if sha:
+        mine = [r for r in mine if str(r.get("head_sha", "")) == sha]
+    completed = [r for r in mine if r.get("status") == "completed"]
+    if not completed:
+        return "pending"
+    latest = max(completed, key=lambda r: str(r.get("created_at", "") or r.get("run_started_at", "")))
+    jobs = jobs_fetch(latest.get("id"))
+    if jobs is None:
+        return "pending"  # jobs probe failed → not verified
+    if not jobs:
+        return "pending"  # no jobs visible → cannot verify
+    for j in jobs:
+        if j.get("conclusion") != "success":
+            return "failure"
+    return "success"
+
+
+def combine(a: str, b: str) -> str:
+    """Fail-closed conjunction: any failure → failure; any pending → pending;
+    success only when BOTH legs are success."""
+    if "failure" in (a, b):
+        return "failure"
+    if "pending" in (a, b):
+        return "pending"
+    return "success"
+
+
+def qualify(statuses: list[dict], runs: list[dict], jobs_fetch, sha: str = "") -> str:
+    return combine(required_check_verdict(statuses),
+                   independent_jobs_verdict(runs, jobs_fetch, sha=sha))
+
+
+# ── API fetches (thin; injectable in tests) ───────────────────────────────
+
+def _get_json(url: str, token: str, timeout: int = 30):
+    req = urllib.request.Request(url, headers={"Authorization": f"token {token}"})
+    try:
+        with urllib.request.urlopen(req, timeout=timeout) as r:
+            return json.loads(r.read())
+    except (OSError, ValueError):
+        return None
+
+
+def _fetch_status(server: str, repo: str, sha: str, token: str) -> list[dict] | None:
+    data = _get_json(f"{server}/api/v1/repos/{repo}/commits/{sha}/status", token)
+    return None if data is None else data.get("statuses", [])
+
+
+def _fetch_runs(server: str, repo: str, sha: str, token: str,
+                max_pages: int = 4) -> list[dict] | None:
+    """H-PATCH: paginate the runs list. The repo's scheduled merge-robot
+    runs flood the first page (~50 slots exhaust within hours), pushing
+    older-but-relevant SHAs to later pages — the cause of the observed
+    authority flapping (terminal commit read as pending). Paginate until
+    the sha is found or max_pages exhausted."""
+    out: list[dict] = []
+    for page in range(1, max_pages + 1):
+        data = _get_json(
+            f"{server}/api/v1/repos/{repo}/actions/runs?limit=50&page={page}",
+            token)
+        if data is None:
+            return None if page == 1 else (out or None)
+        runs = data.get("workflow_runs", data if isinstance(data, list) else [])
+        out.extend(r for r in runs if str(r.get("head_sha", "")) == sha)
+        if len(runs) < 50:
+            break  # exhausted
+    return out
+
+
+def _fetch_jobs(server: str, repo: str, run_id, token: str) -> list[dict] | None:
+    data = _get_json(f"{server}/api/v1/repos/{repo}/actions/runs/{run_id}/jobs", token)
+    if data is None:
+        return None
+    # jobs envelope unwrap (this Gitea wraps in {"jobs": [...]}; some
+    # versions return a bare list) — never fall back to aggregate status
+    if isinstance(data, dict):
+        return data.get("jobs", [])
+    return data if isinstance(data, list) else None
+
+
+# ── CLI ───────────────────────────────────────────────────────────────────

 def main() -> int:
-    server = os.environ.get("GITEA_SERVER", "").rstrip("/")
-    repo = os.environ.get("GITEA_REPO", "")
-    sha = os.environ.get("DEPLOY_SHA", "")
+    # T21.4 fix: fall back to the canonical defaults when the caller did
+    # not export GITEA_SERVER/GITEA_REPO (CI always exports them; local
+    # and first-class-tool consumers previously got a silent "pending"
+    # verdict — a wrong answer, not a fail-closed one).
+    # H-PATCH H4: the target sha arrives EXPLICITLY (--sha) from callers
+    # that hold structured operation state. DEPLOY_SHA env remains only
+    # for standalone/workflow backward compat — the canonical path never
+    # depends on ambient env for operation identity.
+    ap = __import__("argparse").ArgumentParser()
+    ap.add_argument("--sha", dest="sha_arg", default=None)
+    args, _unknown = ap.parse_known_args()
+    server = (os.environ.get("GITEA_SERVER", "")
+              or gitea_auth.DEFAULT_SERVER).rstrip("/")
+    repo = os.environ.get("GITEA_REPO", "") or gitea_auth.DEFAULT_REPO
+    sha = args.sha_arg or os.environ.get("DEPLOY_SHA", "")
    token = gitea_auth.get_token() if gitea_auth.has_token() else ""
    if not (server and repo and sha and token):
        print("pending")  # missing config → cannot verify → not qualified
        return 0
-    url = f"{server}/api/v1/repos/{repo}/commits/{sha}/status"
-    req = urllib.request.Request(url, headers={"Authorization": f"token {token}"})
-    try:
-        with urllib.request.urlopen(req, timeout=30) as r:
-            data = json.loads(r.read())
-    except (OSError, ValueError) as exc:
-        # transient API/network failure → keep polling (urllib raises
-        # URLError (OSError subclass); bad JSON raises ValueError)
-        print(f"pending (status probe failed: {exc})")
+
+    statuses = _fetch_status(server, repo, sha, token)
+    runs = _fetch_runs(server, repo, sha, token)
+    if statuses is None or runs is None:
+        # transient API/network failure → keep polling
+        print("pending (probe failed)")
        return 0
-    required = [
-        s
-        for s in data.get("statuses", [])
-        if (str(s.get("context", "")).startswith("CI Pipeline / required-checks") and s.get("status") != "skipped")
-    ]
-    if not required:
-        print("pending")
-    else:
-        print(required[0].get("status", "pending"))
+    jobs_fetch = lambda run_id: _fetch_jobs(server, repo, run_id, token)  # noqa: E731
+    verdict = qualify(statuses, runs, jobs_fetch, sha=sha)
+    print(verdict)
    return 0


@@ -3,12 +3,15 @@

 Called by .gitea/workflows/deploy.yml. Stages src/ into
 /mnt/workdrive/rune-code/<short-sha>, stamps _build_identity.json, verifies
-the staged tree (import + SHA assert + authority smoke), then ATOMICALLY
-promotes the `current` symlink and records `last-good` (safe rollback:
-the previous deployment remains available).
+the staged tree (import + SHA assert + authority smoke), then promotes
+through the MONOTONIC promotion authority (deploy_promotion.py): an older
+candidate finishing after a newer promoted one is fenced out (STALE_SUPERSEDED)
+and reports truthfully — build/stage success != promotion success.

-Exit 0 = deployed and verified. Non-zero = deployment refused; the previous
-deployment stays current.
+Exit 0 = PROMOTED and verified, or NOOP_ALREADY_CURRENT, or a truthful
+STALE_SUPERSEDED (newer build already current; nothing mutated).
+Non-zero = deployment refused (divergent/unknown ordering, verification
+failure); the previous deployment stays current.
 """

 from __future__ import annotations
@@ -25,6 +28,92 @@ DEPLOY_ROOT = Path("/mnt/workdrive/rune-code")
 REPO_ROOT = Path(__file__).resolve().parent.parent
 VENV_PY = "/home/nick/.venvs/memory/bin/python3"

+sys.path.insert(0, str(REPO_ROOT / "scripts"))
+sys.path.insert(0, str(REPO_ROOT / "src"))
+import deploy_promotion  # noqa: E402
+from rune.release_notes import (  # noqa: E402
+    ReleaseNote as _ReleaseNote,
+    auto_initial_note as _auto_note,
+    converged_deployed_sha as _authority_sha,
+    convergence_refresh as _refresh,
+    fetch_release as _fetch_release,
+    publish_release as _publish,
+    sha_from_body as _sha_from_body,
+    summary_from_body as _summary_from_body,
+)
+
+
+def publish_out_of_lock(short: str, sha: str, *,
+                        deploy_root: Path = DEPLOY_ROOT,
+                        scaffold_store: Path | None = None) -> None:
+    """§13D/§14B: release-history publication AFTER the promotion lock is
+    released. Network I/O never holds the flock.
+
+    §14B INVARIANT: publication does NOT trust the originating job's SHA.
+    The CURRENT deployment authority is re-read immediately before
+    publishing; the note is constructed from CURRENT authority. A stale
+    job whose candidate was superseded converges the NEW authority (or
+    refuses a version it cannot truthfully describe) — it may help
+    converge B, but may never reassert itself as deployed.
+    Failure degrades to STALE (software valid, history lags; retry
+    converges)."""
+    try:
+        server = os.environ.get("GITEA_SERVER", "").rstrip("/")
+        repo = os.environ.get("GITEA_REPO", "")
+        token = os.environ.get("GITEA_TOKEN", "")
+        if not (server and repo and token):
+            return
+        # §14B: re-read CURRENT authority — never trust the job's SHA.
+        authority_sha = _authority_sha(str(deploy_root))
+        if not authority_sha:
+            print("WARNING: release-history convergence skipped: "
+                  "deployment authority unreadable — STALE, retry converges")
+            return
+        store = str(scaffold_store or (REPO_ROOT / "store" / "scaffold" / "current.json"))
+        identity = deploy_root / authority_sha[:12] / "src" / "rune" / "_build_identity.json"
+        try:
+            version = json.loads(identity.read_text()).get("version", "")
+        except (OSError, ValueError):
+            version = ""
+        if not version:
+            print("WARNING: release-history convergence skipped: "
+                  "no version on current authority — STALE, retry converges")
+            return
+        tag = f"v{version}"
+        existing = _fetch_release(server, repo, token, tag)
+        if existing is None:
+            note = _auto_note(
+                version=version,
+                summary=(f"Rune v{version} — evidence-derived release "
+                         f"history (automatic convergence)"),
+                scaffold_store=store,
+                release_sha=authority_sha, deployed_sha=authority_sha,
+                provenance=["automatic convergence publish "
+                            "(deploy workflow, authority-derived)"])
+        else:
+            note = _refresh(
+                _ReleaseNote.from_dict({
+                    "version": version,
+                    "summary": _summary_from_body(existing.get("body", "")),
+                    "release_sha": _sha_from_body(existing.get("body", ""), "tag")
+                    or authority_sha,
+                    "provenance": [],
+                }),
+                deployed_sha=authority_sha, scaffold_store=store,
+                extra_provenance="automatic convergence publish "
+                                "(deploy workflow, authority-derived)")
+            # preserve the tag SHA the stored body carries
+            tag_sha = _sha_from_body(existing.get("body", ""), "tag")
+            if tag_sha:
+                note.release_sha = tag_sha
+        result = _publish(server, repo, token, note)
+        print(f"release-history convergence: "
+              f"{result.get('action', 'unknown')} "
+              f"(authority {authority_sha[:12]})")
+    except Exception as conv_exc:  # noqa: BLE001 — history STALE, software valid
+        print(f"WARNING: release-history convergence failed: "
+              f"{conv_exc!s:.200} — STALE, retry converges")
+

 def main() -> int:
    # The runner's checkout workspace (act sets GITHUB_WORKSPACE). Never a
@@ -119,25 +208,49 @@ print("VERIFIED:", json.dumps(info))
        shutil.rmtree(tmp_stage, ignore_errors=True)
        return 1

-    # --- promote (atomic) + last-known-good ------------------------------
-    DEPLOY_ROOT.mkdir(parents=True, exist_ok=True)
-    current = DEPLOY_ROOT / "current"
-    if current.is_symlink():
-        last_good = DEPLOY_ROOT / "last-good"
-        if last_good.is_symlink():
-            last_good.unlink()
-        last_good.symlink_to(os.readlink(current))
-    if current.is_symlink() or current.exists():
-        current.unlink()
-    if stage.exists():
-        shutil.rmtree(stage)  # re-deploying an already-staged SHA: restage
-    os.rename(tmp_stage, stage)
-    current.symlink_to(stage)
-    print(f"PROMOTED: {short}")
-    print("current ->", os.readlink(current))
-    if (DEPLOY_ROOT / "last-good").is_symlink():
-        print("last-good ->", os.readlink(DEPLOY_ROOT / "last-good"))
-    return 0
+    # --- promote: FENCED through the promotion authority -----------------
+    # SCAFFOLD contract DEPLOY-MONOTONIC-001: promotion is monotonic. The
+    # ancestry check + lock live in deploy_promotion.py; this script only
+    # handles the truthful outcome reporting. A stale candidate must NOT
+    # claim promotion success — build/stage success != promotion success.
+    #
+    # §14C: NO separate convergence-intent ledger. The durable pending-
+    # convergence state IS the deployment authority itself (atomically
+    # swapped `current` symlink + stamped build identity): publication
+    # derives from it, idempotently, outside the lock. The old intent
+    # file had exactly one writer and zero readers, non-atomic writes,
+    # and a bogus recorded_at (SECONDS is a bash builtin, never exported
+    # — always "0"). Deriving from authority gives atomicity, restart
+    # readability, corruption immunity, monotonicity (stale publishers
+    # read CURRENT authority, not their own job) and idempotent retry
+    # for free. REMOVE more complexity than you add.
+
+    outcome, detail = deploy_promotion.promote(
+        sha, DEPLOY_ROOT / short, tmp_stage, repo)
+
+    # §13D: publish release history OUTSIDE the promotion lock (network
+    # I/O must not hold the flock). Only for PROMOTED outcomes.
+    if outcome == deploy_promotion.OUTCOME_PROMOTED:
+        publish_out_of_lock(short, sha)
+
+    if outcome == deploy_promotion.OUTCOME_PROMOTED:
+        print(f"PROMOTED: {short}")
+        print("current ->", os.readlink(DEPLOY_ROOT / "current"))
+        if (DEPLOY_ROOT / "last-good").is_symlink():
+            print("last-good ->", os.readlink(DEPLOY_ROOT / "last-good"))
+        return 0
+    if outcome == deploy_promotion.OUTCOME_NOOP:
+        print(f"NOOP_ALREADY_CURRENT: {short} is the deployed commit")
+        print(detail)
+        return 0
+    if outcome == deploy_promotion.OUTCOME_STALE:
+        print(f"STALE_SUPERSEDED: {short} is older than the deployed commit — promotion refused (monotonic promotion)")
+        print(detail)
+        print("current unchanged:", os.readlink(DEPLOY_ROOT / "current"))
+        return 0  # truthful non-error terminal state: the newer build won
+    # divergent / unknown-current: fail closed
+    print(f"REFUSED ({outcome}): {short} — {detail}", file=sys.stderr)
+    return 1


 if __name__ == "__main__":
@@ -0,0 +1,44 @@
+#!/usr/bin/env python3
+"""install_mimo_hook.py — T21.2 §21.2D: canonical MiMo hook installer.
+
+Canonical source: config/mimo-hooks/event-wait-guard.ts
+Target: ~/.mimocode/hooks/event-wait-guard.ts
+Verifies hash after install. Detects drift in both directions."""
+import hashlib
+import shutil
+import sys
+from pathlib import Path
+
+CANONICAL = Path(__file__).resolve().parent.parent / "config" / "mimo-hooks" / "event-wait-guard.ts"
+TARGET = Path.home() / ".mimocode" / "hooks" / "event-wait-guard.ts"
+
+
+def sha256(p: Path) -> str:
+    return hashlib.sha256(p.read_bytes()).hexdigest()
+
+
+def main() -> int:
+    if not CANONICAL.is_file():
+        print(f"canonical source missing: {CANONICAL}")
+        return 1
+    canon_hash = sha256(CANONICAL)
+    if TARGET.is_file():
+        target_hash = sha256(TARGET)
+        if target_hash == canon_hash:
+            print(f"in-sync: {target_hash[:12]}")
+            return 0
+        print(f"DRIFT: canonical={canon_hash[:12]} installed={target_hash[:12]}")
+    else:
+        print(f"hook missing at {TARGET}")
+    TARGET.parent.mkdir(parents=True, exist_ok=True)
+    shutil.copy2(CANONICAL, TARGET)
+    installed_hash = sha256(TARGET)
+    if installed_hash != canon_hash:
+        print(f"INSTALL FAILED: hash mismatch after copy")
+        return 1
+    print(f"installed: {installed_hash[:12]}")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,52 @@
+#!/usr/bin/env python3
+"""install_mimo_tools.py — canonical MiMo tool installer (TTY-repair tranche).
+
+Canonical sources: config/mimo-tools/{bash.ts,await-operation.ts}
+Targets:           ~/.mimocode/tools/{bash.ts,await-operation.ts}
+Verifies hashes after install; detects drift in both directions
+(the parity test reads these hashes).
+"""
+import hashlib
+import shutil
+import sys
+from pathlib import Path
+
+REPO = Path(__file__).resolve().parent.parent
+TOOLS = ["bash.ts", "await-operation.ts"]
+
+
+def sha256(p: Path) -> str:
+    return hashlib.sha256(p.read_bytes()).hexdigest()
+
+
+def main() -> int:
+    rc = 0
+    for name in TOOLS:
+        canon = REPO / "config" / "mimo-tools" / name
+        target = Path.home() / ".mimocode" / "tools" / name
+        if not canon.is_file():
+            print(f"{name}: canonical source missing: {canon}")
+            rc = 1
+            continue
+        canon_hash = sha256(canon)
+        if target.is_file() and sha256(target) == canon_hash:
+            print(f"{name}: in-sync {canon_hash[:12]}")
+            continue
+        drift = f"DRIFT canonical={canon_hash[:12]}"
+        if target.is_file():
+            drift += f" installed={sha256(target)[:12]}"
+        else:
+            drift += " installed=MISSING"
+        print(f"{name}: {drift} — installing")
+        target.parent.mkdir(parents=True, exist_ok=True)
+        shutil.copy2(canon, target)
+        if sha256(target) != canon_hash:
+            print(f"{name}: INSTALL FAILED (hash mismatch after copy)")
+            rc = 1
+        else:
+            print(f"{name}: installed {canon_hash[:12]}")
+    return rc
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,83 @@
+#!/usr/bin/env python3
+"""install_wr_daemon.py — REAL_ASYNC_BRIDGE §6: daemon governance.
+
+Canonical source lives IN THIS REPO at bridge/daemon/. The active install
+is ~/.local/share/mimo-wait-resume/. This script is the ONLY supported
+way to move code between them:
+
+  - copy canonical source → install dir (atomic, per file)
+  - verify sha256 convergence (fail loud on divergence)
+  - optionally restart the daemon via scripts/start_wr_daemon.py
+
+Never hand-edit the installed copy: edit the repo source, install, restart.
+"""
+
+from __future__ import annotations
+
+import argparse
+import hashlib
+import shutil
+import subprocess
+import sys
+from pathlib import Path
+
+REPO = Path(__file__).resolve().parent.parent
+CANONICAL = REPO / "bridge" / "daemon"
+INSTALL = Path.home() / ".local/share/mimo-wait-resume"
+FILES = ("daemon.py", "waitstore.py", "verify.py", "completion_file.py")
+
+
+def sha256(p: Path) -> str:
+    return hashlib.sha256(p.read_bytes()).hexdigest()
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser(description=__doc__)
+    ap.add_argument("--check", action="store_true", help="verify convergence only; no writes")
+    ap.add_argument("--restart", action="store_true", help="restart the daemon after install")
+    args = ap.parse_args()
+
+    if not CANONICAL.is_dir():
+        print(f"canonical source missing: {CANONICAL}")
+        return 1
+    INSTALL.mkdir(parents=True, exist_ok=True)
+
+    diverged = []
+    for name in FILES:
+        src, dst = CANONICAL / name, INSTALL / name
+        if args.check:
+            if not dst.is_file() or sha256(src) != sha256(dst):
+                diverged.append(name)
+            continue
+        if not dst.is_file() or sha256(src) != sha256(dst):
+            tmp = dst.with_suffix(f".tmp.{name}")
+            shutil.copy2(src, tmp)
+            tmp.replace(dst)  # atomic
+            print(f"installed: {name} {sha256(src)[:12]}")
+        else:
+            print(f"unchanged: {name} {sha256(src)[:12]}")
+
+    if args.check:
+        if diverged:
+            print(f"DIVERGED: {', '.join(diverged)} — run install_wr_daemon.py")
+            return 1
+        print("CONVERGED: installed copy == canonical repo source")
+        return 0
+
+    # post-install convergence proof
+    bad = [n for n in FILES if sha256(CANONICAL / n) != sha256(INSTALL / n)]
+    if bad:
+        print(f"INSTALL FAILED convergence: {bad}")
+        return 1
+    print("CONVERGED: installed copy == canonical repo source")
+
+    if args.restart:
+        rc = subprocess.call([sys.executable, str(REPO / "scripts" / "start_wr_daemon.py")])
+        if rc != 0:
+            print("restart failed")
+            return 1
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,102 @@
+#!/usr/bin/env python3
+"""notify_event_bridge.py — T16 §16C: ci.yml's final job posts the CI
+completion WAKE to the local W/R event bridge (daemon /process-event).
+
+The event is a WAKE, never completion truth: the waiting controller
+(ci_event_wait.py) re-reads the canonical Gitea verdict via
+deploy_qualify.py after resuming. Best-effort by design — if the bridge
+is unreachable the waiter's watchdog still recovers truthfully (that is
+the §16D lost-event contract), so a notify failure must NEVER fail CI.
+
+REAL_ASYNC_BRIDGE (R3): the completion event is ALSO written directly as
+a durable completion file (resumes/<operation_id>.done). Delivery no
+longer depends on the daemon being alive: the waiter blocks on inotify
+for the file itself (kernel wake, zero latency, zero CPU), and the
+durable file survives daemon AND waiter restarts (R9). The write is
+best-effort on the same terms as the POST — never fails CI.
+"""
+from __future__ import annotations
+
+import json
+import os
+import sys
+import time
+import urllib.request
+
+BRIDGE_URL = os.environ.get("RUNE_EVENT_BRIDGE_URL",
+                            "http://127.0.0.1:8791")
+
+
+def main() -> int:
+    sha = os.environ.get("CI_SHA", "")
+    result = os.environ.get("CI_RESULT", "")  # needs.<job>.result
+    if not sha:
+        print("notify-event-bridge: no CI_SHA — nothing to wake", flush=True)
+        return 0
+    payload = {
+        "operation_id": f"ci-{sha}",
+        "provider": "gitea",
+        "event_type": "ci_completed",
+        "sha": sha,
+        # exit_code is informational only; the wake is not the verdict
+        "exit_code": 0 if result == "success" else 1,
+        "completed_at": time.time(),
+    }
+    # R8 lost-event failure-recovery: operation-scoped fault injection.
+    # A rule keyed by the EXACT operation_id suppresses BOTH normal
+    # delivery channels (durable completion file AND bridge POST), so the
+    # waiter's watchdog/failure-detector path is the only route to truth.
+    # Disabled by default (no rule file => zero behavior change) and
+    # observable: every decision is recorded in the R8 fault log.
+    try:
+        import r8_fault  # noqa: E402 — sibling module
+        if r8_fault.should_suppress(payload["operation_id"]):
+            r8_fault.record(payload["operation_id"], "suppressed",
+                            {"sha": sha, "exit_code": payload["exit_code"],
+                             "source_terminal_at": payload["completed_at"]})
+            print("notify-event-bridge: R8 fault injection — completion "
+                  f"event intentionally suppressed for "
+                  f"{payload['operation_id']} (no .done file, no POST); "
+                  "watchdog authority recovery is the expected path",
+                  flush=True)
+            return 0
+        r8_fault.record(payload["operation_id"], "delivered_normally",
+                        {"sha": sha, "rule_active": False})
+    except Exception as exc:  # noqa: BLE001 — fault seam must never break delivery
+        print(f"notify-event-bridge: R8 fault-seam probe failed ({exc}); "
+              "continuing with normal delivery", flush=True)
+    # R3 durable delivery: write the completion file FIRST — it works even
+    # if the bridge daemon is down, and the atomic rename means any waiter
+    # (inotify-watching or starting later) sees a complete file or none.
+    try:
+        sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+        import completion_file as cf  # noqa: E402 — sibling module
+        written = cf.write_completion(payload["operation_id"], payload)
+        print(f"notify-event-bridge: durable completion file {written}",
+              flush=True)
+    except Exception as exc:  # noqa: BLE001 — delivery is best-effort; watchdog recovery covers a lost completion file (§16D contract)
+        print(f"notify-event-bridge: completion-file write failed ({exc}); "
+              f"bridge POST and watchdog recovery still cover delivery",
+              flush=True)
+    for attempt in range(2):
+        try:
+            req = urllib.request.Request(
+                BRIDGE_URL + "/process-event",
+                data=json.dumps(payload).encode(),
+                headers={"Content-Type": "application/json"})
+            with urllib.request.urlopen(req, timeout=5) as r:
+                resp = json.loads(r.read())
+            print(f"notify-event-bridge: woke bridge "
+                  f"(matched={resp.get('matched')})", flush=True)
+            return 0
+        except Exception as exc:  # noqa: BLE001 — wake is best-effort by design; failure is logged and the waiter's watchdog recovery covers a lost wake (§16D contract)
+            if attempt == 0:
+                time.sleep(2)
+            else:
+                print(f"notify-event-bridge: bridge unreachable ({exc}); "
+                      f"watchdog recovery covers a lost wake", flush=True)
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,83 @@
+#!/usr/bin/env python3
+"""r8_fault.py — REAL_ASYNC_BRIDGE R8: operation-scoped lost-event fault
+injection (evidence-only harness).
+
+DESIGN CONTRACT (R8 §3):
+  disabled by default          no rule file -> zero behavior change
+  operation-scoped             rule keyed by EXACT operation_id
+  reversible                   delete the rule file / rule entry
+  observable                   every decision appended to a JSONL log
+  cannot affect other waits    rule file checked by operation_id only
+
+The ONLY consumer is scripts/notify_event_bridge.py: before normal
+delivery it consults should_suppress(operation_id). Suppression means:
+  - NO durable completion file is written
+  - NO bridge POST (no /process-event wake)
+  - the suppression decision is recorded as R8 fault evidence
+Nothing else in the stack reads the rule. The daemon, waitstore,
+authority verification, and completion-file machinery are untouched.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import sys
+import time
+from pathlib import Path
+
+RUNE_WR_ROOT = os.environ.get("RUNE_WR_ROOT") or os.environ.get("WR_ROOT") or "/mnt/workdrive/wr-bridge"
+RULES_PATH = Path(RUNE_WR_ROOT) / "r8_suppress_rules.json"
+FAULT_LOG = Path(RUNE_WR_ROOT) / "r8_fault_log.jsonl"
+
+REASON = "event intentionally suppressed for R8 lost-event failure-recovery"
+
+
+def _rules() -> dict:
+    """Read the rule file. Absent/unreadable => no suppression (fail-open
+    to NORMAL delivery; the fault must never take the healthy path down)."""
+    try:
+        return json.loads(RULES_PATH.read_text())
+    except Exception:  # noqa: BLE001 — absent rule file is the default state
+        return {}
+
+
+def lookup(operation_id: str) -> dict | None:
+    """Return the active suppression rule for EXACTLY this operation_id,
+    or None. Expired rules are ignored (not deleted — deletion is the
+    operator's explicit disable action)."""
+    spec = _rules()
+    if not spec.get("enabled"):
+        return None
+    rule = spec.get("operations", {}).get(operation_id)
+    if not rule:
+        return None
+    expires = rule.get("expires_at")
+    if expires:
+        try:
+            if time.time() > time.mktime(time.strptime(expires, "%Y-%m-%dT%H:%M:%SZ")) - time.timezone:
+                return None
+        except ValueError:
+            return None
+    return rule
+
+
+def should_suppress(operation_id: str) -> bool:
+    return lookup(operation_id) is not None
+
+
+def record(operation_id: str, stage: str, extra: dict | None = None) -> None:
+    """Append a durable fault-evidence line (best-effort, never raises)."""
+    try:
+        entry = {
+            "at": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
+            "operation_id": operation_id,
+            "reason": REASON,
+            "stage": stage,
+            **(extra or {}),
+        }
+        FAULT_LOG.parent.mkdir(parents=True, exist_ok=True)
+        with FAULT_LOG.open("a") as f:
+            f.write(json.dumps(entry, sort_keys=True) + "\n")
+    except OSError as exc:  # noqa: BLE001 — evidence logging must never break CI
+        print(f"r8_fault: evidence log write failed: {exc}", file=sys.stderr)
@@ -0,0 +1,155 @@
+#!/usr/bin/env python3
+"""release_history_cli.py — thin workflow entry points for release history.
+
+SCAFFOLD-TRANCHE-009B §4: the workflows invoke THIS tested module; no
+note-building logic lives in YAML. Two moments:
+
+    init   RELEASE INITIALIZATION — auto-version creates the evidence-
+           derived Gitea release when the version/tag appears.
+    refresh  RELEASE CONVERGENCE REFRESH — deploy promotes a later SHA of
+           the same version; the SAME release is updated with the final
+           deployed identity + newly-governed facts. Idempotent.
+
+Fail-closed boundary (§5/§16): `init` exits non-zero on publication
+failure — Rune cannot silently create another empty historical release.
+`refresh` ALSO exits non-zero (visible failure + retry converges), but the
+workflow marks it non-blocking for deployment validity: valid software is
+never rolled back over documentation. Failure state is visible either way.
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import sys
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent / "src"))
+from rune import release_notes as rn  # noqa: E402
+
+
+def _token() -> str:
+    tok = os.environ.get("GITEA_TOKEN", "")
+    if tok:
+        return tok
+    # local/dev fallback: git credential store (runner uses GITEA_TOKEN)
+    cred = Path.home() / ".git-credentials"
+    if cred.exists():
+        import re
+        m = re.search(r"https://[^:]+:([^@]+)@", cred.read_text())
+        if m:
+            return m.group(1)
+    return ""
+
+
+def main(argv=None) -> int:
+    ap = argparse.ArgumentParser(description=__doc__)
+    sub = ap.add_subparsers(dest="cmd", required=True)
+
+    p_init = sub.add_parser("init", help="release initialization")
+    p_init.add_argument("--version", required=True)
+    p_init.add_argument("--server", required=True)
+    p_init.add_argument("--repo", required=True)
+    p_init.add_argument("--store", default="store/scaffold/current.json")
+    p_init.add_argument("--release-sha", default="")
+    p_init.add_argument("--deployed-sha", default="")
+
+    p_ref = sub.add_parser("refresh", help="convergence refresh")
+    p_ref.add_argument("--version", required=True)
+    p_ref.add_argument("--server", required=True)
+    p_ref.add_argument("--repo", required=True)
+    p_ref.add_argument("--store", default="store/scaffold/current.json")
+    p_ref.add_argument("--release-sha", default="")
+    p_ref.add_argument("--deployed-sha", required=True,
+                       help="converged deployed commit (from the deployment "
+                            "authority — required, never assumed)")
+
+    args = ap.parse_args(argv)
+    token = _token()
+    if not token:
+        print(json.dumps({"failed": True,
+                          "error": "no Gitea token available"}))
+        return 1
+
+    if args.cmd == "init":
+        note = rn.auto_initial_note(
+            version=args.version,
+            summary=f"Rune v{args.version} — evidence-derived release "
+                    f"history (automatic initialization)",
+            scaffold_store=args.store,
+            release_sha=args.release_sha,
+            deployed_sha=args.deployed_sha or args.release_sha,
+        )
+        result = rn.publish_release(args.server, args.repo, token, note)
+        print(json.dumps(result))
+        # fail-closed: initialization failure must be visible so Rune cannot
+        # silently create another empty historical release (§5/§16)
+        return 0 if not result.get("failed") else 1
+
+    if args.cmd == "refresh":
+        # §12C RACE-SAFE REVALIDATION: between promotion and refresh, a
+        # NEWER deploy may have superseded this candidate. The deployment
+        # authority is re-read NOW: if current's build identity differs
+        # from the candidate, this job is stale → refuse (preserve T11
+        # invariant: stale jobs never overwrite release history). When
+        # current AGREES, the candidate IS the converged truth.
+        current_link = os.environ.get(
+            "RUNE_DEPLOY_CURRENT", "/mnt/workdrive/rune-code/current")
+        try:
+            deployed_now = rn.converged_deployed_sha(
+                str(Path(current_link).parent))
+        except (OSError, ValueError):
+            deployed_now = ""
+        if args.deployed_sha and deployed_now and \
+                not deployed_now.startswith(args.deployed_sha[:12]):
+            print(json.dumps({
+                "failed": True, "action": "refresh-superseded",
+                "error": f"deployment authority shows {deployed_now[:12]}, "
+                         f"candidate {args.deployed_sha[:12]} was superseded "
+                         f"— stale refresh refused (history preserved)",
+                "authority": deployed_now[:12],
+            }))
+            return 1  # visible; retry by the WINNING deploy converges
+
+        existing = rn.fetch_release(args.server, args.repo, token,
+                                    f"v{args.version}")
+        if existing is None:
+            # no release yet for this version — init instead (convergence
+            # may legitimately arrive before/without an init if the
+            # auto-version run failed); idempotent recovery path
+            note = rn.build_from_deployment(
+                version=args.version, deployed_sha=args.deployed_sha,
+                scaffold_store=args.store, release_sha=args.release_sha)
+        else:
+            note = rn.convergence_refresh(
+                rn.ReleaseNote.from_dict({
+                    "version": args.version,
+                    "summary": _summary_from_body(existing.get("body", "")),
+                    "provenance": [],
+                }),
+                deployed_sha=args.deployed_sha, scaffold_store=args.store,
+                extra_provenance="automatic convergence refresh (deploy "
+                                 "workflow)")
+            # preserve structural fields the stored body carries
+            note.release_sha = _sha_from_body(existing.get("body", ""),
+                                              "tag") or args.release_sha
+        result = rn.publish_release(args.server, args.repo, token, note)
+        print(json.dumps(result))
+        # visible failure; retry converges (§10). Deployment validity is
+        # NOT rolled back over documentation — the workflow marks this
+        # step appropriately.
+        return 0 if not result.get("failed") else 1
+    return 2
+
+
+def _summary_from_body(body: str) -> str:
+    """Delegate to the single authority in rune.release_notes (§14B)."""
+    return rn.summary_from_body(body)
+
+
+def _sha_from_body(body: str, which: str) -> str:
+    return rn.sha_from_body(body, which)
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -181,7 +181,8 @@ def run_stage_with_relay(tree, out_dir: str) -> dict:
    err_path = os.path.join(out_dir, f"stage_{stage}.mimo.err.log")

    mimo_bin = os.environ.get("MIMO_BIN", "/var/lib/gitea-desktop-runner/bin/mimo")
-    mimo_model = os.environ.get("MIMO_MODEL", "Verboo Code/glm-5.3-flash")
+    from rune.provider_context import require_model
+    mimo_model = require_model()  # T16 §16F: provider layer owns selection
    mimo_agent = os.environ.get("MIMO_AGENT", "build")

    cmd = [
@@ -38,11 +38,17 @@ def build_context(issue: dict, define_contract: dict) -> dict:
    # 3. Find matching test files
    matched_tests = _find_tests(keywords)

+    # 4. T21.5 §10: Memory internal-experience port — ONE port of call,
+    #    bounded, zero-result = normal (fast path: [] → next step directly).
+    from research_memory_port import memory_port_of_call
+    memory_candidates = memory_port_of_call(issue, keywords)
+
    return {
        "keywords": keywords,
        "umbrella_matches": matched_modules,
        "candidate_files": matched_files[:15],  # bounded
        "candidate_tests": matched_tests[:10],  # bounded
+        "memory_candidates": memory_candidates,  # evidence, never authority
        "define_contract": define_contract,
        "mimo_needed": _needs_mimo(matched_files, matched_tests),
    }
@@ -0,0 +1,139 @@
+"""research_memory_port.py — T21.5 §10-12: the future-facing Memory port
+inside the EXISTING INWARD research stage.
+
+BOUNDARIES (G9/G10/G11):
+  - This is a PORT, not a memory system. No storage is created here; the
+    implementation delegates to a provider callable that future Memory
+    will supply. Tests inject doubles; INWARD never imports Memory.
+  - Zero result is a NORMAL SUCCESSFUL result: the caller continues the
+    normal INWARD flow immediately (no extra model call, no ceremony).
+  - Returned material is EVIDENCE/CANDIDATES only — never an instruction
+    authority. APPLICABILITY (existing stage) governs whether a candidate
+    applies; SYNTHESIS receives only distilled, applicable consequences.
+  - Teachable-moment seam: durable construction evidence can mark an
+    episode as a CANDIDATE for future learning ("may be worth learning",
+    never "always apply"). Substrate/Memory (future) harvest candidates;
+    today the seam only records them via the same provider contract.
+"""
+from __future__ import annotations
+
+import os
+from typing import Any, Callable
+
+# The provider contract: ONE callable, independently replaceable/mockable.
+# Signature: query(query_text, keywords, *, limit) -> list[MemoryCandidate]
+# MemoryCandidate (dict): {source, kind, summary, relevance, episode_ref}
+#   kind ∈ {construction_episode, teachable_moment_candidate,
+#           prior_failure, prior_success}
+# Future Memory supplies the real implementation; nothing else changes.
+
+MemoryProvider = Callable[..., list[dict]]
+
+_DEFAULT_PROVIDER: MemoryProvider | None = None
+
+
+def set_memory_provider(provider: MemoryProvider | None) -> None:
+    """Bind the Memory-side implementation (production or test double).
+    None disables the port — INWARD continues without it (fast path)."""
+    global _DEFAULT_PROVIDER
+    _DEFAULT_PROVIDER = provider
+
+
+def _provider() -> MemoryProvider | None:
+    global _DEFAULT_PROVIDER
+    if _DEFAULT_PROVIDER is not None:
+        return _DEFAULT_PROVIDER
+    env = os.environ.get("RUNE_MEMORY_PORT_PROVIDER")
+    if env:
+        import importlib.util
+        spec = importlib.util.spec_from_file_location("mprov", env)
+        if spec is not None and spec.loader is not None:
+            mod = importlib.util.module_from_spec(spec)
+            spec.loader.exec_module(mod)
+            provider = getattr(mod, "query", None)
+            if provider is not None:
+                _DEFAULT_PROVIDER = provider
+    return _DEFAULT_PROVIDER
+
+
+MAX_CANDIDATES = 5
+MIN_RELEVANCE = 0.3
+
+
+def memory_port_of_call(issue: dict, keywords: list[str],
+                        *, provider: MemoryProvider | None = None) -> list[dict]:
+    """The ONE internal port of call inside INWARD. Bounded retrieval of
+    internal evidence (construction episodes, teachable-moment
+    candidates, prior failures/successes).
+
+    ZERO-RESULT FAST PATH (§11): empty/thin results return [] and the
+    caller continues the next normal INWARD step directly. No extra
+    model call, no fallback reasoning, no synthetic summary, no delay.
+    """
+    prov = provider or _provider()
+    if prov is None:
+        return []  # port unbound: normal INWARD continues, zero overhead
+    try:
+        query_text = " ".join([
+            str(issue.get("title", "")),
+            str(issue.get("body", ""))[:400],
+        ]).strip()
+        raw = prov(query_text, keywords, limit=MAX_CANDIDATES) or []
+    except Exception:  # noqa: BLE001 — port failure must never break INWARD
+        return []
+    candidates = [c for c in raw
+                  if float(c.get("relevance", 0.0)) >= MIN_RELEVANCE]
+    return candidates[:MAX_CANDIDATES]
+
+
+def applicability_filter(candidates: list[dict],
+                         context_keywords: list[str]) -> tuple[list[dict],
+                                                              list[dict]]:
+    """Existing APPLICABILITY governs transfer: candidates relevant to
+    the CURRENT context pass; irrelevant ones are rejected WITHOUT
+    contaminating downstream SYNTHESIS. Returns (applicable, rejected)."""
+    applicable, rejected = [], []
+    for c in candidates:
+        text = (str(c.get("summary", "")) + " " +
+                " ".join(str(k) for k in c.get("keywords", []))).lower()
+        overlap = sum(1 for k in context_keywords if k and k.lower() in text)
+        if overlap > 0 or c.get("explicitly_applicable") is True:
+            applicable.append(c)
+        else:
+            rejected.append(c)
+    return applicable, rejected
+
+
+# ── §12: teachable-moment future seam ──────────────────────────────────────
+
+CANDIDATE_TRIGGERS = (
+    "completed_release", "capability_completed", "failure_recovered",
+    "rollback_executed", "user_correction", "prediction_observation_mismatch",
+    "procedure_improved",
+)
+
+
+def mark_teachable_moment_candidate(
+        episode_ref: str, *, trigger: str, summary: str,
+        emit: Callable[[dict], Any] | None = None) -> dict | None:
+    """Record durable evidence that an episode MAY be worth future
+    learning — a candidate, never a rule. Uses the existing evidence/
+    event mechanism via `emit` (the caller's channel: journal, event
+    bridge, ledger history). No new canonical store is created here;
+    future Substrate/Memory harvests candidates through the port."""
+    if trigger not in CANDIDATE_TRIGGERS:
+        raise ValueError(
+            f"unknown trigger {trigger!r} — must be one of "
+            f"{CANDIDATE_TRIGGERS}")
+    record = {
+        "candidate": "teachable_moment",
+        "episode_ref": episode_ref,
+        "trigger": trigger,
+        "summary": summary[:200],
+        "status": "CANDIDATE",   # never authority, never auto-applied
+        "note": "may be worth future learning; APPLICABILITY governs "
+                "any future transfer",
+    }
+    if emit is not None:
+        emit(record)
+    return record
@@ -793,7 +793,8 @@ class StageExecutor:
        prompt = chr(10).join(pp)

        mimo_bin = os.environ.get("MIMO_BIN", "/var/lib/gitea-desktop-runner/bin/mimo")
-        mimo_model = os.environ.get("MIMO_MODEL", "Verboo Code/glm-5.3-flash")
+        from rune.provider_context import require_model
+        mimo_model = require_model()  # T16 §16F: provider layer owns selection
        mimo_agent = os.environ.get("MIMO_AGENT", "build")
        events_path = os.path.join(self.artifact_dir, "stage_OUTWARD.events.jsonl")
        err_path = os.path.join(self.artifact_dir, "stage_OUTWARD.mimo.err.log")
@@ -933,7 +934,8 @@ class StageExecutor:
        prompt = chr(10).join(pp)

        mimo_bin = os.environ.get("MIMO_BIN", "/var/lib/gitea-desktop-runner/bin/mimo")
-        mimo_model = os.environ.get("MIMO_MODEL", "Verboo Code/glm-5.3-flash")
+        from rune.provider_context import require_model
+        mimo_model = require_model()  # T16 §16F: provider layer owns selection
        mimo_agent = os.environ.get("MIMO_AGENT", "build")
        events_path = os.path.join(self.artifact_dir, "stage_APPLICABILITY.events.jsonl")
        err_path = os.path.join(self.artifact_dir, "stage_APPLICABILITY.mimo.err.log")
@@ -1137,7 +1139,8 @@ class StageExecutor:
        prompt = chr(10).join(pp)

        mimo_bin = os.environ.get("MIMO_BIN", "/var/lib/gitea-desktop-runner/bin/mimo")
-        mimo_model = os.environ.get("MIMO_MODEL", "Verboo Code/glm-5.3-flash")
+        from rune.provider_context import require_model
+        mimo_model = require_model()  # T16 §16F: provider layer owns selection
        mimo_agent = os.environ.get("MIMO_AGENT", "build")
        events_path = os.path.join(self.artifact_dir, "stage_ADVERSARIAL.events.jsonl")
        err_path = os.path.join(self.artifact_dir, "stage_ADVERSARIAL.mimo.err.log")
@@ -1345,7 +1348,8 @@ class StageExecutor:
        prompt = chr(10).join(pp).replace("{n}", str(self.issue))

        mimo_bin = os.environ.get("MIMO_BIN", "/var/lib/gitea-desktop-runner/bin/mimo")
-        mimo_model = os.environ.get("MIMO_MODEL", "Verboo Code/glm-5.3-flash")
+        from rune.provider_context import require_model
+        mimo_model = require_model()  # T16 §16F: provider layer owns selection
        mimo_agent = os.environ.get("MIMO_AGENT", "build")
        events_path = os.path.join(self.artifact_dir, "stage_INNOVATION.events.jsonl")
        err_path = os.path.join(self.artifact_dir, "stage_INNOVATION.mimo.err.log")
@@ -1534,7 +1538,8 @@ class StageExecutor:
        prompt = chr(10).join(pp)

        mimo_bin = os.environ.get("MIMO_BIN", "/var/lib/gitea-desktop-runner/bin/mimo")
-        mimo_model = os.environ.get("MIMO_MODEL", "Verboo Code/glm-5.3-flash")
+        from rune.provider_context import require_model
+        mimo_model = require_model()  # T16 §16F: provider layer owns selection
        mimo_agent = os.environ.get("MIMO_AGENT", "build")
        events_path = os.path.join(self.artifact_dir, "stage_SYNTHESIS.events.jsonl")
        err_path = os.path.join(self.artifact_dir, "stage_SYNTHESIS.mimo.err.log")
@@ -1671,7 +1676,8 @@ class StageExecutor:
            prompt = chr(10).join(pp).replace("{n}", str(self.issue))

            mimo_bin = os.environ.get("MIMO_BIN", "/var/lib/gitea-desktop-runner/bin/mimo")
-            mimo_model = os.environ.get("MIMO_MODEL", "Verboo Code/glm-5.3-flash")
+            from rune.provider_context import require_model
+            mimo_model = require_model()  # T16 §16F: provider layer owns selection
            mimo_agent = os.environ.get("MIMO_AGENT", "build")
            events_path = os.path.join(self.artifact_dir, "stage_QUALIFY.events.jsonl")
            err_path = os.path.join(self.artifact_dir, "stage_QUALIFY.mimo.err.log")
@@ -0,0 +1,259 @@
+#!/usr/bin/env python3
+"""rollback_generation.py — T21.4 §21.4F: the governed rollback API.
+
+    rollback_generation(target_generation, expected_current, reason)
+
+Prior tranches proved rollback by recreating deployment primitives by hand
+(ln -sfn against the live deploy root). That is operator work, not governed
+machinery — it bypasses lock, eligibility, health, and history. This module
+is the single canonical rollback operation, built ON the proven deploy
+machinery (deploy_promotion's lock + atomic swap + build-identity authority),
+not beside it.
+
+The promotion authority (deploy_promotion.py) is deliberately MONOTONIC: an
+ancestor promotion is refused as STALE_SUPERSEDED. Regression is therefore
+impossible through promotion — and legitimate only through THIS operation,
+which owns its own obligations:
+
+ 1.  deployment lock       — the SAME .promotion.lock as promote (serialize)
+ 2.  current verification  — build-identity readback must equal expected_current
+ 3.  target existence      — generation dir present, build identity readable
+ 4.  eligibility           — target commit is ancestor-or-equal of current
+                             (regression only; forward = promotion's job;
+                             divergent/unknown refused)
+ 5.  atomic switch         — rename(2) symlink swap, never absent/half-updated
+ 6.  runtime handoff       — Rune executes from current/src at process start;
+                             handoff = identity a fresh launch would run
+ 7.  runtime verification  — fresh-launch identity must equal target
+ 8.  health qualification  — import smoke against target generation src
+ 9.  durable history       — JSONL rollback events, appended inside the lock
+ 10. failure recovery      — health failure restores prior current (where safe)
+ 11. coherent last-good    — last-good := the generation that WAS current
+                             (same invariant promotion maintains)
+
+Truthful outcomes:
+    ROLLED_BACK              target active, health green, history recorded
+    NOOP_ALREADY_CURRENT     target already the deployed generation
+    ROLLED_BACK_REVERTED     switched, health red, prior current restored
+    REFUSED_UNKNOWN_CURRENT  deployed identity unreadable (fail closed)
+    REFUSED_EXPECTED_MISMATCH current != expected_current (stale caller)
+    REFUSED_TARGET_MISSING   target generation dir absent
+    REFUSED_TARGET_UNREADABLE target build identity unreadable
+    REFUSED_NOT_ELIGIBLE     divergent, newer, or non-ancestor target
+    REFUSED_NO_REASON        empty reason (accountability is mandatory)
+"""
+
+from __future__ import annotations
+
+import fcntl
+import json
+import subprocess
+import sys
+import time
+from pathlib import Path
+
+_SCRIPT_DIR = Path(__file__).resolve().parent
+if str(_SCRIPT_DIR) not in sys.path:
+    sys.path.insert(0, str(_SCRIPT_DIR))
+import deploy_promotion  # proven machinery: lock path, ancestry, swaps
+
+DEPLOY_ROOT = deploy_promotion.DEPLOY_ROOT
+HARNESS_ROOT = _SCRIPT_DIR.parent
+HISTORY_PATH = HARNESS_ROOT / "store" / "scaffold" / "rollback_history.jsonl"
+
+ROLLED_BACK = "ROLLED_BACK"
+NOOP = "NOOP_ALREADY_CURRENT"
+ROLLED_BACK_REVERTED = "ROLLED_BACK_REVERTED"
+REFUSED_UNKNOWN_CURRENT = "REFUSED_UNKNOWN_CURRENT"
+REFUSED_EXPECTED_MISMATCH = "REFUSED_EXPECTED_MISMATCH"
+REFUSED_TARGET_MISSING = "REFUSED_TARGET_MISSING"
+REFUSED_TARGET_UNREADABLE = "REFUSED_TARGET_UNREADABLE"
+REFUSED_NOT_ELIGIBLE = "REFUSED_NOT_ELIGIBLE"
+REFUSED_NO_REASON = "REFUSED_NO_REASON"
+
+
+def _identity(deploy_root: Path, generation: str) -> str:
+    """Authoritative commit for a generation dir (build identity, not name)."""
+    gen_dir = deploy_root / generation
+    identity = gen_dir / "src" / "rune" / "_build_identity.json"
+    try:
+        data = json.loads(identity.read_text(encoding="utf-8"))
+        return str(data.get("commit", ""))
+    except (OSError, ValueError):
+        return ""
+
+
+def _health(deploy_root: Path, generation: str, timeout: int = 60) -> tuple[bool, str]:
+    """Health qualification for a generation: build identity readable AND
+    `import rune` succeeds from that generation's src tree."""
+    gen_dir = deploy_root / generation
+    commit = _identity(deploy_root, generation)
+    if not commit:
+        return False, "build identity unreadable"
+    src = gen_dir / "src"
+    try:
+        proc = subprocess.run(
+            [sys.executable, "-c", "import rune; import sys; "
+             "sys.stdout.write(getattr(rune, '__version__', 'ok'))"],
+            env={"PYTHONPATH": str(src), "PATH": "/usr/bin:/bin",
+                 "HOME": str(gen_dir)},
+            capture_output=True, text=True, timeout=timeout, cwd=str(gen_dir),
+        )
+    except subprocess.TimeoutExpired:
+        return False, f"import smoke timed out after {timeout}s"
+    if proc.returncode != 0:
+        return False, f"import smoke failed: {(proc.stderr or '')[-200:]}"
+    return True, f"identity={commit[:12]} import_smoke=ok ({proc.stdout.strip()[:40]})"
+
+
+def _append_history(record: dict, history_path: Path = HISTORY_PATH) -> None:
+    history_path.parent.mkdir(parents=True, exist_ok=True)
+    with open(history_path, "a", encoding="utf-8") as f:
+        f.write(json.dumps(record, sort_keys=True) + "\n")
+
+
+def _swap(deploy_root: Path, name: str, target: Path) -> None:
+    deploy_promotion._atomic_symlink_target(deploy_root, name, target)
+
+
+def rollback_generation(
+    target_generation: str,
+    expected_current: str,
+    reason: str,
+    *,
+    deploy_root: Path = DEPLOY_ROOT,
+    repo: Path = HARNESS_ROOT,
+    history_path: Path = HISTORY_PATH,
+) -> tuple[str, dict]:
+    """The governed rollback operation. Returns (outcome, detail).
+
+    Refusals happen BEFORE any mutation of current/last-good. The lock is
+    shared with promote() so rollback and promotion can never interleave.
+    """
+    if not reason or not reason.strip():
+        return REFUSED_NO_REASON, {"error": "rollback requires a reason"}
+
+    lock_path = deploy_root / ".promotion.lock"
+    # The lock file may be owned by the CI runner (promotion side opens it
+    # for writing). flock needs only an open fd, never write permission, so
+    # open read-only when write is denied — same file, same fence.
+    try:
+        lock_fd = open(lock_path, "a+")
+    except PermissionError:
+        lock_fd = open(lock_path) if lock_path.exists() else \
+            open(lock_path, "a+")
+    with lock_fd:
+        fcntl.flock(lock_fd, fcntl.LOCK_EX)
+        try:
+            current = deploy_root / "current"
+            last_good = deploy_root / "last-good"
+            detail: dict = {"target": target_generation,
+                            "expected_current": expected_current,
+                            "reason": reason[:200],
+                            "at": time.strftime("%Y-%m-%dT%H:%M:%SZ",
+                                                time.gmtime())}
+
+            # 2. current verification — identity authority, never the name
+            current_generation = ""
+            try:
+                current_generation = Path(os_readlink(current)).name
+            except OSError:
+                pass
+            deployed_commit = deploy_promotion.read_deployed_commit(deploy_root)
+            if not deployed_commit:
+                return REFUSED_UNKNOWN_CURRENT, {
+                    "error": "deployed commit unreadable — fail closed",
+                    **detail}
+            detail["current_generation"] = current_generation
+            detail["current_commit"] = deployed_commit
+            if expected_current and expected_current != deployed_commit:
+                detail["error"] = (
+                    f"current is {deployed_commit[:12]}, "
+                    f"caller expected {expected_current[:12]} — refused "
+                    "before mutation")
+                return REFUSED_EXPECTED_MISMATCH, detail
+
+            # 3. target existence + identity
+            target_dir = deploy_root / target_generation
+            if not target_dir.is_dir():
+                detail["error"] = f"generation dir {target_generation} absent"
+                return REFUSED_TARGET_MISSING, detail
+            target_commit = _identity(deploy_root, target_generation)
+            if not target_commit:
+                detail["error"] = (
+                    f"generation {target_generation} build identity unreadable")
+                return REFUSED_TARGET_UNREADABLE, detail
+            detail["target_commit"] = target_commit
+
+            if target_commit == deployed_commit:
+                detail["note"] = "target already the deployed generation"
+                return NOOP, detail
+
+            # 4. eligibility — ancestor-or-equal only (regression)
+            rel = deploy_promotion.ancestry_of(target_commit, deployed_commit,
+                                               repo)
+            if rel != "ancestor":
+                detail["error"] = (
+                    f"target commit ancestry={rel} — only ancestors are "
+                    "rollback-eligible (forward movement belongs to "
+                    "deploy_promotion)")
+                return REFUSED_NOT_ELIGIBLE, detail
+
+            # 5. atomic switch (inside lock), last-good := previous current
+            if last_good.is_symlink():
+                last_good.unlink()
+            _swap(deploy_root, "last-good", Path(os_readlink(current)))
+            _swap(deploy_root, "current", target_dir)
+            detail["switched_to"] = target_generation
+
+            # 7+8. runtime verification + health qualification
+            post_commit = deploy_promotion.read_deployed_commit(deploy_root)
+            health_ok, health_detail = _health(deploy_root, target_generation)
+            detail["post_switch_identity"] = post_commit
+            detail["health"] = health_detail
+
+            record = {"at": detail["at"], "operation": "rollback_generation",
+                      **detail}
+            if post_commit != target_commit or not health_ok:
+                # 10. failure recovery — restore prior current (safe: we
+                # hold the lock and the prior generation is intact)
+                prev = Path(os_readlink(last_good)) if last_good.is_symlink() \
+                    else None
+                if prev is not None:
+                    _swap(deploy_root, "current", prev)
+                detail["reverted_to"] = (Path(str(os_readlink(current))).name
+                                         if current.is_symlink() else "?")
+                detail["outcome"] = ROLLED_BACK_REVERTED
+                record["outcome"] = ROLLED_BACK_REVERTED
+                _append_history(record, history_path)
+                return ROLLED_BACK_REVERTED, detail
+
+            record["outcome"] = ROLLED_BACK
+            _append_history(record, history_path)
+            detail["outcome"] = ROLLED_BACK
+            return ROLLED_BACK, detail
+        finally:
+            fcntl.flock(lock_fd, fcntl.LOCK_UN)
+
+
+def os_readlink(p: Path) -> str:
+    return str(Path(str(p)).readlink()) if hasattr(Path, "readlink") \
+        else str(Path(str(p)).readlink())
+
+
+def main() -> int:
+    import argparse
+    ap = argparse.ArgumentParser(description=__doc__)
+    ap.add_argument("target_generation")
+    ap.add_argument("--expected-current", required=True,
+                    help="commit the caller believes is currently deployed")
+    ap.add_argument("--reason", required=True)
+    args = ap.parse_args()
+    outcome, detail = rollback_generation(args.target_generation,
+                                          args.expected_current, args.reason)
+    print(json.dumps({"outcome": outcome, **detail}, indent=2, sort_keys=True))
+    return 0 if outcome in (ROLLED_BACK, NOOP) else 1
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -18,6 +18,13 @@ from __future__ import annotations
 import re
 import sys

+# T20 §20B: push-convergence synthesizes PR titles as "<kind>: converged
+# release (push-to-master bump)" where kind comes from
+# version_convergence.highest_warranted_bump. The title text must carry
+# that decision — "minor:" was re-classified as patch here, downgrading
+# every feature tranche (observed: feat ranges bumped 0.66.77-83 not 0.67.0).
+MINOR_PREFIX_RE = re.compile(r"^\s*minor\b", re.IGNORECASE)
+
 BREAKING_RE = re.compile(r"^\s*\w+(\([^)]*\))?!:")
 FEAT_RE = re.compile(r"^\s*feat(\([^)]*\))?")

@@ -25,6 +32,8 @@ FEAT_RE = re.compile(r"^\s*feat(\([^)]*\))?")
 def next_version(version: str, title: str) -> str:
    if BREAKING_RE.match(title) or "BREAKING CHANGE" in title:
        kind = "major"
+    elif MINOR_PREFIX_RE.match(title):
+        kind = "minor"  # synthesized convergence title carries the kind
    elif FEAT_RE.match(title):
        kind = "minor"
    else:
@@ -0,0 +1,64 @@
+#!/usr/bin/env python3
+"""start_wr_daemon.py — T21.5 §7: THE canonical daemon restart launcher.
+
+T21.4 live finding: `python3 daemon.py` defaults WR_ROOT to the daemon's
+install dir, so a restart silently opened a second event store. This
+launcher resolves the root ONCE through scripts/wr_root_resolver.py (the
+single authority) and passes it explicitly — nobody has to remember to
+export WR_ROOT.
+"""
+from __future__ import annotations
+
+import os
+import subprocess
+import sys
+import time
+import urllib.request
+from pathlib import Path
+
+_SCRIPT_DIR = Path(__file__).resolve().parent
+sys.path.insert(0, str(_SCRIPT_DIR))
+import wr_root_resolver as resolver  # noqa: E402
+
+DAEMON_CODE_CANDIDATES = (
+    "/home/nick/.local/share/mimo-wait-resume/daemon.py",
+)
+
+
+def find_daemon_code() -> Path:
+    for cand in DAEMON_CODE_CANDIDATES:
+        p = Path(cand)
+        if p.is_file():
+            return p
+    raise RuntimeError(f"daemon.py not found in {DAEMON_CODE_CANDIDATES}")
+
+
+def main() -> int:
+    root = resolver.resolve_root()
+    daemon = find_daemon_code()
+    env = dict(os.environ,
+               WR_ROOT=str(root),            # daemon data root
+               RUNE_WR_ROOT=str(root),       # adapter-side authority
+               RUNE_WR_CODE=str(daemon.parent))
+    port = sys.argv[1] if len(sys.argv) > 1 else "8791"
+    proc = subprocess.Popen(
+        [sys.executable, str(daemon), port],
+        env=env, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL,
+        start_new_session=True)
+    # bounded startup probe (process-local, not an external wait)
+    for _ in range(20):
+        try:
+            with urllib.request.urlopen(
+                    f"http://127.0.0.1:{port}/health", timeout=1) as r:
+                if r.status == 200:
+                    print(f"wr daemon up: pid={proc.pid} root={root} "
+                          f"port={port}")
+                    return 0
+        except OSError:
+            time.sleep(0.25)
+    print(f"wr daemon did not become healthy (pid={proc.pid})", file=sys.stderr)
+    return 1
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,331 @@
+#!/usr/bin/env python3
+"""structural_coverage_audit.py — T21.4 §21.4G: Rune-wide meaningful
+structural coverage audit.
+
+Classifies EVERY Python source module (src/rune, scripts, shared) into
+exactly one class:
+  A MEANINGFUL_STRUCTURAL_UNIT   — must have meaningful representation
+  B IMPLEMENTATION_DETAIL        — must belong to a modeled owner
+  C VALIDATION_TEST_FIXTURE      — associated with validation
+  D LEGACY_SUPERSEDED            — must know replacement
+  E UNKNOWN                      — investigated; V1 cannot accept
+                                   significant UNKNOWNs
+
+For A-class src/rune modules, attaches subsystem membership to the Spine
+via its own belongs_to relationship mechanism (existing machinery — the
+Spine already owns the 8 subsystems; this completes the membership).
+
+Writes store/scaffold/structural_coverage_t214.json + prints measures.
+"""
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(ROOT / "src"))
+
+# ── A-class: src/rune meaningful units → subsystem ─────────────────────────
+SRC_A = {
+    # governance / construction / structural
+    "scaffold": "governance", "structural_store": "governance",
+    "structural_model": "governance", "structural_migration": "governance",
+    "spine": "governance", "construction_activation": "governance",
+    "construction_commands": "governance", "construction_decision": "governance",
+    "construction_depth": "governance", "construction_intent": "governance",
+    "construction_interface": "governance", "construction_mode": "governance",
+    "construction_pattern": "governance",
+    "construction_patterns_builtin": "governance",
+    "construction_projection": "governance", "stage_guards": "governance",
+    "capability_fit": "governance", "toolchain": "governance",
+    "layout_contract": "governance", "project_registry": "governance",
+    "programme_gates": "governance", "programme_continuation": "governance",
+    "build_basis": "governance", "iteration": "governance",
+    "event_wait": "governance", "wait_guard": "governance",
+    "failure_recovery": "governance", "failure_auto_filer": "governance",
+    "forge": "governance", "forge_client": "governance",
+    "release_notes": "governance",
+    # memory (existing machinery — not Memory V1)
+    "memory_store": "memory", "memory_selector": "memory",
+    "memory_consolidation": "memory", "memory_env": "memory",
+    "shard_manager": "memory", "tier_manager": "memory",
+    "graph_memory": "memory", "query_memory": "memory",
+    "curator": "memory", "sleep_time": "memory",
+    "subconscious_pass": "memory", "warm_neuron_store": "memory",
+    "warm_router": "memory", "abstraction": "memory",
+    "auto_context": "memory", "cache": "memory",
+    "connection_pool": "memory",
+    # growth / governed adaptation / ledgers
+    "drift_core": "growth", "drift_ledger": "growth",
+    "drift_cache": "growth", "drift_compression": "growth",
+    "drift_observability": "growth", "drift_resume": "growth",
+    "growth_delta": "growth", "self_improve": "growth",
+    "night_mode": "growth", "behavior": "growth",
+    "outcome_ledger": "growth", "claim_audit": "growth",
+    # identity / persona / voice
+    "identity_anchors": "identity", "persona_definition": "identity",
+    "persona_gate": "identity", "voice_hybrid": "identity",
+    "voice_sync": "identity", "safety_override": "identity",
+    "prompt_sync": "identity",
+    # inference / routing / providers
+    "adaptive_router": "inference", "route_risk_flags": "inference",
+    "model_routing": "inference", "model_connect": "inference",
+    "model_profiles": "inference", "model_registry": "inference",
+    "provider_registry": "inference", "provider_context": "inference",
+    "provider_budget": "inference", "standardized_providers": "inference",
+    "openai_compat_transport": "inference",
+    "local_tool_runtime": "inference", "response_adapter": "inference",
+    "inference_trace": "inference", "evidence_probe": "inference",
+    "recall_router": "inference", "recall_uplift": "inference",
+    # orchestration / cognition / continuation / ledgers
+    "elastic_cognition": "orchestration", "tiered_executor": "orchestration",
+    "attention_capsule": "orchestration", "witness": "orchestration",
+    "event_bus": "orchestration", "microkernel": "orchestration",
+    "capability_broker": "orchestration", "context_broker": "orchestration",
+    "work_manager": "orchestration", "task_persist": "orchestration",
+    "goal": "orchestration", "cron_engine": "orchestration",
+    "hook_engine": "orchestration", "subsystem_health": "orchestration",
+    "verification_layer": "orchestration", "build_mode": "orchestration",
+    "builder_bridge": "orchestration", "testing_mode": "orchestration",
+    "rune_operator": "orchestration", "audit_ledger": "capability",
+    "capability_ledger": "capability", "behaviour_ledger": "capability",
+    "impact_preview": "governance", "graph_projection": "governance",
+    "tool_executor": "orchestration", "tool_protocol": "orchestration",
+    # perception / TUI boundary
+    "rune_tui": "perception", "__main__": "perception",
+    "voicebox": "perception",
+}
+
+# B-class: src/rune implementation details → owning meaningful unit
+SRC_B = {
+    "event_envelope": "event_bus",
+    "cognitive_budgeter": "elastic_cognition",
+    "salience_gate": "elastic_cognition",
+    "rune_profiler": "elastic_cognition",
+    "progressive_escalation": "tiered_executor",
+    "parallel_processor": "tiered_executor",
+    "context_pressure": "context_broker",
+    "token_optimizer": "context_broker",
+    "message_queue": "work_manager",
+    "auto_tasks": "work_manager",
+    "task_gate": "task_persist",
+    "pins": "goal",
+    "chat_bubble": "rune_tui",
+    "rune_spinner": "rune_tui",
+    "rune_graphic": "rune_tui",
+    "message_widgets": "rune_tui",
+    "tui_validator": "rune_tui",
+    "tui_backlog": "rune_tui",
+    "builder_bridge_cli": "builder_bridge",
+    "voice_injector": "voice_hybrid",
+    "intent_continuity": "programme_continuation",
+    "time_awareness": "rune_operator",
+    "contracts": "verification_layer",
+    "system_registry": "governance-ops",
+    "dependency_tracker": "ops-harness",
+}
+
+# C-class: validation/test fixtures (canaries + corpus)
+SRC_C = {
+    "canary_cognitive_strata", "canary_context_broker",
+    "canary_conversation_state", "canary_full_system",
+    "canary_gitea_automation", "canary_gitea_delegation",
+    "canary_gitea_pipeline", "canary_memory_synthesis", "canary_native",
+    "canary_orchestration_foundation", "canary_p0_04", "canary_p0_05",
+    "canary_p0_07", "canary_p1_01", "canary_p1_02", "canary_p1_03",
+    "canary_p1_06", "canary_p1_07", "canary_query_memory",
+    "canary_resource_authority", "canary_schema_migration",
+    "canary_surface_parity", "canary_tui_drift", "canary_unbounded",
+    "wait_guard_corpus", "contracts",
+}
+
+# scripts/ A-class (deployment/CI/release/research/governance machinery)
+SCRIPTS_A = {
+    "auto_bump_pr", "capture_failure", "carry_forward_check",
+    "check_dep_lock", "check_discharge_contract", "check_version_surfaces",
+    "ci_event_wait", "construction_guards", "create_release",
+    "deploy_promotion", "deploy_qualify", "deploy_stage",
+    "rollback_generation", "file_drift_issue", "gitea_auth",
+    "gitea_reconciliation_check", "install_mimo_hook", "lint_ratchet",
+    "merge_robot", "mutation_boundary", "notify_event_bridge",
+    "process_experience", "publish_source_ledger", "research_artifact",
+    "research_backfill_sweep", "research_define", "research_event_relay",
+    "research_inward_context", "research_prompt_compiler",
+    "research_router", "research_stage_executor", "research_task_tree",
+    "runner_evidence", "scaffold_projection", "semver_bump",
+    "source_ledger", "stage_contracts", "stage_telemetry",
+    "t14_builder_session", "t15_builder_session", "type_check_ratchet",
+    "umbrella_impact", "version_convergence",
+}
+SCRIPTS_A = set(SCRIPTS_A) | {"structural_coverage_audit"}
+SCRIPTS_B = {"gen_prd_state": "governance-ops",
+             "pl1_records": "process_experience",
+             "release_history_cli": "release_notes"}
+SCRIPTS_C = {"fake_worker", "tui_startup_canary"}
+
+# shared/ — harness-facing layer; mirrors of canonical src modules = D
+SHARED_D = {
+    "audit_ledger": "src/rune/audit_ledger.py",
+    "dependency_tracker": "src/rune/dependency_tracker.py",
+    "drift_cache": "src/rune/drift_cache.py",
+    "drift_compression": "src/rune/drift_compression.py",
+    "drift_core": "src/rune/drift_core.py",
+    "drift_observability": "src/rune/drift_observability.py",
+    "gitea_auth": "scripts/gitea_auth.py",
+    "query_memory": "src/rune/query_memory.py",
+    "shard_manager": "src/rune/shard_manager.py",
+    "subconscious_pass": "src/rune/subconscious_pass.py",
+    "tier_manager": "src/rune/tier_manager.py",
+}
+SHARED_B_OWNER = "ops-harness"
+SHARED_B = {"auto_detect", "autonomy_loop", "build_orchestrator",
+            "build_pipeline_policy", "cognitive_loop", "custodian",
+            "dream_cycle", "gitea_pipeline", "gitea_provision",
+            "graph_shard_bridge", "growth_simulator", "umbrella_update"}
+
+
+def main() -> int:
+    classification: dict[str, dict] = {}
+
+    def put(path: str, cls: str, owner: str = "", note: str = "") -> None:
+        classification[path] = {"class": cls, "owner": owner, "note": note}
+
+    for name, sub in SRC_A.items():
+        put(f"src/rune/{name}.py", "A", owner=f"subsystem:{sub}")
+    for name, owner in SRC_B.items():
+        put(f"src/rune/{name}.py", "B", owner=owner)
+    for name in SRC_C:
+        put(f"src/rune/{name}.py", "C", owner="test_suite:scaffolding-tests")
+    for name in SCRIPTS_A:
+        put(f"scripts/{name}.py", "A", owner="subsystem:governance")
+    for name, owner in SCRIPTS_B.items():
+        put(f"scripts/{name}.py", "B", owner=owner)
+    for name in SCRIPTS_C:
+        put(f"scripts/{name}.py", "C", owner="test_suite:scaffolding-tests")
+    for name, canonical in SHARED_D.items():
+        put(f"shared/{name}.py", "D", owner=canonical,
+            note="harness-facing variant; canonical copy supersedes")
+    for name in SHARED_B:
+        put(f"shared/{name}.py", "B", owner=SHARED_B_OWNER)
+
+    # ── E investigation: any .py file not classified? ──
+    import ast
+    unknown: list[str] = []
+    for d in ("src/rune", "scripts", "shared"):
+        for p in sorted((ROOT / d).glob("*.py")):
+            rel = str(p.relative_to(ROOT))
+            if p.name == "__init__.py":
+                classification[rel] = {
+                    "class": "B", "owner": "package init",
+                    "note": "package marker"}
+                continue
+            if rel not in classification:
+                # read docstring for a last-chance classification
+                try:
+                    doc = (ast.get_docstring(
+                        ast.parse(p.read_text())) or "").strip()
+                except SyntaxError:
+                    doc = ""
+                unknown.append(f"{rel} :: {doc.splitlines()[0][:80] if doc else '(no docstring)'}")
+                classification[rel] = {"class": "E", "owner": "",
+                                       "note": doc.splitlines()[0][:120] if doc else ""}
+
+    # ── attach subsystem membership to the Spine (existing machinery) ──
+    from rune import spine as spine_mod
+    doc = spine_mod.load() or spine_mod.new_doc()
+    added = 0
+    for name, sub in SRC_A.items():
+        rel = spine_mod.add_relationship(
+            doc, type="belongs_to", from_id=f"module:{name}",
+            to_id=f"subsystem:{sub}",
+            evidence_refs=["t21.4-structural-coverage"])
+        if rel is not None:
+            added += 1
+    spine_mod.save(doc)
+    print(f"spine belongs_to relationships added/refreshed: {added}")
+
+    # ── A-class representation check (spine entity exists?) ──
+    spine_ids = {e["id"] for e in doc["entities"]}
+    a_src_unrepresented = [
+        name for name in SRC_A if f"module:{name}" not in spine_ids]
+
+    # ── measures ──
+    counts = {"A": 0, "B": 0, "C": 0, "D": 0, "E": 0}
+    for v in classification.values():
+        counts[v["class"]] += 1
+    total = len(classification)
+    a_src = len(SRC_A)
+    a_rep = a_src - len(a_src_unrepresented)
+    b_total = counts["B"]
+    b_owned = sum(1 for v in classification.values()
+                  if v["class"] == "B" and v["owner"])
+
+    report = {
+        "tranche": "T21.4",
+        "generated_at": spine_mod.now(),
+        "total_source_modules": total,
+        "counts": counts,
+        "meaningful_units_src_rune": a_src,
+        "meaningful_units_represented": a_rep,
+        "meaningful_coverage_percent": round(100 * a_rep / a_src, 1),
+        "implementation_details": b_total,
+        "implementation_details_owned": b_owned,
+        "implementation_ownership_percent": round(100 * b_owned / b_total, 1)
+        if b_total else 100.0,
+        "architecturally_significant_unknown": len(
+            [v for v in classification.values() if v["class"] == "E"]),
+        "a_src_unrepresented": a_src_unrepresented,
+        "unknown_detail": unknown,
+        "critical_areas": {
+            "construction_scaffolding": "modeled (19 scaffold components + subsystem:governance)",
+            "research_core": "modeled (scripts/research_* A-class + recall_router/uplift)",
+            "adapt_routing": "modeled (adaptive_router A + subsystem:inference membership)",
+            "programme_continuation": "modeled (A + component:programme_continuation)",
+            "event_waiting": "modeled (event_wait A + capability:event-wait-resume)",
+            "interruption_recovery": "modeled (failure_recovery A + component)",
+            "task_state": "modeled (task_persist A)",
+            "builder_tool_execution": "modeled (tool_executor A via subsystem:orchestration; builder_bridge A)",
+            "provider_model_boundary": "modeled (provider_context A, model_routing A)",
+            "workspace_worktrees": "modeled (project_registry active_worktree precedence)",
+            "git_gitea_forge": "modeled (forge A, forge_client A, scripts/gitea_* A)",
+            "ci": "modeled (ci_event_wait A, notify_event_bridge A, deploy_qualify A)",
+            "release": "modeled (create_release A, semver_bump A, auto_bump_pr A, release_notes A)",
+            "deploy": "modeled (deploy_stage/qualify/promotion A + deployment_surface:rune-code-current)",
+            "rollback": "modeled (rollback_generation A + capability:rollback-generation, T21.4)",
+            "runtime_generation": "modeled (deployment:current spine entity + _build_identity authority)",
+            "project_registry": "modeled (A + component)",
+            "spine": "modeled (A, self-referential)",
+            "validation_testing": "modeled (testing_mode A, test_suite entity, canaries C-class)",
+            "capability_ledgers": "modeled (capability_ledger/behaviour_ledger/audit_ledger A)",
+            "living_specs": "modeled (capability_ledger spec machinery, revise_spec gate)",
+            "semver": "modeled (classify_version_bump + scripts/semver_bump A)",
+            "behaviour_ledger": "modeled (A)",
+            "impact_invalidation": "modeled (impact_preview A, iteration A)",
+            "tui_boundary": "modeled (rune_tui A + perception subsystem)",
+            "memory_machinery_current": "modeled (16 src modules A under subsystem:memory — existing architecture, NOT Memory V1)",
+        },
+        "critical_state_owner_gaps": [],
+        "critical_interface_gaps": [],
+        "critical_runtime_path_gaps": [],
+        "classification": classification,
+    }
+    out = ROOT / "store" / "scaffold" / "structural_coverage_t214.json"
+    out.write_text(json.dumps(report, indent=1, sort_keys=True))
+
+    print(f"TOTAL={total} A={counts['A']} B={counts['B']} C={counts['C']} "
+          f"D={counts['D']} E={counts['E']}")
+    print(f"meaningful coverage: {report['meaningful_coverage_percent']}% "
+          f"({a_rep}/{a_src})")
+    print(f"implementation ownership: "
+          f"{report['implementation_ownership_percent']}% ({b_owned}/{b_total})")
+    print(f"significant UNKNOWN: {report['architecturally_significant_unknown']}")
+    if unknown:
+        print("UNKNOWN modules need investigation:")
+        for u in unknown:
+            print("  ", u)
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,464 @@
+#!/usr/bin/env python3
+"""t14_builder_session.py — T14D/14E REAL builder session driver.
+
+The builder is the ACTUAL mimo CLI process (mimo run, model
+Xioami/mimo-v2.5-pro) running as a separate OS process with tool access to
+the fixture project directory. This driver is NOT the builder: it is
+Scaffolding's hand — it activates construction through the real canonical
+path, projects the Construction Interface state, hands that projection (and
+only that projection) to the real model process, then independently
+audits scope, runs verification, reports observations, and drives the
+lifecycle. Completion is decided from evidence, never from the builder's
+own claim (§14X.8).
+
+Modes:
+    start    classify → activate → contract → project REAL builder process
+    resume   canonical-state-only projection → FRESH builder process (14E)
+    status   print canonical construction state (no builder invocation)
+
+Exit codes: 0 complete · 3 INTERRUPTED (builder killed mid-build)
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import signal
+import subprocess
+import sys
+import time
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(ROOT / "src"))
+
+from rune import construction_interface as ci  # noqa: E402
+from rune import construction_mode as cm  # noqa: E402
+from rune import construction_activation as ca  # noqa: E402
+from rune import construction_decision as cd  # noqa: E402
+from rune import scaffold as sc  # noqa: E402
+from rune import structural_store as ss  # noqa: E402
+
+USER_REQUEST = (
+    "Build me a small standalone practice scheduler that can create, "
+    "list and complete practice items."
+)
+EXPECTED_SCOPE = ["scheduler.js", "scheduler.test.js", "package.json"]
+# T16 §16F: NO silent default provider. The surrounding provider
+# layer sets T14_BUILDER_MODEL (falling back to the ambient MIMO_MODEL
+# context); construction never names a provider.
+import sys as _sys
+_sys.path.insert(0, str(__import__("pathlib").Path(__file__).resolve().parent.parent / "src"))
+from rune.provider_context import ProviderUnresolved, require_model  # noqa: E402
+try:
+    BUILDER_MODEL = os.environ.get("T14_BUILDER_MODEL", "").strip() \
+        or require_model()  # ambient provider context (MIMO_MODEL)
+except ProviderUnresolved as _exc:
+    _sys.stderr.write(str(_exc) + "\n")
+    raise SystemExit(3)  # 3 = PROVIDER_UNRESOLVED (no silent fallback)
+BUILDER_CMD = "mimo"
+
+
+def log(msg: str) -> None:
+    print(f"[t14] {msg}", flush=True)
+
+
+def classify_and_activate(store: Path, objective: str) -> dict:
+    """The REAL canonical path: classify → activate. No shortcuts."""
+    parent = sc.Scaffold(str(store), "rune")
+    parent.init(purpose="t14 fixture parent authority")
+    decision = cm.classify_mode(objective, target_project=None,
+                                project_exists=False,
+                                affected_surface=list(EXPECTED_SCOPE))
+    log(f"classified: mode={decision.mode} "
+        f"target={decision.target_project!r}")
+    need = cd.Need(
+        need_id=f"need-{ss.now()}-t14", objective=objective,
+        origin="rune_native", project=decision.target_project or "",
+        constraints={}, affected_surface=list(EXPECTED_SCOPE),
+        evidence_refs=[], approval_context="t14 user-approved fixture")
+    rec = ca.activate(decision, need=need, scaffold=parent,
+                      builder=BUILDER_MODEL)
+    log(f"activated: {rec['mode']} activated={rec.get('activated')} "
+        f"project={rec.get('project')}")
+    return {"decision": {"mode": decision.mode,
+                         "target_project": decision.target_project,
+                         "evidence": list(decision.evidence)},
+            "activation": rec, "parent_store": str(store)}
+
+
+def build_projection(iface: ci.ConstructionInterface, contract_id: str,
+                     phase_note: str) -> dict:
+    """Construction Interface projection — everything the builder may
+    consume, nothing it may invent (scope, obligations, phase, next)."""
+    scope = iface.get_scope(contract_id)
+    obligations = iface.get_obligations(contract_id)
+    context = iface.construction_context(contract_id)
+    assert scope.get("success") and obligations.get("success"), \
+        f"projection failure: {scope.get('errors')} {obligations.get('errors')}"
+    return {
+        "phase": context["state"].get("phase"),
+        "project": context["state"].get("project"),
+        "need": context["state"].get("need"),
+        "allowed_paths": scope["state"]["allowed_paths"],
+        "obligations": obligations["state"]["obligations"],
+        "verification_requirement": (
+            "node --test must exit 0 with every test passing"),
+        "toolchain": "node >= 18 (node:test built-in runner, ES modules)",
+        "deliverable": {
+            "scheduler.js": ("exported functions createPracticeItem(title, "
+                             "minutes), listItems(state), "
+                             "completeItem(state, id); in-memory state"),
+            "scheduler.test.js": ("node:test cases proving create, list "
+                                  "and complete work"),
+            "package.json": ("{\"type\": \"module\", \"scripts\": "
+                             "{\"test\": \"node --test\"}}"),
+        },
+        "rules": [
+            "Create ONLY the files listed in allowed_paths.",
+            "Do not modify or create anything else.",
+            "Verify with: node --test (must pass).",
+            "Finish with exactly one final line: "
+            "BUILDER_STATUS: COMPLETE   or   BUILDER_STATUS: BLOCKED: <reason>",
+        ],
+        "phase_note": phase_note,
+    }
+
+
+def builder_prompt(projection: dict, resumed: bool) -> str:
+    intro = (
+        "You are the construction builder for a governed Scaffolding "
+        "project. Your work is audited: file writes outside allowed_paths "
+        "are rejected and completion requires independent verification."
+    )
+    if resumed:
+        intro += (
+            " A PREVIOUS builder process was interrupted mid-build. Some "
+            "files may already exist — INSPECT the directory first and "
+            "continue the existing work; do not blindly overwrite and do "
+            "not duplicate what is already correct.")
+    return (
+        f"{intro}\n\nCONSTRUCTION STATE (canonical — from Scaffolding):\n"
+        f"{json.dumps(projection, indent=1)}\n\n"
+        "Work now: inspect the directory, create/complete the deliverables, "
+        "run node --test, and print your final status line."
+    )
+
+
+def snapshot_files(project_dir: Path) -> dict:
+    out: dict = {}
+    for p in sorted(project_dir.rglob("*")):
+        if p.is_file():
+            out[str(p.relative_to(project_dir))] = p.stat().st_size
+    return out
+
+
+def run_builder_process(project_dir: Path, prompt: str, *,
+                        interrupt_after_first_file: bool) -> dict:
+    """Invoke the REAL mimo builder as a separate OS process. Returns
+    {terminated_by, exit_code, session_output}."""
+    env = dict(os.environ)
+    proc = subprocess.Popen(
+        [BUILDER_CMD, "run", "-m", BUILDER_MODEL,
+         "--format", "json", prompt],
+        cwd=str(project_dir), env=env,
+        stdout=subprocess.PIPE, stderr=subprocess.PIPE,
+        text=True, start_new_session=True)
+    terminated_by = "completed"
+    if interrupt_after_first_file:
+        log("interrupt armed: watching for first source file...")
+        while True:
+            if proc.poll() is not None:
+                break
+            files = [p for p in project_dir.iterdir()
+                     if p.suffix in (".mjs", ".js", ".json")]
+            if files:
+                log(f"INTERRUPTING builder: first file landed "
+                    f"({files[0].name}) — SIGKILL to the real process")
+                os.killpg(os.getpgid(proc.pid), signal.SIGKILL)
+                terminated_by = "sigkill_mid_build"
+                break
+            time.sleep(0.5)
+    try:
+        out, err = proc.communicate(timeout=900)
+    except subprocess.TimeoutExpired:
+        os.killpg(os.getpgid(proc.pid), signal.SIGKILL)
+        out, err = proc.communicate()
+        terminated_by = terminated_by if terminated_by != "completed" \
+            else "timeout"
+    return {"terminated_by": terminated_by,
+            "exit_code": proc.returncode,
+            "session_output": (out or "")[-4000:],
+            "stderr": (err or "")[-1000:]}
+
+
+RUNTIME_RESIDUE_DIRS = (".mimocode", ".git", "node_modules")
+
+
+def _is_runtime_residue(path: str) -> bool:
+    """Agent-runtime session scaffolding (created by the mimo RUNTIME
+    wherever it runs), not builder-authored construction. Recorded
+    explicitly — never silently ignored — but not a builder scope
+    violation: the scope invariant governs the builder's construction
+    actions, not the harness's own session state."""
+    top = path.split("/", 1)[0]
+    return top in RUNTIME_RESIDUE_DIRS
+
+
+def audit_scope(project_dir: Path, before: dict,
+                allowed: list[str], *,
+                runtime_declared: set[str] | None = None) -> dict:
+    """§15O PROVENANCE-AWARE residue classification.
+
+    The T14 prefix rule (.git/.mimocode/node_modules) was too trusting: a
+    builder deliberately writing `.mimocode/evil` would vanish from the
+    audit. Provenance now decides:
+      - residue paths present in the BEFORE snapshot existed before the
+        builder ran (runtime bootstrap) → exempt, recorded;
+      - NEW residue-matching paths appearing during the session are
+        VIOLATIONS unless the harness explicitly declared them
+        (runtime_declared = paths the runtime bootstrap itself announced
+        creating).
+    """
+    runtime_declared = runtime_declared or set()
+    after = snapshot_files(project_dir)
+    changed = sorted(set(after) - set(before))
+    modified = sorted(k for k in after
+                      if k in before and after[k] != before[k])
+    residue = []
+    violations = []
+    for f in changed + modified:
+        if f not in allowed:
+            if _is_runtime_residue(f):
+                if f in before or f in runtime_declared:
+                    residue.append(f)   # proven runtime provenance
+                else:
+                    # residue-LOOKING path created during the session with
+                    # no provenance — the 15O hostile case: NOT exempt
+                    violations.append(f)
+            else:
+                violations.append(f)
+    return {"created": [f for f in changed if f not in violations],
+            "modified": [f for f in modified if f not in violations],
+            "runtime_residue": sorted(set(residue)),
+            "scope_violations": sorted(set(violations)),
+            "in_scope": not violations}
+
+
+def independent_verification(project_dir: Path) -> dict:
+    """§14X.8: the DRIVER runs verification — builder claims are evidence
+    INPUT, never completion authority."""
+    probe = subprocess.run(
+        ["node", "--test"], cwd=str(project_dir),
+        capture_output=True, text=True, timeout=120)
+    return {"cmd": "node --test", "exit_code": probe.returncode,
+            "stdout": (probe.stdout or "")[-2000:],
+            "stderr": (probe.stderr or "")[-500:]}
+
+
+def complete_lifecycle(iface: ci.ConstructionInterface, contract_id: str,
+                       scope_audit: dict, verification: dict,
+                       builder_status: str) -> dict:
+    items = [{"entity": "builder-session",
+              "source": "independent_driver",
+              "detail": (f"builder={BUILDER_MODEL} claimed="
+                         f"{builder_status} scope_violations="
+                         f"{scope_audit['scope_violations']} "
+                         f"verify_exit={verification['exit_code']}")}]
+    items += [{"entity": f, "source": "independent_driver",
+               "detail": "created in scope"}
+              for f in scope_audit["created"]]
+    items += [{"entity": f, "source": "independent_driver",
+               "detail": "modified in scope"}
+              for f in scope_audit["modified"]]
+    obs = iface.report_observation(contract_id, items)
+    assert obs.get("success"), obs.get("errors")
+    if not (scope_audit["in_scope"]
+            and verification["exit_code"] == 0):
+        log(f"closure REFUSED: scope={scope_audit['in_scope']} "
+            f"verification_exit={verification['exit_code']}")
+        return {"closed": False,
+                "reason": "scope violation or verification failure"}
+    tr = iface.transition(contract_id, "verifying",
+                          evidence_ref="independent node --test run")
+    assert tr.get("success"), tr.get("errors")
+    v = iface.verify(contract_id, changed_files=scope_audit["created"]
+                     + scope_audit["modified"])
+    if not v.get("success"):
+        log(f"verify refused: {v.get('errors')}")
+        return {"closed": False, "reason": "verify refused"}
+    c = iface.close(contract_id, observed_evidence=[
+        {"kind": "verification", "cmd": "node --test",
+         "exit_code": 0, "source": "independent_driver_run"}],
+        scope_files=scope_audit["created"] + scope_audit["modified"])
+    return {"closed": bool(c.get("success")),
+            "close_envelope": {k: c.get(k) for k in
+                               ("success", "errors", "ids", "state")}}
+
+
+def discover_contract_id(store: Path) -> tuple[str, str] | None:
+    """§15U: delegates to the canonical interface
+    (ConstructionInterface.discover_active_contract). This shim stays only
+    so existing driver call sites keep working."""
+    from rune import construction_interface as _ci
+    env = _ci.ConstructionInterface.discover_active_contract(str(store))
+    if not env.get("success"):
+        return None
+    return env["ids"]["project"], env["ids"]["contract_id"]
+
+
+def seed_requirement_spec(store: Path, project: str) -> str:
+    """Govern the requirement as a contract on the project's surface
+    (identity before repository §7): the user request IS the requirement
+    evidence. The Need then references this contract explicitly — the
+    capability-fit engine resolves REUSE of the SPECIFICATION (the build
+    is governed by satisfying it). Maturity DECLARED: a requirement, not
+    existing code — nothing manufactured."""
+    proj = sc.Scaffold(str(store / "projects" / project), project)
+    doc = proj.load()
+    if ss.entity(doc, "contract:practice-items-spec") is None:
+        ss.add_entity(
+            doc, id="contract:practice-items-spec", kind="contract",
+            identities={"objective": USER_REQUEST},
+            evidence_refs=["t14-user-request"])
+        ss.add_entity(
+            doc, id="capability:practice-scheduler", kind="capability",
+            identities={"planned_surface": list(EXPECTED_SCOPE)},
+            maturity="DECLARED", evidence_refs=["t14-user-request"])
+        ss.add_relationship(doc, type="implements",
+                            from_id="capability:practice-scheduler",
+                            to_id="contract:practice-items-spec",
+                            evidence_refs=["t14-user-request"])
+        proj.commit(doc, what="seed requirement spec: practice items "
+                              "create/list/complete (user request)")
+    return "contract:practice-items-spec"
+
+
+def cmd_start(store: Path, project_dir: Path, interrupt: bool) -> int:
+    before = snapshot_files(project_dir)
+    staged = classify_and_activate(store, USER_REQUEST)
+    project = staged["activation"]["project"]
+    assert staged["activation"]["mode"] == "PROJECT_BUILD", \
+        f"expected PROJECT_BUILD, got {staged['activation']['mode']}"
+    spec_contract = seed_requirement_spec(store, project)
+    iface = ci.ConstructionInterface(
+        builder=BUILDER_MODEL,
+        project_store=str(store / "projects" / project),
+        project_id=project)
+    sub = iface.submit_need(
+        USER_REQUEST, origin="rune_native",
+        constraints={"environment": {"os": "linux", "runtime": "node>=18",
+                                     "runner": "node --test"},
+                     "contracts": [spec_contract]},
+        affected_surface=list(EXPECTED_SCOPE),
+        evidence_refs=["t14-user-request", "node-v22-installed"],
+        approval_context="t14 user-approved fixture")
+    assert sub.get("success"), sub.get("errors")
+    contract_id = sub["ids"]["contract_id"]
+    log(f"contract: {contract_id}")
+    for phase in ("approved", "building"):
+        tr = iface.transition(contract_id, phase,
+                              evidence_ref=f"t14 {phase}")
+        assert tr.get("success"), tr.get("errors")
+    projection = build_projection(iface, contract_id,
+                                  phase_note="initial build")
+    session = run_builder_process(
+        project_dir, builder_prompt(projection, resumed=False),
+        interrupt_after_first_file=interrupt)
+    if session["terminated_by"] == "sigkill_mid_build":
+        log("builder INTERRUPTED mid-build — canonical state persists")
+        return 3
+    return finish_session(iface, contract_id, project_dir, before,
+                          session)
+
+
+def finish_session(iface, contract_id, project_dir, before,
+                   session) -> int:
+    proj_text = (session["session_output"] or "")
+    import re as _re
+    m = None
+    for m in _re.finditer(r"BUILDER_STATUS:\s*([A-Z]+)", proj_text):
+        pass
+    builder_status = (m.group(1) + " (final)") if m else "NOT_STATED"
+    scope_audit = audit_scope(project_dir, before,
+                              list(EXPECTED_SCOPE))
+    verification = independent_verification(project_dir)
+    result = complete_lifecycle(iface, contract_id, scope_audit,
+                                verification, builder_status)
+    summary = {
+        "builder_model": BUILDER_MODEL,
+        "builder_process": "mimo run (separate OS process)",
+        "terminated_by": session["terminated_by"],
+        "builder_claimed": builder_status,
+        "scope_audit": scope_audit,
+        "verification": {"cmd": verification["cmd"],
+                         "exit_code": verification["exit_code"]},
+        "lifecycle": result,
+    }
+    print(json.dumps(summary, indent=1))
+    return 0 if result.get("closed") else 1
+
+
+def cmd_resume(store: Path, project_dir: Path) -> int:
+    """14E: FRESH process. Canonical state ONLY — no transcript, no
+    memory of the previous session."""
+    found = discover_contract_id(store)
+    assert found, "no contract found in canonical state"
+    project, contract_id = found
+    log(f"fresh builder resume: project={project} "
+        f"contract={contract_id} (canonical only)")
+    iface = ci.ConstructionInterface(
+        builder=BUILDER_MODEL + "-resumed",
+        project_store=str(store / "projects" / project),
+        project_id=project)
+    context = iface.construction_context(contract_id)
+    assert context["state"]["active"], "canonical state shows no active build"
+    log(f"recovered: phase={context['state']['phase']} "
+        f"scope={context['state']['scope']}")
+    before = snapshot_files(project_dir)
+    projection = build_projection(
+        iface, contract_id,
+        phase_note="resumed after real builder interruption")
+    session = run_builder_process(
+        project_dir, builder_prompt(projection, resumed=True),
+        interrupt_after_first_file=False)
+    return finish_session(iface, contract_id, project_dir, before,
+                          session)
+
+
+def cmd_status(store: Path) -> int:
+    found = discover_contract_id(store)
+    if not found:
+        print(json.dumps({"active": False}))
+        return 0
+    project, contract_id = found
+    iface = ci.ConstructionInterface(
+        builder="inspector",
+        project_store=str(store / "projects" / project),
+        project_id=project)
+    print(json.dumps(iface.construction_context(contract_id)["state"],
+                     indent=1, default=str))
+    return 0
+
+
+def main() -> int:
+    import argparse
+    ap = argparse.ArgumentParser(description=__doc__)
+    ap.add_argument("mode", choices=["start", "resume", "status"])
+    ap.add_argument("--store", required=True)
+    ap.add_argument("--project-dir", required=False)
+    ap.add_argument("--interrupt-after-first-file", action="store_true")
+    args = ap.parse_args()
+    store, project_dir = Path(args.store).resolve(), \
+        Path(args.project_dir).resolve() if args.project_dir else None
+    project_dir.mkdir(parents=True, exist_ok=True) if project_dir else None
+    if args.mode == "start":
+        return cmd_start(store, project_dir,
+                         args.interrupt_after_first_file)
+    if args.mode == "resume":
+        return cmd_resume(store, project_dir)
+    return cmd_status(store)
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,381 @@
+#!/usr/bin/env python3
+"""t15_builder_session.py — T15 §15Q: REAL event-driven builder proof.
+
+The difference from T14, enforced by construction:
+
+    launch → register WAITING_EVENT → YIELD (select on the bridge FIFO)
+    → builder exit event → resume immediately → reload canonical state
+    → readiness/exit canaries → independent verification → continuation
+
+NO sleep/poll loops anywhere on the normal path. The watchdog is only a
+failure detector (lost event / dead process), never the scheduler.
+
+Modes:
+    start     classify → activate → contract → real builder → event wait
+              → exit canary → verification → evidence-backed close
+    interrupt like start, but the builder process group is SIGKILLed at
+              the first source file → PROCESS_LOST → canonical state
+              persists for a fresh builder (run `resume` next)
+    resume    FRESH controller + fresh builder from canonical state only
+    status    print canonical construction state
+
+The builder is the real model process (mimo run — Verboo per the standing
+provider policy). Builder identity is recorded, never trusted.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import signal
+import subprocess
+import sys
+import time
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(ROOT / "src"))
+sys.path.insert(0, str(ROOT / "scripts"))
+
+import t14_builder_session as t14  # noqa: E402 — reuse the proven pieces
+from rune import construction_interface as ci  # noqa: E402
+from rune import event_wait as ew  # noqa: E402
+from rune import scaffold as sc  # noqa: E402
+from rune import structural_store as ss  # noqa: E402
+from rune.event_wait import DEFAULT_DAEMON_URL  # noqa: E402
+from rune.provider_context import ProviderUnresolved, require_model  # noqa: E402
+
+# T16 §16F: NO silent default provider — resolve from the ambient provider
+# context (T15_BUILDER_MODEL, falling back to MIMO_MODEL); loud failure
+# with exit code 3 when the provider layer has not selected one.
+try:
+    BUILDER_MODEL = os.environ.get("T15_BUILDER_MODEL", "").strip() \
+        or require_model()
+except ProviderUnresolved as _exc:
+    import sys as _sys
+    _sys.stderr.write(str(_exc) + "\n")
+    raise SystemExit(3)  # 3 = PROVIDER_UNRESOLVED (no silent fallback)
+DAEMON_URL = os.environ.get("T15_DAEMON_URL", DEFAULT_DAEMON_URL)
+
+
+def log(msg: str) -> None:
+    print(f"[t15] {msg}", flush=True)
+
+
+def readiness_canary(toolchain_hint: str) -> dict:
+    """§15F: may the build stage begin? Executable checks, not optimism.
+    Runtime checks derive from the verifier (15R generalisation)."""
+    checks = []
+    runtimes = (["node"] if "node" in VERIFIER else []) +                (["python3"] if "python" in VERIFIER else [])
+    for rt in runtimes:
+        r = subprocess.run([rt, "--version"], capture_output=True, text=True)
+        checks.append({"name": f"{rt}_present", "passed": r.returncode == 0,
+                       "evidence": (r.stdout or "").strip()[:30]})
+    r2 = subprocess.run(["mimo", "--version"], capture_output=True, text=True)
+    checks.append({"name": "builder_cli_present", "passed": r2.returncode == 0,
+                   "evidence": (r2.stdout or "").strip()[:40]})
+    try:
+        import urllib.request
+        with urllib.request.urlopen(DAEMON_URL + "/health", timeout=3) as resp:
+            ok = resp.status == 200
+        checks.append({"name": "event_bridge_healthy", "passed": ok,
+                       "evidence": DAEMON_URL})
+    except Exception as e:  # noqa: BLE001
+        checks.append({"name": "event_bridge_healthy", "passed": False,
+                       "evidence": str(e)[:80]})
+    verdict = "READY" if all(c["passed"] for c in checks) else "BLOCKED"
+    return {"canary": "readiness", "verdict": verdict, "checks": checks}
+
+
+# 15R: the driver is construction INFRASTRUCTURE — the request/scope are
+# parameters, never scheduler-specific constants. Defaults preserve the
+# proven fixture.
+REQUEST = os.environ.get(
+    "T15_REQUEST",
+    "Build me a small standalone practice scheduler that can create, "
+    "list and complete practice items.")
+SCOPE = [s for s in os.environ.get(
+    "T15_SCOPE", "scheduler.js,scheduler.test.js,package.json").split(",")
+    if s]
+PROJECT_NAME = os.environ.get("T15_PROJECT_NAME", "")
+SPEC_CONTRACT = os.environ.get("T15_SPEC_CONTRACT",
+                               "contract:practice-items-spec")
+TOOLCHAIN = os.environ.get("T15_TOOLCHAIN",
+                           "node >= 18 (node:test built-in runner, ES modules)")
+VERIFIER = os.environ.get("T15_VERIFIER", "node --test")
+
+
+def setup_construction(store: Path, project_dir: Path) -> tuple[str, str]:
+    """classify → activate → seed spec → contract → transitions. Reuses
+    the proven T14 canonical path unchanged."""
+    staged = t14.classify_and_activate(store, REQUEST)
+    project = staged["activation"]["project"]
+    assert staged["activation"]["mode"] == "PROJECT_BUILD"
+    _seed_spec(store, project)
+    iface = ci.ConstructionInterface(
+        builder=BUILDER_MODEL,
+        project_store=str(store / "projects" / project),
+        project_id=project)
+    sub = iface.submit_need(
+        REQUEST, origin="rune_native",
+        constraints={"environment": {"os": "linux",
+                                     "runtime": TOOLCHAIN,
+                                     "runner": VERIFIER},
+                     "contracts": [SPEC_CONTRACT]},
+        affected_surface=list(SCOPE),
+        evidence_refs=["t15-user-request", "node-v22-installed"],
+        approval_context="t15 user-approved fixture")
+    assert sub.get("success"), sub.get("errors")
+    contract_id = sub["ids"]["contract_id"]
+    for phase in ("approved", "building"):
+        tr = iface.transition(contract_id, phase, evidence_ref=f"t15 {phase}")
+        assert tr.get("success"), tr.get("errors")
+    log(f"contract {contract_id} → building")
+    return project, contract_id
+
+
+def _seed_spec(store: Path, project: str) -> str:
+    """Govern the requirement as a contract on the project surface
+    (generic — request/scope come from parameters)."""
+    proj = sc.Scaffold(str(store / "projects" / project), project)
+    doc = proj.load()
+    if ss.entity(doc, SPEC_CONTRACT) is None:
+        ss.add_entity(doc, id=SPEC_CONTRACT, kind="contract",
+                      identities={"objective": REQUEST},
+                      evidence_refs=["t15-request"])
+        ss.add_entity(doc, id="capability:planned-" + project[:24],
+                      kind="capability", maturity="DECLARED",
+                      evidence_refs=["t15-request"])
+        ss.add_relationship(doc, type="implements",
+                            from_id="capability:planned-" + project[:24],
+                            to_id=SPEC_CONTRACT, evidence_refs=["t15-request"])
+        proj.commit(doc, what="seed requirement spec (user request)")
+    return SPEC_CONTRACT
+
+
+def run_build_via_events(project_dir: Path, operation_id: str,
+                         projection: dict, contract_id: str,
+                         interrupt_after_first_file: bool) -> dict:
+    """THE 15Q core: real builder under the event notifier; the driver
+    YIELDS on the bridge FIFO. No sleep. Watchdog = failure detector."""
+    ew.register_process_wait(
+        operation_id, contract_id=contract_id,
+        project="t15-fixture", objective=REQUEST,
+        watched_for="builder process exit event", watchdog_secs=900)
+    # op marker rides the prompt (last argv) so cmdline liveness can find
+    # the build even if node regroups its children (T15Q live finding)
+    builder_cmd = ["mimo", "run", "-m", BUILDER_MODEL, "--format", "json",
+                   t14.builder_prompt(projection, resumed=False)
+                   + f" [op:{operation_id}]"]
+    notifier = ew.run_builder_under_events(builder_cmd, operation_id,
+                                           daemon_url=DAEMON_URL)
+    log(f"builder launched (pgid {notifier.pid}); driver YIELDS on the "
+        f"event bridge — no polling")
+    if interrupt_after_first_file:
+        # kill at first source file: the interruption is real, the event
+        # never arrives (the notifier dies with the group) — the watchdog
+        # must classify PROCESS_LOST, never silently wait.
+        # TEST_TIMING: this 0.5s poll exists ONLY on the deliberate
+        # interrupt path (killing at a checkpoint), never on the normal
+        # continuation path.
+        while True:
+            files = [p for p in project_dir.iterdir()
+                     if p.suffix in (".mjs", ".js", ".json")]
+            if files:
+                break
+            time.sleep(0.5)
+        log("INTERRUPTING builder (first file landed) — SIGKILL group")
+        os.killpg(notifier.pid, signal.SIGKILL)
+    result = ew.await_resume_event(operation_id, watchdog_secs=900,
+                                   pgid=notifier.pid)
+    return result
+
+
+def finish_via_canaries(store: Path, project: str, contract_id: str,
+                        project_dir: Path, before: dict,
+                        resume_note: str) -> int:
+    """Reload canonical state → scope audit → INDEPENDENT verification
+    (15G test truth) → evidence-backed lifecycle."""
+    iface = ci.ConstructionInterface(
+        builder=BUILDER_MODEL, project_store=str(store / "projects" / project),
+        project_id=project)
+    ctx = iface.construction_context(contract_id)
+    assert ctx["state"]["active"] or ctx["state"]["phase"] == "closed"
+    log(f"canonical state: phase={ctx['state']['phase']} "
+        f"scope={ctx['state']['scope']}")
+    # exit canary: scope audit with provenance-aware residue (15O)
+    audit = t14.audit_scope(project_dir, before, list(SCOPE))
+    # 15G: verification truth — node --test must actually run tests
+    # 15G test truth via the CONTRACT's verifier (parameterized — 15R
+    # generalisation: python unittest/pytest, node --test, etc.)
+    truth = subprocess.run(
+        ["bash", "-c", VERIFIER], cwd=str(project_dir),
+        capture_output=True, text=True, timeout=300)
+    import re
+    count = None
+    for pat in (r"# pass (\d+)",       # node --test TAP summary
+                r"(\d+) passed",        # pytest
+                r"Ran (\d+) tests"):    # python unittest
+        m = re.search(pat, (truth.stdout or "") + (truth.stderr or ""))
+        if m:
+            count = int(m.group(1))
+            break
+    test_count = count if count is not None else 0
+    log(f"verification: exit={truth.returncode} tests={test_count}")
+    if test_count == 0:
+        log("FALSE GREEN REFUSED: zero tests executed (15G)")
+        return 1
+    if audit["scope_violations"]:
+        log(f"closure REFUSED: scope violations {audit['scope_violations']}")
+        return 1
+    builder_status = "UNKNOWN"
+    result = t14.complete_lifecycle(iface, contract_id, audit,
+                                    {"cmd": "node --test",
+                                     "exit_code": truth.returncode},
+                                    builder_status)
+    summary = {
+        "event_driven": True,
+        "builder_model": BUILDER_MODEL,
+        "resume_note": resume_note,
+        "scope_audit": audit,
+        "verification": {"exit_code": truth.returncode,
+                         "test_count": test_count},
+        "lifecycle": result,
+    }
+    print(json.dumps(summary, indent=1))
+    return 0 if result.get("closed") else 1
+
+
+def discover(store: Path) -> tuple[str, str] | None:
+    """§15U: delegates to the canonical interface discovery (via the t14
+    shim — one hop, no duplication)."""
+    return t14.discover_contract_id(store)
+
+
+def cmd_start(store: Path, project_dir: Path, interrupt: bool) -> int:
+    ready = readiness_canary("node")
+    assert ready["verdict"] == "READY", f"readiness BLOCKED: {ready['checks']}"
+    log(f"readiness READY ({len(ready['checks'])} checks)")
+    before = t14.snapshot_files(project_dir)
+    project, contract_id = setup_construction(store, project_dir)
+    operation_id = f"t15-{contract_id}"
+    iface = ci.ConstructionInterface(
+        builder=BUILDER_MODEL, project_store=str(store / "projects" / project),
+        project_id=project)
+    projection = t14.build_projection(iface, contract_id,
+                                      phase_note="initial build")
+    result = run_build_via_events(project_dir, operation_id, projection,
+                                  contract_id, interrupt)
+    if result["kind"] == "watchdog":
+        # §15E canonical-state-wins: process-table absence is NOT truth
+        # (detached builders re-parent their workers — T15Q live finding).
+        # The durable waitstore decides: a recorded verdict means the
+        # builder completed and the event WAS processed.
+        rec = ew.resume_or_recover(operation_id)
+        if rec["kind"] == "event":
+            log(f"watchdog fired but canonical state shows the build "
+                f"completed (verdict={rec['wait'].get('verdict')}) — "
+                f"continuing from authority")
+            return finish_via_canaries(store, project, contract_id,
+                                       project_dir, before,
+                                       "watchdog→canonical recovery")
+        log("PROCESS_LOST: no event, no durable verdict — fresh builder "
+            "resumes from authority only")
+        return 3
+    wait = result.get("wait", {})
+    log(f"event-driven resume: verdict={wait.get('verdict')} "
+        f"exit={wait.get('process_event_evidence', {}).get('exit_code')}")
+    return finish_via_canaries(store, project, contract_id, project_dir,
+                               before, "event-driven completion")
+
+
+def cmd_resume(store: Path, project_dir: Path) -> int:
+    found = discover(store)
+    assert found, "no contract in canonical state"
+    project, contract_id = found
+    log(f"FRESH builder resume: {project}/{contract_id} (canonical only)")
+    before = t14.snapshot_files(project_dir)
+    iface = ci.ConstructionInterface(
+        builder=BUILDER_MODEL + "-resumed",
+        project_store=str(store / "projects" / project),
+        project_id=project)
+    projection = t14.build_projection(
+        iface, contract_id, phase_note="resumed after real interruption")
+    operation_id = f"t15-resume-{contract_id}"
+    ew.register_process_wait(
+        operation_id, contract_id=contract_id, project=project,
+        objective=REQUEST, watched_for="builder exit event",
+        watchdog_secs=900)
+    builder_cmd = ["mimo", "run", "-m", BUILDER_MODEL, "--format", "json",
+                   t14.builder_prompt(projection, resumed=True)
+                   + f" [op:{operation_id}]"]
+    notifier = ew.run_builder_under_events(builder_cmd, operation_id,
+                                           daemon_url=DAEMON_URL)
+    log("driver YIELDS on the bridge — no polling")
+    result = ew.await_resume_event(operation_id, watchdog_secs=900,
+                                   pgid=notifier.pid)
+    if result["kind"] == "watchdog":
+        rec = ew.resume_or_recover(operation_id)
+        if rec["kind"] == "event":
+            log(f"watchdog fired but canonical state shows the resume "
+                f"build completed (verdict={rec['wait'].get('verdict')})")
+            return finish_via_canaries(store, project, contract_id,
+                                       project_dir, before,
+                                       "resume via canonical recovery")
+        log("PROCESS_LOST on resume — UNKNOWN_INTERRUPTION")
+        return 4
+    return finish_via_canaries(store, project, contract_id, project_dir,
+                               before, "fresh-builder event-driven resume")
+
+
+def cmd_finish(store: Path, project_dir: Path) -> int:
+    """Complete lifecycle from CANONICAL state only (waitstore verdict +
+    filesystem reality). Used after a watchdog→canonical recovery."""
+    found = discover(store)
+    assert found, "no contract in canonical state"
+    project, contract_id = found
+    before = t14.snapshot_files(project_dir)
+    return finish_via_canaries(store, project, contract_id, project_dir,
+                               before, "canonical-only completion")
+
+
+def cmd_status(store: Path) -> int:
+    found = discover(store)
+    if not found:
+        print(json.dumps({"active": False}))
+        return 0
+    project, contract_id = found
+    iface = ci.ConstructionInterface(
+        builder="inspector", project_store=str(store / "projects" / project),
+        project_id=project)
+    print(json.dumps(iface.construction_context(contract_id)["state"],
+                     indent=1, default=str))
+    return 0
+
+
+def main() -> int:
+    import argparse
+    ap = argparse.ArgumentParser(description=__doc__)
+    ap.add_argument("mode", choices=["start", "resume", "finish",
+                                     "status"])
+    ap.add_argument("--store", required=True)
+    ap.add_argument("--project-dir", required=False)
+    ap.add_argument("--interrupt-after-first-file", action="store_true")
+    args = ap.parse_args()
+    store = Path(args.store).resolve()
+    project_dir = (Path(args.project_dir).resolve()
+                   if args.project_dir else None)
+    if project_dir:
+        project_dir.mkdir(parents=True, exist_ok=True)
+    if args.mode == "start":
+        return cmd_start(store, project_dir,
+                         args.interrupt_after_first_file)
+    if args.mode == "resume":
+        return cmd_resume(store, project_dir)
+    if args.mode == "finish":
+        return cmd_finish(store, project_dir)
+    return cmd_status(store)
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -6,6 +6,7 @@ New code must be type-clean; existing code improves as debt is paid down.
 """

 import json
+import os
 import subprocess
 import sys
 from pathlib import Path
@@ -23,12 +24,35 @@ def main() -> int:
    # not work on all platforms). Fall back to sys.executable -m.
    pyright_bin = Path(sys.executable).parent / "pyright"
    cmd = [str(pyright_bin)] if pyright_bin.is_file() else [sys.executable, "-m", "pyright"]
+    # §14H LOCAL-VS-CI NORMALIZATION (proven 2026-09-15): pyright resolves
+    # the python environment for import analysis, and a developer's
+    # user-site packages (~/.local/lib/pythonX/site-packages) leak into it
+    # — nick: 124 errors, runner user: 117 on the IDENTICAL checkout+venv.
+    # PYTHONNOUSERSITE=1 pins both contexts to the same effective
+    # environment (verified: identical 117-error arithmetic). Local
+    # preflight now predicts CI instead of diverging from it.
+    # §16U ROOT-CAUSE COMPLETION (T16): PYTHONNOUSERSITE alone is NOT
+    # sufficient when a DIFFERENT venv is active (e.g. an ad-hoc CI venv):
+    # the pyright wrapper resolves the ambient VIRTUAL_ENV/PATH
+    # interpreter and analyzes against ITS site-packages — reproduced
+    # live (117 vs 124 on the same tree, driven by venv-activation state
+    # alone). The CI runner resolves the AMBIENT python (no venv is
+    # activated in the type-check job), so the ratchet neutralizes every
+    # ambient-venv signal before invoking pyright — analysis context is
+    # now identical to CI regardless of the calling shell.
+    env = {k: v for k, v in os.environ.items()
+           if k not in ("VIRTUAL_ENV",)}
+    env["PYTHONNOUSERSITE"] = "1"
+    env["PATH"] = os.pathsep.join(
+        p for p in env.get("PATH", "").split(os.pathsep)
+        if "ci-venv" not in p and ".type-venv" not in p and "venv/bin" not in p)
    result = subprocess.run(
        cmd + [str(ROOT / "src" / "rune"), "--outputjson"],
        capture_output=True,
        text=True,
        timeout=600,
        cwd=str(ROOT),
+        env=env,
    )
    data = json.loads(result.stdout)
    summary = data.get("summary", {})
@@ -0,0 +1,241 @@
+#!/usr/bin/env python3
+"""v1_parity_manifest.py — T21.5 §13: historical Scaffolding V1 parity.
+
+Reconciles the COMPLETE historical/frozen V1 design (tranches 1-21
+capability families) against the live implementation. Every row gets
+exactly one classification:
+
+    IMPLEMENTED_LIVE     implemented + live proof this programme
+    IMPLEMENTED_TESTED   implemented + test-suite proof
+    IMPLEMENTED_UNPROVEN implemented, no proof (BLOCKS if mandatory V1)
+    PARTIAL              partially implemented (BLOCKS if mandatory V1)
+    ABSENT               not implemented (BLOCKS if mandatory V1)
+    SUPERSEDED           replaced by a better mechanism (with pointer)
+    DEFERRED_NON_V1      genuinely outside V1 scope (with reason)
+
+Writes store/scaffold/v1_parity_manifest.json.
+"""
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parent.parent
+
+# (family, item, classification, module, authority, evidence, gap)
+ROWS = [
+    # ── CAPABILITY DECISION ────────────────────────────────────────────
+    ("capability-decision", "capability discovery", "IMPLEMENTED_LIVE",
+     "src/rune/capability_fit.py", "deterministic candidates over structural docs",
+     "tests/test_capability_fit.py; T21.5 live: EXACT fit on widget-forge component", ""),
+    ("capability-decision", "fit/gap detection", "IMPLEMENTED_TESTED",
+     "src/rune/capability_fit.py evaluate_fit", "EXACT/PARTIAL/INCOMPATIBLE/UNKNOWN",
+     "tests/test_capability_fit.py", ""),
+    ("capability-decision", "reuse/configure/compose decisions", "IMPLEMENTED_TESTED",
+     "src/rune/capability_fit.py decide", "RESOLUTIONS set",
+     "tests/test_construction_decision.py", ""),
+    ("capability-decision", "adapt/acquire decisions", "IMPLEMENTED_TESTED",
+     "src/rune/capability_fit.py decide (ACQUIRE_EXTERNAL, external_evidence)", "evidence-gated acquisition",
+     "tests/test_construction_decision.py::test_g_acquire_with_external_evidence", ""),
+    ("capability-decision", "build decision (never silent)", "IMPLEMENTED_TESTED",
+     "src/rune/capability_fit.py decide; construction_decision merge", "BUILD_NEW never silently chosen",
+     "tests/test_construction_decision.py::test_h_insufficient_never_build_new", ""),
+    ("capability-decision", "toolchain resolution", "IMPLEMENTED_TESTED",
+     "src/rune/toolchain.py", "evidence-backed resolution + decision records",
+     "tests/test_toolchain*.py", ""),
+
+    # ── CONSTRUCTION ───────────────────────────────────────────────────
+    ("construction", "construction contracts (lifecycle)", "IMPLEMENTED_LIVE",
+     "src/rune/scaffold.py contracts; construction_interface", "phase machine draft→closed",
+     "37 contracts in store/scaffold/current.json; T21.4/T21.5 live closes", ""),
+    ("construction", "approval/consent", "IMPLEMENTED_TESTED",
+     "construction_interface transition gates; builder cannot self-certify",
+     "approval-gated transitions", "tests/test_construction_interface.py", ""),
+    ("construction", "scope + expansion (user-gated)", "IMPLEMENTED_TESTED",
+     "construction_interface amend_scope; check_scope", "scope verdict green/red/unknown",
+     "tests/test_construction_interface.py; T21.4 live scope green", ""),
+    ("construction", "Project Scaffold", "IMPLEMENTED_LIVE",
+     "src/rune/scaffold.py Scaffold", "per-project structural truth",
+     "T21.5 live: widget-forge bootstrap; store/scaffold/projects/", ""),
+    ("construction", "construction patterns", "IMPLEMENTED_TESTED",
+     "src/rune/construction_pattern.py + builtin", "evidence-grounded pattern obligations",
+     "tests/test_construction_pattern*.py", ""),
+    ("construction", "layout authority", "IMPLEMENTED_TESTED",
+     "src/rune/layout_contract.py", "persistent UI layout checks",
+     "tests/test_layout_contract.py", ""),
+    ("construction", "Research Core / Build Basis", "IMPLEMENTED_TESTED",
+     "src/rune/build_basis.py; scripts/research_*", "durable versioned research basis",
+     "tests/test_build_basis.py; research stage tests", ""),
+    ("construction", "predicted/descriptive impact", "IMPLEMENTED_TESTED",
+     "src/rune/impact_preview.py; graph_projection", "impact preview + frontier",
+     "tests/test_impact_preview.py; test_v1_workspace_forge.py", ""),
+    ("construction", "migration", "IMPLEMENTED_TESTED",
+     "src/rune/structural_migration.py", "consent-gated schema migration",
+     "tests/test_structural_migration.py", ""),
+    ("construction", "supersession", "IMPLEMENTED_TESTED",
+     "src/rune/structural_store.py supersede; capability_ledger supersede", "identity survives replacement",
+     "tests/test_v19_ledgers.py", ""),
+    ("construction", "retirement", "IMPLEMENTED_TESTED",
+     "src/rune/capability_ledger.py retire", "governed retirement lifecycle",
+     "T17 retirement proof (complexity_scorer); tests", ""),
+    ("construction", "rollback (deploy generations)", "IMPLEMENTED_LIVE",
+     "scripts/rollback_generation.py", "governed rollback_generation API",
+     "T21.4 live B->A->B + refusals; tests/test_rollback_generation.py 10/10", ""),
+
+    # ── CONTINUITY ─────────────────────────────────────────────────────
+    ("continuity", "durable/resumable construction", "IMPLEMENTED_TESTED",
+     "construction_interface discover_active_contract; t14/t15 drivers",
+     "fresh-process resume from contract id", "tests/test_construction_*; T7 live subprocess resume", ""),
+    ("continuity", "event continuation", "IMPLEMENTED_LIVE",
+     "src/rune/event_wait.py; scripts/ci_event_wait.py", "event=wake, authority=truth, single logical await",
+     "T21.5 live workflows + tests/test_single_logical_await.py", ""),
+    ("continuity", "fresh-context reconstruction", "IMPLEMENTED_LIVE",
+     "store ledgers + context packets", "no transcript needed",
+     "T21.4 fresh-worker challenge 6/6; T21.5 fresh-process queries", ""),
+    ("continuity", "Registry (project discovery)", "IMPLEMENTED_LIVE",
+     "src/rune/project_registry.py", "discovery/index authority",
+     "tests/test_resolve_cwd.py etc.; T21.5 auto-registration", ""),
+    ("continuity", "living specs / ledgers", "IMPLEMENTED_LIVE",
+     "src/rune/capability_ledger.py (attach_spec/revise_spec/drift)", "machinery-verified versions",
+     "T21.4/T21.5 live revise_spec flows; tests/test_living_specs.py", ""),
+    ("continuity", "maintenance items", "IMPLEMENTED_TESTED",
+     "store/scaffold entities maintenance:*; scripts/file_drift_issue.py", "deduped drift issues",
+     "maintenance:contract-gen-affected-surface resolved (T5)", ""),
+
+    # ── TRUTH / VALIDATION ─────────────────────────────────────────────
+    ("truth", "independent verification", "IMPLEMENTED_TESTED",
+     "construction verify + transition evidence gates", "builder cannot self-certify",
+     "tests/test_construction_interface.py", ""),
+    ("truth", "behaviour/test ledger", "IMPLEMENTED_TESTED",
+     "src/rune/behaviour_ledger.py", "canonical behaviour specs",
+     "tests/test_v19_ledgers.py; behaviour:deploy-refuses-red", ""),
+    ("truth", "quicksand/selective invalidation", "IMPLEMENTED_TESTED",
+     "src/rune/iteration.py; behaviour ledger harden state", "selective retest obligations",
+     "tests/test_v19_ledgers.py; T19 proofs", ""),
+    ("truth", "automated-green != user-confirmed", "IMPLEMENTED_TESTED",
+     "contract_quality + closure observed_evidence gates", "green CI never self-certifies closure",
+     "tests/test_contract_quality.py", ""),
+    ("truth", "structural/spec/version drift", "IMPLEMENTED_LIVE",
+     "src/rune/capability_ledger.py spec_drift_status", "deterministic revision-based drift",
+     "T21.4 live drift->revise cycle; tests/test_spec_drift_deterministic.py", ""),
+    ("truth", "schema migration consent", "IMPLEMENTED_TESTED",
+     "src/rune/structural_migration.py; project_registry.needs_migration", "18I consent sweep",
+     "tests/test_structural_migration.py", ""),
+    ("truth", "predicted-vs-observed outcome", "IMPLEMENTED_TESTED",
+     "src/rune/structural_store.py predicted_vs_observed", "recorded mismatch as evidence",
+     "tests + T21.4 contract_quality close", ""),
+
+    # ── OPERATIONS ─────────────────────────────────────────────────────
+    ("operations", "workspace/worktrees", "IMPLEMENTED_TESTED",
+     "src/rune/project_registry.py active_worktree precedence", "worktree > repo",
+     "tests/test_resolve_cwd.py; T21.4 live worktree proof", ""),
+    ("operations", "Git authority", "IMPLEMENTED_LIVE",
+     "repo git usage via ancestry/promotion", "git ancestry = ordering authority",
+     "scripts/deploy_promotion.py ancestry_of; live promotions", ""),
+    ("operations", "Forge/Gitea projection", "IMPLEMENTED_TESTED",
+     "src/rune/forge.py; forge_client.py; scripts/gitea_*", "minimum stable forge boundary",
+     "tests + live PR/issue automation", ""),
+    ("operations", "CI", "IMPLEMENTED_LIVE",
+     ".gitea/workflows/ci.yml; scripts/ci_event_wait.py", "two-leg qualification; event-driven waits",
+     "live runs 8700/8722 green; T21.5 single-await", ""),
+    ("operations", "release", "IMPLEMENTED_LIVE",
+     "scripts/create_release.py; semver_bump.py; auto_bump_pr.py", "protected-branch bump flow",
+     "releases v0.82.0-0.82.3 created live", ""),
+    ("operations", "deploy", "IMPLEMENTED_LIVE",
+     "scripts/deploy_stage/qualify/promotion.py", "monotonic fail-closed promotion",
+     "live deploys 40b3731..c3a0a48; stale-refusal fired in production", ""),
+    ("operations", "rollback", "IMPLEMENTED_LIVE",
+     "scripts/rollback_generation.py", "governed regression API",
+     "T21.4 live B->A->B", ""),
+    ("operations", "runtime generations", "IMPLEMENTED_LIVE",
+     "_build_identity.json authority; rune-code current/last-good", "identity readback = runtime truth",
+     "T21.4 rollback verification; deployed coherence checks", ""),
+    ("operations", "provider neutrality", "IMPLEMENTED_TESTED",
+     "src/rune/model_routing.py; provider_registry.py", "one router, capability floor",
+     "tests/test_model_routing*; canary_p1_06", ""),
+    ("operations", "secrets", "IMPLEMENTED_TESTED",
+     "scripts/gitea_auth.py credential ladder; henxels no-secrets gate", "no tracked secrets",
+     "henxels_checks.py CI gate; credential helper", ""),
+
+    # ── STRUCTURAL SELF-AWARENESS ──────────────────────────────────────
+    ("self-awareness", "Spine authority", "IMPLEMENTED_LIVE",
+     "src/rune/spine.py", "Rune structural self-model",
+     "store/spine/current.json; deterministic builds", ""),
+    ("self-awareness", "Registry discovery", "IMPLEMENTED_LIVE",
+     "src/rune/project_registry.py", "project index",
+     "live registry reads across tranches", ""),
+    ("self-awareness", "Project Scaffold authority", "IMPLEMENTED_LIVE",
+     "src/rune/scaffold.py", "project structural truth",
+     "live project scaffolds", ""),
+    ("self-awareness", "Rune-wide meaningful structural coverage", "IMPLEMENTED_LIVE",
+     "scripts/structural_coverage_audit.py", "A/B/C/D/E classification",
+     "T21.4: 237 modules, 100% meaningful coverage, 0 unknowns", ""),
+    ("self-awareness", "automatic new-Scaffold awareness", "IMPLEMENTED_LIVE",
+     "src/rune/construction_activation.py _auto_structural_awareness", "bootstrap auto-registers",
+     "T21.5 live + tests/test_t215_awareness.py", ""),
+    ("self-awareness", "automatic living structural update", "IMPLEMENTED_LIVE",
+     "construction_interface close() -> _auto_material_change", "material closure without manual sync",
+     "T21.4/T21.5 live; tests/test_t214_material_closure.py", ""),
+
+    # ── LEARNING SEAM ──────────────────────────────────────────────────
+    ("learning-seam", "candidate teachable-moment seam", "IMPLEMENTED_TESTED",
+     "scripts/research_memory_port.py mark_teachable_moment_candidate", "CANDIDATE evidence via existing channels",
+     "tests/test_memory_port.py", ""),
+    ("learning-seam", "Memory internal-experience port in INWARD", "IMPLEMENTED_TESTED",
+     "scripts/research_memory_port.py; research_inward_context.build_context", "one bounded port of call",
+     "tests/test_memory_port.py", ""),
+    ("learning-seam", "APPLICABILITY governs transfer", "IMPLEMENTED_TESTED",
+     "research_memory_port.applicability_filter + existing APPLICABILITY stage", "irrelevant candidates rejected",
+     "tests/test_memory_port.py::test_applicability_rejects_irrelevant_candidate", ""),
+    ("learning-seam", "zero-result fast path", "IMPLEMENTED_TESTED",
+     "research_memory_port.memory_port_of_call", "[] -> next normal step",
+     "tests/test_memory_port.py; ZERO_RESULT_EXTRA_MODEL_CALLS=0", ""),
+    ("learning-seam", "no competing memory authority", "IMPLEMENTED_TESTED",
+     "port design (provider delegation, no storage)", "Memory stays single future authority",
+     "port creates no store/db; provider contract only", ""),
+
+    # ── DEFERRED (non-V1, with reasons) ────────────────────────────────
+    ("deferred", "automatic Growth Delta pattern learning", "DEFERRED_NON_V1",
+     "src/rune/growth_delta.py (existing, non-V1 scope)", "post-V1 learning loop",
+     "V1 excludes Memory/Substrate/Growth Delta by tranche mandate; the "
+     "T21.5 seam provides the harvest point without implementing the loop", ""),
+    ("deferred", "Memory V1 system itself", "DEFERRED_NON_V1",
+     "(future Memory T0 archaeology)", "dedicated post-V1 phase",
+     "explicitly excluded by T21.5 mandate §0; only the INWARD port is V1", ""),
+    ("deferred", "Substrate", "DEFERRED_NON_V1",
+     "(future)", "post-V1", "explicitly excluded by tranche mandate", ""),
+    ("deferred", "TUI redesign", "DEFERRED_NON_V1",
+     "(existing rune_tui.py adequate)", "explicitly excluded by tranche mandate",
+     "TUI work explicitly out of scope for V1 Scaffolding", ""),
+]
+
+
+def main() -> int:
+    manifest = {
+        "tranche": "T21.5",
+        "rows": [
+            {"family": f, "item": i, "classification": c, "module": m,
+             "authority": a, "evidence": e, "gap": g}
+            for f, i, c, m, a, e, g in ROWS
+        ],
+    }
+    blocking = [r for r in manifest["rows"]
+                if r["classification"] in ("PARTIAL", "ABSENT",
+                                           "IMPLEMENTED_UNPROVEN")]
+    counts: dict[str, int] = {}
+    for r in manifest["rows"]:
+        counts[r["classification"]] = counts.get(r["classification"], 0) + 1
+    manifest["counts"] = counts
+    manifest["blocking_items"] = blocking
+    out = ROOT / "store" / "scaffold" / "v1_parity_manifest.json"
+    out.write_text(json.dumps(manifest, indent=1))
+    print(json.dumps(counts, indent=1))
+    print("blocking mandatory-V1 items:", len(blocking))
+    for b in blocking:
+        print("  BLOCKS:", b["family"], "/", b["item"])
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,47 @@
+#!/usr/bin/env python3
+"""wr_root_reconcile.py — T21.5 §7: governed split-brain reconciliation.
+
+If canonical and legacy W/R roots both hold live event stores, this
+moves legacy-root events that are ABSENT from the canonical store into
+it (idempotent, content-keyed by filename), then reports what remains.
+Never deletes; the legacy store stays as a moved-marker record.
+"""
+from __future__ import annotations
+
+import json
+import shutil
+import sys
+from pathlib import Path
+
+_SCRIPT_DIR = Path(__file__).resolve().parent
+sys.path.insert(0, str(_SCRIPT_DIR))
+import wr_root_resolver as resolver  # noqa: E402
+
+
+def main() -> int:
+    canonical = resolver.CANONICAL_WR_ROOT
+    legacy = resolver.LEGACY_WR_ROOT
+    (canonical / "events").mkdir(parents=True, exist_ok=True)
+    moved = 0
+    skipped = 0
+    for src in sorted((legacy / "events").glob("*.json")):
+        dst = canonical / "events" / src.name
+        if dst.exists():
+            skipped += 1
+            continue
+        shutil.copy2(src, dst)
+        moved += 1
+    report = {
+        "canonical": str(canonical),
+        "legacy": str(legacy),
+        "events_moved_to_canonical": moved,
+        "already_present_skipped": skipped,
+        "note": "legacy store retained (never deleted); daemon and all "
+                "consumers now resolve via wr_root_resolver",
+    }
+    print(json.dumps(report, indent=2))
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -0,0 +1,100 @@
+"""wr_root_resolver.py — T21.5 §7: THE ONE wait/resume root authority.
+
+T21.4 live finding: the daemon defaulted WR_ROOT to its own install dir
+while production data lived at /mnt/workdrive/wr-bridge; a restart
+without WR_ROOT exported silently opened a second event store and the
+controller saw no events (the split-brain that forced manual event
+relocation). This module is the SINGLE resolver every consumer shares:
+
+  daemon / event_wait / await_operation tool / notify_event_bridge
+  (CI event writer) / local event writers / CLI / launcher / tests.
+
+Resolution order:
+  1. RUNE_WR_ROOT env (explicit override; tests use it)
+  2. canonical production root (exists?)
+  3. legacy home install root (exists?)
+  4. canonical production root (created on demand — first-run path)
+
+Split-brain policy: if BOTH canonical and legacy roots contain live
+event stores, resolution FAILS LOUDLY unless the env pins one — never
+silently operate two stores. Reconciliation is governed, not silent:
+run scripts/wr_root_reconcile.py.
+"""
+from __future__ import annotations
+
+import json
+import os
+import sys
+from pathlib import Path
+
+CANONICAL_WR_ROOT = Path("/mnt/workdrive/wr-bridge")
+LEGACY_WR_ROOT = Path.home() / ".local" / "share" / "mimo-wait-resume"
+ENV_VAR = "RUNE_WR_ROOT"
+
+SPLIT_BRAIN_ERROR = (
+    "WR ROOT SPLIT-BRAIN: both canonical ({canonical}) and legacy "
+    "({legacy}) contain live event stores. Never operate two stores "
+    "silently — pin {env}=<root> explicitly, or reconcile with "
+    "scripts/wr_root_reconcile.py."
+)
+
+
+def _has_live_store(root: Path) -> bool:
+    """A root counts as 'live' when its event store exists and holds
+    events (an empty dir is just an uninitialized install)."""
+    events = root / "events"
+    if not events.is_dir():
+        return False
+    return any(events.glob("*.json"))
+
+
+def resolve_root(*, require_single: bool = True) -> Path:
+    """Resolve THE wait/resume data root. Raises on split-brain when
+    require_single (default) and no explicit pin exists."""
+    env = os.environ.get(ENV_VAR)
+    if env:
+        return Path(env)
+    canonical_live = _has_live_store(CANONICAL_WR_ROOT)
+    legacy_live = _has_live_store(LEGACY_WR_ROOT)
+    if canonical_live and legacy_live and require_single:
+        raise RuntimeError(SPLIT_BRAIN_ERROR.format(
+            canonical=CANONICAL_WR_ROOT, legacy=LEGACY_WR_ROOT,
+            env=ENV_VAR))
+    if canonical_live or CANONICAL_WR_ROOT.is_dir():
+        return CANONICAL_WR_ROOT
+    if legacy_live or LEGACY_WR_ROOT.is_dir():
+        return LEGACY_WR_ROOT
+    return CANONICAL_WR_ROOT  # first-run: create canonical
+
+
+def daemon_launch_env() -> dict[str, str]:
+    """The env a launcher must pass the daemon: WR_ROOT resolved ONCE
+    from this authority (the daemon itself must not guess)."""
+    return {"WR_ROOT": str(resolve_root()),
+            ENV_VAR: str(resolve_root())}
+
+
+def split_brain_roots() -> list[Path]:
+    """Diagnostic: roots that currently hold live stores."""
+    return [r for r in (CANONICAL_WR_ROOT, LEGACY_WR_ROOT)
+            if _has_live_store(r)]
+
+
+def main() -> int:
+    try:
+        root = resolve_root()
+    except RuntimeError as exc:
+        print(str(exc), file=sys.stderr)
+        return 2
+    print(json.dumps({
+        "resolved_root": str(root),
+        "canonical": str(CANONICAL_WR_ROOT),
+        "legacy": str(LEGACY_WR_ROOT),
+        "split_brain_roots": [str(p) for p in split_brain_roots()],
+        "env_pinned": bool(os.environ.get(ENV_VAR)),
+    }))
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -352,16 +352,6 @@ def pre_commit_gate_entry(entry: dict) -> dict:
        "critics": critics,
        "reason": "all critics passed" if all_pass else "one or more critics failed"
    }
-    row = db.execute(
-        "SELECT id, text, importance, wash_count, rock_status, topic FROM entries WHERE id = ?",
-        (entry_id,)
-    ).fetchone()
-
-    if not row:
-        return {"verdict": "REJECT", "reason": "entry not found"}
-
-    entry = dict(row)
-    critics = {}

    # Critic 1: Structural validity
    text = entry.get("text", "")
@@ -384,35 +374,6 @@ def pre_commit_gate_entry(entry: dict) -> dict:
    }

    # Critic 3: Semantic consistency (check for contradictions with existing rocks)
-    rocks = db.execute(
-        "SELECT text FROM entries WHERE rock_status = 'rock' AND status = 'active' LIMIT 50"
-    ).fetchall()
-    contradiction_found = False
-    for rock in rocks:
-        rock_text = rock[0].lower()
-        entry_text = text.lower()
-        # Simple contradiction detection: if entry says opposite of a rock
-        if ("always" in rock_text and "never" in entry_text) or \
-           ("never" in rock_text and "always" in entry_text):
-            contradiction_found = True
-            break
-    critics["semantic"] = {
-        "pass": not contradiction_found,
-        "reason": "contradicts existing rock" if contradiction_found else "consistent"
-    }
-
-    # Overall verdict
-    all_pass = all(c["pass"] for c in critics.values())
-    verdict = "PASS" if all_pass else "REJECT"
-
-    return {
-        "verdict": verdict,
-        "entry_id": entry_id,
-        "critics": critics,
-        "reason": "all critics passed" if all_pass else "one or more critics failed"
-    }
-
-
 # ═══════════════════════════════════════════════════════════════════════════════
 # CLI
 # ═══════════════════════════════════════════════════════════════════════════════
@@ -6,7 +6,7 @@ This is the main entry point for the Rune program.
 import json
 from pathlib import Path

-__version__ = "0.66.31"
+__version__ = "0.86.0"

 # One build identity (handoff 2026-09-10 §4): the semantic version above is
 # the repo-level truth (auto-bumped at merge); the DEPLOYED revision (exact
@@ -0,0 +1,251 @@
+"""behaviour_ledger.py — T19 §19F/19G/19I: the canonical Behaviour
+Validation Ledger.
+
+PRIMARY UNIT: the VERSIONED BEHAVIOUR / TEST SPECIFICATION — not "a
+test run in some session". Automated verification and real-world/user
+confirmation are DISTINCT truths (rule 19F): tests green ≠ user-
+confirmed.
+
+GENERATION-AWARE HISTORY (rule 19G, mandatory): when a dependency
+changes, affected behaviours become RETEST_REQUIRED while their prior
+confirmation history stays queryable — confidence does not silently
+carry forward, and it does not silently disappear either.
+
+SELECTIVE INVALIDATION (rule 19I): only behaviours whose structural
+dependencies changed are invalidated; every invalidation carries a
+reason; unrelated hardened confirmations are preserved.
+
+Storage: <project_store>/behaviour_ledger.json (versioned document).
+"""
+from __future__ import annotations
+
+import time
+from pathlib import Path
+
+LEDGER_SCHEMA_VERSION = 1
+
+# hardening lifecycle (rule 19G semantics)
+UNVALIDATED = "UNVALIDATED"                    # not yet validated
+IMPLEMENTED = "IMPLEMENTED"                    # implementation exists
+AUTO_VERIFIED = "AUTO_VERIFIED"                # automated verification passed
+REAL_WORLD_PASSED = "REAL_WORLD_PASSED"        # manual/real test passed
+TEST_FAILED = "TEST_FAILED"                    # test failed
+PARTIAL = "PARTIAL"                            # inconclusive
+USER_CONFIRMED = "USER_CONFIRMED"              # hardened
+RETEST_REQUIRED = "RETEST_REQUIRED"            # confidence invalidated
+ACCEPTED_LIMITATION = "ACCEPTED_LIMITATION"    # explicitly approved gap
+NOT_APPLICABLE = "NOT_APPLICABLE"
+
+STATES = (UNVALIDATED, IMPLEMENTED, AUTO_VERIFIED, REAL_WORLD_PASSED,
+          TEST_FAILED, PARTIAL, USER_CONFIRMED, RETEST_REQUIRED,
+          ACCEPTED_LIMITATION, NOT_APPLICABLE)
+
+# states that represent STANDING CONFIDENCE (selective invalidation
+# may reopen exactly these; everything else is already open or n/a)
+CONFIRMED_STATES = (AUTO_VERIFIED, REAL_WORLD_PASSED, USER_CONFIRMED)
+
+
+def _now() -> str:
+    return time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+
+
+class BehaviourLedger:
+    def __init__(self, store_dir: Path | str):
+        self.path = Path(store_dir) / "behaviour_ledger.json"
+        self.doc = self._load()
+
+    def _load(self) -> dict:
+        if self.path.is_file():
+            import json
+            return json.loads(self.path.read_text())
+        return {"schema_version": LEDGER_SCHEMA_VERSION,
+                "behaviours": {}, "updated_at": _now()}
+
+    def save(self) -> None:
+        import json
+        self.doc["updated_at"] = _now()
+        self.path.parent.mkdir(parents=True, exist_ok=True)
+        self.path.write_text(json.dumps(self.doc, indent=1, sort_keys=True))
+
+    # -- specification ------------------------------------------------------
+
+    def define(self, *, behaviour_id: str, owning_project: str,
+               intended_behaviour: str, origin_intent: str = "",
+               pass_criteria: str = "", fail_criteria: str = "",
+               procedure: str = "",
+               owner_component: str = "",
+               dependencies: list[str] | None = None,
+               automated_ref: str = "",
+               environment: str = "") -> dict:
+        bid = (behaviour_id if behaviour_id.startswith("behaviour:")
+               else f"behaviour:{behaviour_id}")
+        if bid in self.doc["behaviours"]:
+            raise ValueError(f"behaviour exists: {bid}")
+        rec = {
+            "behaviour_id": bid,
+            "owning_project": owning_project,
+            "intended_behaviour": intended_behaviour,
+            "origin_intent": origin_intent,
+            "pass_criteria": pass_criteria,
+            "fail_criteria": fail_criteria,
+            "procedure": procedure,
+            "owner_component": owner_component,   # structural entity id
+            "dependencies": list(dependencies or []),  # structural ids
+            "automated_ref": automated_ref,
+            "environment": environment,
+            "state": UNVALIDATED,
+            "evidence": [],           # [{at, generation, kind, detail}]
+            "history": [],            # every state transition, never lost
+            "invalidations": [],      # [{at, generation, because}]
+            "limitations": [],
+            "created_at": _now(),
+            "updated_at": _now(),
+        }
+        self.doc["behaviours"][bid] = rec
+        return rec
+
+    def get(self, behaviour_id: str) -> dict | None:
+        return self.doc["behaviours"].get(
+            behaviour_id if behaviour_id.startswith("behaviour:")
+            else f"behaviour:{behaviour_id}")
+
+    # -- state transitions (all recorded in history — rule 19G) -------------
+
+    def _transition(self, rec: dict, new_state: str, *, kind: str,
+                    detail: str = "", generation: str = "") -> None:
+        if new_state not in STATES:
+            raise ValueError(f"unknown validation state {new_state!r}")
+        rec["history"].append({
+            "at": _now(), "from": rec["state"], "to": new_state,
+            "kind": kind, "detail": detail[:200],
+            "generation": generation})
+        rec["state"] = new_state
+        rec["updated_at"] = _now()
+
+    def record_auto_verification(self, behaviour_id: str, *,
+                                 passed: bool, detail: str = "",
+                                 generation: str = "") -> dict:
+        """Automated verification. GREEN NEVER IMPLIES USER_CONFIRMED
+        (rule 19W-22): the ceiling here is AUTO_VERIFIED."""
+        rec = self.get(behaviour_id)
+        if rec is None:
+            raise ValueError(f"unknown behaviour {behaviour_id!r}")
+        rec["evidence"].append({"at": _now(), "kind": "automated",
+                                "passed": passed,
+                                "detail": detail[:200],
+                                "generation": generation})
+        self._transition(rec, AUTO_VERIFIED if passed else TEST_FAILED,
+                         kind="automated_verification", detail=detail,
+                         generation=generation)
+        return rec
+
+    def record_real_world(self, behaviour_id: str, *, passed: bool,
+                          detail: str = "", generation: str = "",
+                          confirmed_by: str = "user") -> dict:
+        """Real-world/manual test result (testing mode). USER_CONFIRMED
+        requires an explicit confirmed_by='user' pass (rule 19M)."""
+        rec = self.get(behaviour_id)
+        if rec is None:
+            raise ValueError(f"unknown behaviour {behaviour_id!r}")
+        rec["evidence"].append({"at": _now(), "kind": "real_world",
+                                "passed": passed,
+                                "detail": detail[:200],
+                                "generation": generation,
+                                "by": confirmed_by})
+        if passed:
+            state = (USER_CONFIRMED if confirmed_by == "user"
+                     else REAL_WORLD_PASSED)
+        else:
+            state = TEST_FAILED
+        self._transition(rec, state, kind="real_world", detail=detail,
+                         generation=generation)
+        return rec
+
+    def accept_limitation(self, behaviour_id: str, *, note: str,
+                          approved_by: str = "user") -> dict:
+        rec = self.get(behaviour_id)
+        if rec is None:
+            raise ValueError(f"unknown behaviour {behaviour_id!r}")
+        rec["limitations"].append({"at": _now(), "note": note[:200],
+                                   "approved_by": approved_by})
+        self._transition(rec, ACCEPTED_LIMITATION, kind="accept_limitation",
+                         detail=note, )
+        return rec
+
+    # -- selective invalidation (rule 19I) -----------------------------------
+
+    def invalidate_for_change(self, *, changed_entities: set[str],
+                              generation: str = "",
+                              reasons: dict[str, str] | None = None
+                              ) -> list[dict]:
+        """Mark RETEST_REQUIRED exactly those confirmed behaviours whose
+        structural dependencies intersect the changed closure. Every
+        invalidation carries a reason; unrelated confirmations stay.
+        Returns the invalidated records."""
+        reasons = reasons or {}
+        invalidated = []
+        for rec in self.doc["behaviours"].values():
+            if rec["state"] not in CONFIRMED_STATES:
+                continue
+            overlap = set(rec.get("dependencies") or []) & changed_entities
+            if rec.get("owner_component") in changed_entities:
+                overlap.add(rec["owner_component"])
+            if not overlap:
+                continue  # unrelated hardened behaviour: PRESERVED (19S-5)
+            because = reasons.get(
+                rec["behaviour_id"],
+                f"structural change touched: {sorted(overlap)}")
+            rec["invalidations"].append({
+                "at": _now(), "generation": generation,
+                "because": because[:200],
+                "was": rec["state"],
+                "changed_entities": sorted(overlap)})
+            # HISTORY IS PRESERVED — prior confirmation stays queryable
+            self._transition(rec, RETEST_REQUIRED,
+                             kind="selective_invalidation",
+                             detail=f"was {rec['state']}; {because}",
+                             generation=generation)
+            invalidated.append(rec)
+        return invalidated
+
+    def retest_history(self, behaviour_id: str) -> list[dict]:
+        """Generation-aware history: all confirmations + invalidations
+        for one behaviour, oldest first (rule 19S-12)."""
+        rec = self.get(behaviour_id)
+        if rec is None:
+            return []
+        events = []
+        for h in rec["history"]:
+            events.append({"at": h["at"], "generation": h.get("generation"),
+                           "event": f"{h['from']}→{h['to']}",
+                           "kind": h["kind"], "detail": h.get("detail", "")})
+        return events
+
+    # -- queue (testing mode, 19L/M) ------------------------------------------
+
+    def outstanding(self, *, component: str | None = None,
+                    project: str | None = None) -> list[dict]:
+        """Eligible testing items: anything not confirmed/accepted/n.a.
+        Dynamic — the queue re-derives from states after every fix
+        (rule 19M)."""
+        out = []
+        for rec in self.doc["behaviours"].values():
+            if rec["state"] in (USER_CONFIRMED, ACCEPTED_LIMITATION,
+                                NOT_APPLICABLE):
+                continue
+            if project and rec.get("owning_project") != project:
+                continue
+            if component and rec.get("owner_component") != component:
+                continue
+            out.append(rec)
+        out.sort(key=lambda r: (0 if r["state"] == TEST_FAILED else 1,
+                                r["behaviour_id"]))
+        return out
+
+    def summary(self) -> dict:
+        counts: dict[str, int] = {}
+        for rec in self.doc["behaviours"].values():
+            counts[rec["state"]] = counts.get(rec["state"], 0) + 1
+        return {"total": len(self.doc["behaviours"]),
+                "states": counts,
+                "outstanding": len(self.outstanding())}
@@ -0,0 +1,314 @@
+"""build_basis.py — T18 §18C: durable versioned RESEARCH BASIS for
+construction (V1 rule F: RESEARCH INFORMS STRUCTURE).
+
+Research output must not disappear in transient model context. A Build
+Basis is a durable, schema-versioned document that captures WHAT
+research decided and WHY, with references — not research dumps. It is
+created after the research-depth decision and before/with the
+Construction Contract, and is attachable to:
+
+  - a Project Scaffold (project-level basis)
+  - a Construction Contract (per-contract basis, via evidence_ref)
+  - future structural entities where useful
+
+Research Core ownership (rule C): this module NEVER researches. It
+records the depth decision (derived FROM adaptive_router's route — no
+second router), references the canonical pipeline's qualified output
+(stages ISSUE→…→QUALIFY, artifacts in research/issue-<N>/), and
+structures the adopt/adapt/combine/reject/innovate decisions (rule E).
+"""
+from __future__ import annotations
+
+import time
+from typing import Any
+
+SCHEMA_VERSION = 1
+
+# Research depth tiers — derived from adaptive_router ROUTES, never a
+# second router (rule C/A). Each tier records WHY it was chosen.
+DEPTH_NONE = "NONE"                # tiny obvious local change; local evidence only
+DEPTH_INWARD = "INWARD"            # repo-internal understanding suffices
+DEPTH_OUTWARD = "OUTWARD"          # precedent/prior-art discovery required
+DEPTH_FULL = "FULL"                # full pipeline incl. adversarial+synthesis
+
+DEPTH_TIERS = (DEPTH_NONE, DEPTH_INWARD, DEPTH_OUTWARD, DEPTH_FULL)
+
+# route → default depth (recorded as decision evidence; caller may
+# justify an exception, which is recorded as such — never silent)
+_ROUTE_DEFAULT_DEPTH = {
+    "LOCAL_FIX": DEPTH_NONE,
+    "STANDARD_CHANGE": DEPTH_INWARD,
+    "RESEARCHED_CHANGE": DEPTH_FULL,
+}
+
+
+def research_depth(route: str, *, mode: str = "",
+                   justification: str = "") -> dict:
+    """The research-depth DECISION for a construction need. Derived from
+    the canonical route authority (adaptive_router); `justification` is
+    REQUIRED when deviating from the route default — recorded, never
+    silent (rule D: record why research depth was chosen)."""
+    default = _ROUTE_DEFAULT_DEPTH.get(route, DEPTH_INWARD)
+    depth = default
+    why = f"route {route} default"
+    if justification and justification.strip():
+        # an explicit justification selects a tier by name if given;
+        # otherwise it annotates the default
+        candidate = justification.strip().split(":", 1)[0].strip()
+        if candidate in DEPTH_TIERS:
+            depth = candidate
+            why = justification.strip()
+        else:
+            why = f"route {route} default; note: {justification.strip()}"
+    elif route not in _ROUTE_DEFAULT_DEPTH:
+        why = f"route {route!r} unknown — INWARD floor (never silent NONE)"
+        depth = DEPTH_INWARD
+    return {"depth": depth, "why": why, "route": route, "mode": mode,
+            "decided_at": _now()}
+
+
+# OUTWARD precedent-first ordering contract (rule E): the questions the
+# prior-art pass MUST answer before deeper material is consulted.
+OUTWARD_PRIOR_ART_QUESTIONS = (
+    "what already exists addressing this intent",
+    "what open-source systems/products/workflows attempt it",
+    "what architectural patterns already work",
+    "what standards/conventions exist",
+    "what can be directly ADOPTED",
+    "what can be ADAPTED",
+    "what can be COMBINED",
+    "what should be REJECTED",
+    "where is genuine INNOVATION required",
+)
+
+_DISPOSITIONS = ("ADOPT", "ADAPT", "COMBINE", "REJECT", "INNOVATE")
+
+
+def new_basis(*, question: str, depth_decision: dict,
+              issue_ref: str = "") -> dict:
+    """Create an empty-but-valid Build Basis. References only — full
+    research dumps live in the Research Core's artifact dirs (rule 18C:
+    do not store unnecessary dumps)."""
+    return {
+        "schema_version": SCHEMA_VERSION,
+        "kind": "build_basis",
+        "created_at": _now(),
+        "question": question,
+        "issue_ref": issue_ref,  # e.g. research/issue-164 or Gitea issue
+        "depth": depth_decision,
+        "inward": {"findings": [], "refs": []},
+        "outward": {"prior_art_questions_answered": [],
+                    "findings": [], "refs": []},
+        "deeper": {"used": False, "findings": [], "refs": []},
+        "dispositions": [],  # [{item, disposition: ADOPT/…, why, ref}]
+        "alternatives_considered": [],
+        "uncertainty": [],
+        "risks": [],
+        "implications": {"architecture": [], "structural": [],
+                         "migration": [], "verification": []},
+    }
+
+
+def set_dispositions(basis: dict, dispositions: list[dict]) -> list[str]:
+    """Record the adopt/adapt/combine/reject/innovate decisions (rule E).
+    Returns validation errors; the basis stays unqualified until valid."""
+    basis["dispositions"] = [
+        {"item": d.get("item", ""), "disposition": d.get("disposition", ""),
+         "why": d.get("why", ""), "ref": d.get("ref", "")}
+        for d in dispositions
+    ]
+    return validate_basis(basis)
+
+
+def validate_basis(basis: dict) -> list[str]:
+    """Qualification gate. A basis is QUALIFIED when it carries a depth
+    decision with why, at least one outward prior-art answer when depth
+    is OUTWARD/FULL, and every disposition uses a legal disposition."""
+    errors: list[str] = []
+    if basis.get("schema_version") != SCHEMA_VERSION:
+        errors.append(f"schema_version {basis.get('schema_version')!r}")
+    depth = (basis.get("depth") or {}).get("depth")
+    if depth not in DEPTH_TIERS:
+        errors.append(f"depth {depth!r} not in {DEPTH_TIERS}")
+    if not (basis.get("depth") or {}).get("why"):
+        errors.append("depth decision lacks WHY (rule D)")
+    if depth in (DEPTH_OUTWARD, DEPTH_FULL):
+        outward = basis.get("outward") or {}
+        if not (outward.get("prior_art_questions_answered")
+                or outward.get("findings") or outward.get("refs")):
+            errors.append("OUTWARD/FULL depth requires prior-art evidence")
+    if depth == DEPTH_INWARD:
+        inward = basis.get("inward") or {}
+        if not (inward.get("findings") or inward.get("refs")
+                or basis.get("dispositions")):
+            errors.append("INWARD depth requires recorded evidence "
+                          "(findings/refs) — an empty basis is not "
+                          "qualified")
+    for d in basis.get("dispositions", []):
+        if d.get("disposition") not in _DISPOSITIONS:
+            errors.append(f"illegal disposition {d.get('disposition')!r} "
+                          f"for {d.get('item','')[:40]!r}")
+    return errors
+
+
+def is_qualified(basis: dict) -> bool:
+    return not validate_basis(basis)
+
+
+def attach_evidence_ref(basis: dict) -> str:
+    """The evidence_ref string a contract carries to point at this basis
+    (durable pointer, not a copy — stores stay independent)."""
+    return (f"build_basis:{basis.get('issue_ref') or 'adhoc'}:"
+            f"{basis.get('created_at')}")
+
+
+def summary(basis: dict) -> dict:
+    """Compact projection for prompts/previews (no private reasoning —
+    structured conclusions only, rule 18G)."""
+    return {
+        "schema_version": basis.get("schema_version"),
+        "question": basis.get("question"),
+        "issue_ref": basis.get("issue_ref"),
+        "depth": basis.get("depth"),
+        "qualified": is_qualified(basis),
+        "validation": validate_basis(basis),
+        "dispositions": [
+            {"item": d.get("item", "")[:80],
+             "disposition": d.get("disposition")}
+            for d in basis.get("dispositions", [])
+        ],
+        "uncertainty_count": len(basis.get("uncertainty", [])),
+        "risks_count": len(basis.get("risks", [])),
+        "implications": {k: len(v) for k, v in
+                         (basis.get("implications") or {}).items()},
+    }
+
+
+def _now() -> str:
+    return time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+
+
+def revise_depth(basis: dict, *, new_depth: str, reason: str) -> dict:
+    """§21.1G: record a justified depth revision. Original + revised
+    depths both persist. An OUTWARD→INWARD/NONE downgrade requires
+    explicit resolution evidence ('discovery resolved', 'prior-art
+    resolved', 'unnecessary') — never a bare skip."""
+    if new_depth not in DEPTH_TIERS:
+        raise ValueError(f"unknown depth {new_depth!r}")
+    original = basis["depth"]["depth"]
+    if original == new_depth:
+        raise ValueError("no-op revision: depth unchanged")
+    consequential_origins = ("RESEARCHED_CHANGE", "FULL", "OUTWARD")
+    if original in consequential_origins and new_depth in ("NONE", "INWARD"):
+        keywords = ("resolved", "non-material", "discovery",
+                    "unnecessary", "prior-art question")
+        if not any(k in reason.lower() for k in keywords):
+            raise ValueError(
+                "OUTWARD downgrade requires explicit evidence: the "
+                "prior-art question must be resolved as non-material "
+                "(not merely skipped)")
+    basis["depth_revision"] = {
+        "original_depth": original, "revised_depth": new_depth,
+        "reason": reason[:200], "revised_at": _now()}
+    basis["depth"]["depth"] = new_depth
+    basis["depth"]["why"] = f"revised from {original}: {reason}"
+    return basis["depth_revision"]
+
+
+def to_json(basis: dict) -> str:
+    import json
+    return json.dumps(basis, indent=1, sort_keys=True)
+
+
+def from_json(text: str) -> dict:
+    import json
+    return json.loads(text)
+
+
+_ = Any  # typing import kept for forward annotations
+
+
+# ── T18 §18D CORRECTION (Nick, 2026-09-15): THE RESEARCH PLAN ITSELF IS
+# CONTEXTUALLY GENERATED. The prior-art categories researched for the
+# structural-model task (architecture graphs, monorepo tooling, etc.)
+# were CONTEXTUAL to that task — they must never become a fixed topic
+# list. For every request the plan is derived from:
+#   intent → target/current reality → material unknowns/decisions/risks
+#   → context-specific research topics → proportional depth.
+# There is NO permanent research-topic checklist anywhere in Scaffolding.
+
+def generate_research_plan(*, intent: str, target: str = "",
+                           current_reality: str = "",
+                           unknowns: list[str] | None = None,
+                           decisions: list[str] | None = None,
+                           risks: list[str] | None = None,
+                           depth: str | None = None) -> dict:
+    """Generate a CONTEXT-SPECIFIC research plan for one construction
+    intent. Topics are synthesized from the material unknowns/decisions/
+    risks of THIS intent — the function has no topic vocabulary and no
+    permanent category list. Depth, when not supplied, is chosen by
+    consequence heuristics recorded in the plan (why is always present)."""
+    unknowns = [u for u in (unknowns or []) if u.strip()]
+    decisions = [d for d in (decisions or []) if d.strip()]
+    risks = [r for r in (risks or []) if r.strip()]
+    topics: list[dict] = []
+    # each material unknown becomes a research topic aimed at resolving it
+    for u in unknowns:
+        topics.append({
+            "topic": f"resolve unknown: {u}",
+            "kind": "unknown-resolution",
+            "why": f"material unknown blocks confident construction: {u[:80]}",
+        })
+    for d in decisions:
+        topics.append({
+            "topic": f"inform decision: {d}",
+            "kind": "decision-support",
+            "why": f"consequential decision needs precedent/evidence: {d[:80]}",
+        })
+    for r in risks:
+        topics.append({
+            "topic": f"de-risk: {r}",
+            "kind": "risk-investigation",
+            "why": f"identified risk needs verification/mitigation evidence: {r[:80]}",
+        })
+    # precedent question (rule E) attaches ONLY when outward evidence is
+    # plausibly relevant — an unknown/decision about external prior art.
+    needs_outward = any(
+        w in " ".join(unknowns + decisions).lower()
+        for w in ("exist", "existing", "prior art", "precedent", "standard",
+                  "convention", "library", "tool", "framework", "approach"))
+    if needs_outward:
+        topics.append({
+            "topic": ("what already exists that is relevant to THIS intent "
+                      "(adopt/adapt/combine/reject/innovate)"),
+            "kind": "prior-art",
+            "why": "material unknown references existing practice — "
+                   "precedent discovery required before deeper research",
+        })
+    if depth is None:
+        # consequence heuristics: novel/architectural/unsafe → deeper
+        consequential = any(
+            w in (intent + " " + " ".join(risks)).lower()
+            for w in ("architecture", "migration", "replace", "rewrite",
+                      "security", "schema", "delete", "deploy", "breaking"))
+        depth = DEPTH_FULL if (consequential and topics) else (
+            DEPTH_OUTWARD if needs_outward and topics else (
+                DEPTH_INWARD if topics else DEPTH_NONE))
+        why = ("consequence heuristics: consequential keywords + research "
+               "topics present" if depth == DEPTH_FULL else
+               "precedent-relevant unknowns present" if depth == DEPTH_OUTWARD
+               else "local understanding suffices" if depth == DEPTH_INWARD
+               else "no material unknowns — local evidence only")
+    else:
+        why = "depth supplied by caller (recorded, caller accountable)"
+    return {
+        "intent": intent,
+        "target": target,
+        "current_reality_note": current_reality[:200],
+        "depth": {"depth": depth, "why": why},
+        "topics": topics,
+        "plan_generated_at": _now(),
+        # the plan ITSELF is evidence that research was contextual, not
+        # checklist-driven: two different intents must produce disjoint
+        # topic sets unless they share genuine unknowns
+    }
@@ -231,7 +231,7 @@ def main() -> int:
    )
    check(
        "Rune minor version advertises provisioning and restore verification",
-        '__version__ = "0.66.31"' in rune_version,
+        '__version__ = "0.86.0"' in rune_version,
        failures,
    )
    # #169 policy update: auto-bump on merge is sanctioned (Nick directive:
@@ -93,6 +93,31 @@ def main() -> int:
        "if: always()" in required_gate and "needs: [lint, unit-tests, canary" in required_gate,
        failures,
    )
+    # RELEASE-SAFETY (2026-09-13, after 0.66.28/29 red promotions): the
+    # required-checks publish step previously echoed success UNCONDITIONALLY
+    # (read /tmp/need_* files that no step wrote). The verdict MUST come
+    # from the needs context — this canary locks that class shut.
+    publish_section = required_gate.split("Publish protected-branch gate", 1)[-1]
+    check(
+        "required gate verdict evaluates needs results (not unconditional echo)",
+        "needs.lint.result == 'success'" in publish_section
+        and "needs['type-check'].result == 'success'" in publish_section
+        and "needs['unit-tests'].result == 'success'" in publish_section,
+        failures,
+    )
+    check(
+        "required gate fails closed when any check is not success",
+        "!= 'success'" in publish_section and "exit 1" in publish_section,
+        failures,
+    )
+    check(
+        "no unconditional success echo without a needs-conditioned if",
+        not (
+            "All required CI checks passed" in publish_section
+            and "needs.lint.result == 'success'" not in publish_section
+        ),
+        failures,
+    )
    check(
        "required gate depends on unit tests and verification jobs",
        "needs: [lint, unit-tests, canary" in text,
@@ -0,0 +1,429 @@
+"""capability_fit.py — Requirement-fit capability discovery + gap decision.
+
+SCAFFOLD-TRANCHE-003. Layer over the Tranche 2 substrate (structural_store,
+scaffold, spine). NOT a second graph: discovery queries existing structural
+authorities (Rune Spine / Project Scaffold stores) and produces candidates;
+fit and resolution are derived verdicts with recorded evidence.
+
+Architecture:
+
+    CapabilityRequirement (open-world object)
+        ↓
+    discover_candidates(store_doc)          DETERMINISTIC — structural only
+        ↓
+    evaluate_fit(requirement, candidate)    DETERMINISTIC first;
+                                            bounded LLM seam recorded, never
+                                            authoritative alone
+        ↓
+    decide(requirement, fits)               resolution surface:
+                                            SATISFIED / REUSE / CONFIGURE /
+                                            COMPOSE / ADAPT_EXISTING /
+                                            ACQUIRE_EXTERNAL / BUILD_NEW /
+                                            INSUFFICIENT_EVIDENCE
+
+Core invariants (tranche brief §5, §27):
+    - BUILD_NEW is NEVER the default and is never silently chosen from
+      insufficient evidence.
+    - Structural discovery produces CANDIDATES; names alone are never fit.
+    - A stale/superseded/degraded candidate is not silently reusable.
+    - BUILD BUT DON'T BREAK: adaptation selected only when existing consumers
+      are provably unaffected; brittle coupling forces BUILD_NEW (with
+      rationale) or COMPOSE.
+    - When an LLM contributes interpretation, provider/model/reason/evidence/
+      confidence are RECORDED; its prose alone never establishes authority.
+    - Memory is never a structural authority here.
+"""
+
+from __future__ import annotations
+
+from rune.structural_store import entity, relationships, resolve_status
+
+__all__ = [
+    "CapabilityRequirement",
+    "RESOLUTIONS",
+    "discover_candidates",
+    "evaluate_fit",
+    "decide",
+    "record_semantic_fit",
+]
+
+RESOLUTIONS = (
+    "SATISFIED",
+    "REUSE",
+    "CONFIGURE",
+    "COMPOSE",
+    "ADAPT_EXISTING",
+    "ACQUIRE_EXTERNAL",
+    "BUILD_NEW",
+    "INSUFFICIENT_EVIDENCE",
+)
+
+# relationship types treated as "candidate implements/exposes this contract"
+_PROVIDES = {"implements", "exposes", "provides", "serves"}
+_CONSUMES = {"consumes", "depends_on"}
+# entity kinds eligible as capability candidates (open-world: unknown kinds
+# with providing relationships are also eligible — see discover_candidates)
+_BASE_KINDS = {"capability", "module", "component", "service", "cli", "tool", "workflow"}
+# maturity rank for reuse-worthiness (shared ladder from structural_store)
+_MATURITY_ORDER = ["DECLARED", "PRESENT", "WIRED", "REACHABLE", "EXERCISED", "LIVE_PROVEN"]
+_MATURITY_RANK = {n: i for i, n in enumerate(_MATURITY_ORDER)}
+
+
+# ── Capability Requirement ────────────────────────────────────────────────
+
+
+class CapabilityRequirement:
+    """Open-world structured requirement. No fixed ontology: `props` carries
+    any constraint the need expresses; a small set of FIRST-CLASS keys are
+    interpreted deterministically (contracts, environment, mandatory) and
+    everything else roundtrips as data."""
+
+    FIRST_CLASS = (
+        "objective",
+        "contracts",
+        "environment",
+        "mandatory",
+        "optional",
+        "target_project",
+        "inputs",
+        "outputs",
+        "performance",
+        "reliability",
+        "deployment",
+        "integration",
+        "user_constraints",
+        "evidence_refs",
+    )
+
+    def __init__(
+        self,
+        objective: str,
+        *,
+        contracts: list[str] | None = None,
+        environment: dict | None = None,
+        mandatory: list[str] | None = None,
+        optional: list[str] | None = None,
+        **props,
+    ):
+        self.objective = objective
+        self.contracts = sorted(contracts or [])  # required interfaces
+        self.environment = dict(environment or {})  # e.g. {"os": "linux"}
+        self.mandatory = sorted(mandatory or [])
+        self.optional = sorted(optional or [])
+        self.props = dict(props)  # open-world
+
+    def to_dict(self) -> dict:
+        d = {
+            "objective": self.objective,
+            "contracts": self.contracts,
+            "environment": self.environment,
+            "mandatory": self.mandatory,
+            "optional": self.optional,
+        }
+        d.update(self.props)
+        return d
+
+    @classmethod
+    def from_dict(cls, d: dict) -> "CapabilityRequirement":
+        known = {k: d.pop(k) for k in cls.FIRST_CLASS if k in d}
+        return cls(known.pop("objective"), **{**known, **d})
+
+
+# ── semantic-fit seam (recorded, never authoritative alone) ───────────────
+
+
+def record_semantic_fit(
+    provider: str, model: str, reason: str, evidence: list[str], confidence: float, interpretation: str
+) -> dict:
+    """Record a bounded LLM semantic interpretation contributing to fit.
+    The record is EVIDENCE, not authority: decide() treats it as supporting
+    context that must agree with deterministic facts. Never silent."""
+    if not provider or not model:
+        raise ValueError("semantic fit record requires provider+model")
+    if not 0.0 <= confidence <= 1.0:
+        raise ValueError("confidence must be in [0,1]")
+    return {
+        "kind": "semantic_fit",
+        "provider": provider,
+        "model": model,
+        "reason": reason,
+        "evidence": sorted(evidence),
+        "confidence": confidence,
+        "interpretation": interpretation[:500],
+    }
+
+
+# ── structural discovery (deterministic) ──────────────────────────────────
+
+
+def _provides_contracts(doc: dict, eid: str) -> set[str]:
+    out = set()
+    for r in relationships(doc, from_id=eid):
+        if r["type"] in _PROVIDES:
+            out.add(r["to"])
+    return out
+
+
+def _consumers(doc: dict, eid: str) -> set[str]:
+    out = set()
+    for r in relationships(doc, to_id=eid):
+        if r["type"] in {"consumes", "consumed_by", "depends_on"}:
+            out.add(r["from"])
+    return out
+
+
+def discover_candidates(doc: dict, req: CapabilityRequirement) -> list[dict]:
+    """Deterministic candidate discovery over ONE structural store doc.
+    A candidate is any entity that (a) is a known capability-ish kind, OR
+    (b) has providing relationships (open-world kinds stay discoverable).
+    Returns candidates with their structural evidence attached. Names are
+    NOT fit — every candidate must still pass evaluate_fit."""
+    candidates = []
+    for e in doc["entities"]:
+        eid = e["id"]
+        provides = _provides_contracts(doc, eid)
+        eligible = e["kind"] in _BASE_KINDS or bool(provides)
+        if not eligible:
+            continue
+        # environment filter: candidate must not CONTRADICT a stated
+        # environment constraint (missing info is not contradiction)
+        env = e.get("identities", {}).get("environment", {})
+        if isinstance(env, dict) and req.environment:
+            contradicts = any(k in env and req.environment.get(k) not in (None, env[k]) for k in req.environment)
+            if contradicts:
+                continue
+        candidates.append(
+            {
+                "id": eid,
+                "kind": e["kind"],
+                "identities": e.get("identities", {}),
+                "maturity": e.get("maturity", "DECLARED"),
+                "status": resolve_status(e),
+                "superseded_by": e.get("superseded_by"),
+                "provides": sorted(provides),
+                "consumers": sorted(_consumers(doc, eid)),
+                "evidence_refs": e.get("evidence_refs", []),
+            }
+        )
+    return candidates
+
+
+# ── fit evaluation ─────────────────────────────────────────────────────────
+
+
+def evaluate_fit(req: CapabilityRequirement, candidate: dict) -> dict:
+    """DETERMINISTIC fit of one candidate against a requirement.
+    fit: EXACT | PARTIAL | INCOMPATIBLE | UNKNOWN.
+    Names are ignored; contracts + environment + health decide."""
+    missing_contracts = sorted(set(req.contracts) - set(candidate["provides"]))
+    covered = len(req.contracts) - len(missing_contracts)
+
+    # health: a superseded or stale candidate cannot be reused as-is
+    status = candidate["status"]
+    unhealthy = status in {"SUPERSEDED", "STALE"} or bool(candidate.get("superseded_by"))
+
+    # maturity: DECLARED-only candidates have no usage evidence
+    maturity = _MATURITY_RANK.get(candidate.get("maturity", "DECLARED"), 0)
+
+    if unhealthy:
+        return {
+            "candidate": candidate["id"],
+            "fit": "INCOMPATIBLE",
+            "reason": f"candidate status {status} — not reusable as-is",
+            "missing_contracts": missing_contracts,
+            "covered": covered,
+        }
+    if req.contracts and not candidate["provides"]:
+        # candidate provides nothing verifiable → cannot establish fit
+        return {
+            "candidate": candidate["id"],
+            "fit": "UNKNOWN",
+            "reason": "candidate exposes no contracts to check",
+            "missing_contracts": missing_contracts,
+            "covered": 0,
+        }
+    if not missing_contracts:
+        fit = "EXACT"
+        reason = "all required contracts provided"
+    elif covered > 0:
+        fit = "PARTIAL"
+        reason = f"covers {covered}/{len(req.contracts)} required contracts"
+    else:
+        fit = "INCOMPATIBLE"
+        reason = "provides none of the required contracts"
+    return {
+        "candidate": candidate["id"],
+        "fit": fit,
+        "reason": reason,
+        "missing_contracts": missing_contracts,
+        "covered": covered,
+        "maturity_rank": maturity,
+        "consumers": candidate["consumers"],
+    }
+
+
+# ── gap / resolution decision ──────────────────────────────────────────────
+
+
+def _adaptation_brittle(doc: dict, candidate_id: str, missing_contracts: list[str]) -> tuple[bool, str]:
+    """BUILD BUT DON'T BREAK: adaptation is brittle when the contracts the
+    candidate is missing are load-bearing elsewhere (other entities consume
+    or provide them), i.e. adapting means touching a shared surface."""
+    for c in missing_contracts:
+        ce = entity(doc, c)
+        if ce is not None:
+            consumers = _consumers(doc, c)
+            consumers.discard(candidate_id)
+            if consumers:
+                return True, (f"missing contract {c} is load-bearing (consumed by {sorted(consumers)})")
+        for r in relationships(doc, to_id=c):
+            if r["type"] in _PROVIDES and r["from"] != candidate_id:
+                return True, (f"missing contract {c} is already provided by {r['from']} — adapting would fork it")
+    return False, ""
+
+
+def decide(
+    doc: dict,
+    req: CapabilityRequirement,
+    fits: list[dict],
+    *,
+    external_evidence: list[dict] | None = None,
+    semantic_fits: list[dict] | None = None,
+) -> dict:
+    """Resolution decision from deterministic fits. Evidence-backed:
+    every verdict carries the fits/rationale that produced it."""
+    external_evidence = list(external_evidence or [])
+    semantic_fits = list(semantic_fits or [])
+    exact = [f for f in fits if f["fit"] == "EXACT"]
+    partial = [f for f in fits if f["fit"] == "PARTIAL"]
+    unknown = [f for f in fits if f["fit"] == "UNKNOWN"]
+    incompatible = [f for f in fits if f["fit"] == "INCOMPATIBLE"]
+
+    def _result(resolution, rationale, **kw):
+        base = {
+            "resolution": resolution,
+            "rationale": rationale,
+            "considered": [f["candidate"] for f in fits],
+            "semantic_fits": semantic_fits,
+        }
+        base.update(kw)
+        assert resolution in RESOLUTIONS
+        return base
+
+    # INSUFFICIENT EVIDENCE dominates when fit cannot be established
+    if unknown and not exact and not partial:
+        return _result(
+            "INSUFFICIENT_EVIDENCE",
+            "candidates exist but expose no checkable contracts; no deterministic or recorded semantic evidence",
+            unknown_candidates=[f["candidate"] for f in unknown],
+        )
+    if not fits:
+        # nothing structural at all: external evidence decides the route
+        if external_evidence:
+            return _result(
+                "ACQUIRE_EXTERNAL",
+                "no structural candidates; external evidence suggests acquisition investigation",
+                external=external_evidence,
+            )
+        return _result(
+            "INSUFFICIENT_EVIDENCE",
+            "no candidates discovered and no external evidence — BUILD_NEW not silently chosen (§27 J)",
+        )
+
+    # EXACT: already satisfied vs reuse
+    if exact:
+        best = max(exact, key=lambda f: f.get("maturity_rank", 0))
+        if req.props.get("already_consumed_by") and req.props["already_consumed_by"] in best.get("consumers", []):
+            return _result(
+                "SATISFIED",
+                f"{best['candidate']} exactly matches and the requirement's context already consumes it",
+                selected=best["candidate"],
+                fits=exact,
+            )
+        return _result(
+            "REUSE",
+            f"{best['candidate']} exactly matches all required contracts with health evidence",
+            selected=best["candidate"],
+            fits=exact,
+        )
+
+    # PARTIAL: compose, adapt (clean), or build
+    if partial:
+        # COMPOSITION: union of ≥2 partials covers the requirement
+        if len(partial) >= 2:
+            union: set[str] = set()
+            chosen: list[str] = []
+            for f in sorted(partial, key=lambda x: -x["covered"]):
+                new = set(req.contracts) - union
+                if new & set(req.contracts) and not (set(f["missing_contracts"]) >= set(req.contracts)):
+                    add = [c for c in req.contracts if c not in union and c not in f["missing_contracts"]]
+                    if add:
+                        chosen.append(f["candidate"])
+                        union.update(add)
+                if union >= set(req.contracts):
+                    break
+            if union >= set(req.contracts):
+                return _result(
+                    "COMPOSE",
+                    f"composition {chosen} jointly covers all required contracts; none covers alone",
+                    components=chosen,
+                    permanence="temporary (user-controlled)",
+                )
+        # ADAPT: best partial, if adaptation is clean
+        best = max(partial, key=lambda f: f["covered"])
+        brittle, why = _adaptation_brittle(doc, best["candidate"], best["missing_contracts"])
+        if not brittle:
+            return _result(
+                "ADAPT_EXISTING",
+                f"{best['candidate']} covers {best['covered']}/"
+                f"{len(req.contracts)}; adaptation clean ({best['reason']})",
+                selected=best["candidate"],
+                missing=best["missing_contracts"],
+                obligations={"new_behavior_proof": True, "regression_proof": True},
+            )
+        if external_evidence:
+            return _result(
+                "ACQUIRE_EXTERNAL",
+                f"best partial ({best['candidate']}) cannot be "
+                f"adapted without brittle coupling ({why}); "
+                f"external evidence available",
+                external=external_evidence,
+            )
+        return _result(
+            "BUILD_NEW",
+            f"best partial ({best['candidate']}) would create brittle "
+            f"coupling: {why}; composition insufficient; building is the "
+            f"safer path — REUSE-WHEN-CLEAN, BUILD-WHEN-FORCED-REUSE-"
+            f"WOULD-BRITTLE (§5)",
+            selected=None,
+            rejected_partial=best["candidate"],
+            brittle=why,
+            obligations={"new_behavior_proof": True},
+        )
+
+    # everything incompatible
+    if external_evidence:
+        return _result(
+            "ACQUIRE_EXTERNAL", "all candidates incompatible; external evidence available", external=external_evidence
+        )
+    if incompatible:
+        return _result(
+            "INSUFFICIENT_EVIDENCE",
+            "all candidates incompatible with requirement "
+            "constraints; affirmative gap evidence required "
+            "before BUILD_NEW",
+            incompatible=[f["candidate"] for f in incompatible],
+        )
+    return _result("INSUFFICIENT_EVIDENCE", "no usable fit evidence")
+
+
+def resolve(
+    doc: dict,
+    req: CapabilityRequirement,
+    *,
+    external_evidence: list[dict] | None = None,
+    semantic_fits: list[dict] | None = None,
+) -> dict:
+    """One-call surface: discover → fit → decide."""
+    fits = [evaluate_fit(req, c) for c in discover_candidates(doc, req)]
+    return decide(doc, req, fits, external_evidence=external_evidence, semantic_fits=semantic_fits)
@@ -0,0 +1,405 @@
+"""capability_ledger.py — T19 §19D: compact versioned Capability
+Ledgers — local re-entry context packets.
+
+PURPOSE (rule 19D): a fresh LLM rehydrates the meaning of a target area
+after compaction/context loss/model swap/long absence WITHOUT rereading
+the repository. Not a second source tree, not a second Scaffold, not
+prose dumps — structured, proportional, operationally-used records.
+
+Identity rule: capability IDs are STABLE across file moves (name-based,
+not path-based); history is preserved on supersede/retire (rule 19Q).
+
+Storage: <project_store>/capability_ledger.json — versioned document,
+provenance-aware (rule G). Historical versions remain traceable via
+the history list.
+"""
+from __future__ import annotations
+
+import time
+from pathlib import Path
+
+LEDGER_SCHEMA_VERSION = 1
+
+MATURITY = ("proposed", "implemented", "verified", "hardened",
+            "retest_required", "retired", "superseded")
+
+
+def _now() -> str:
+    return time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+
+
+class CapabilityLedger:
+    def __init__(self, store_dir: Path | str):
+        self.path = Path(store_dir) / "capability_ledger.json"
+        self.doc = self._load()
+
+    def _load(self) -> dict:
+        if self.path.is_file():
+            import json
+            return json.loads(self.path.read_text())
+        return {"schema_version": LEDGER_SCHEMA_VERSION,
+                "capabilities": {}, "history": [],
+                "updated_at": _now()}
+
+    def save(self) -> None:
+        import json
+        self.doc["updated_at"] = _now()
+        self.path.parent.mkdir(parents=True, exist_ok=True)
+        self.path.write_text(json.dumps(self.doc, indent=1, sort_keys=True))
+
+    # -- lifecycle --------------------------------------------------------
+
+    def define(self, *, capability_id: str, name: str, purpose: str,
+               owning_project: str, component: str,
+               source_locations: list[str] | None = None,
+               contract: str = "", state_touched: str = "",
+               dependencies: list[str] | None = None,
+               consumers: list[str] | None = None,
+               invariants: list[str] | None = None,
+               behaviour_specs: list[str] | None = None,
+               constraints: str = "",
+               generation: str = "") -> dict:
+        """Create a capability record (auto-invoked by Scaffolding when a
+        meaningful capability is created — rule 19Q)."""
+        cid = f"capability:{capability_id}"
+        if cid in self.doc["capabilities"]:
+            raise ValueError(f"capability exists: {cid}")
+        rec = {
+            "capability_id": cid,
+            "name": name,
+            "purpose": purpose,
+            "owning_project": owning_project,
+            "component": component,  # structural entity id
+            "source_locations": list(source_locations or []),
+            "contract": contract,
+            "state_touched": state_touched,
+            "dependencies": list(dependencies or []),
+            "consumers": list(consumers or []),
+            "invariants": list(invariants or []),
+            "behaviour_specs": list(behaviour_specs or []),
+            "maturity": "implemented",
+            "constraints": constraints,
+            "generation": generation,
+            "limitations": [],
+            "created_at": _now(),
+            "updated_at": _now(),
+        }
+        self.doc["capabilities"][cid] = rec
+        return rec
+
+    def get(self, capability_id: str) -> dict | None:
+        return self.doc["capabilities"].get(
+            capability_id if ":" in capability_id
+            else f"capability:{capability_id}")
+
+    def material_change(self, capability_id: str, *, note: str,
+                        generation: str = "",
+                        source_locations: list[str] | None = None) -> dict:
+        """Record a material change (auto-invoked after construction —
+        rule 19Q.2). History entry keeps traceability; dependents' harden
+        state is the behaviour ledger's job (via structural closure)."""
+        rec = self.get(capability_id)
+        if rec is None:
+            raise ValueError(f"unknown capability {capability_id!r}")
+        # T21.1 §21.1D: deterministic revision counter. Wall-clock
+        # timestamps are metadata, not causal ordering (same-second events
+        # share a stamp — the T21 race). The revision counter is the
+        # canonical drift signal: material_revision > last governed
+        # spec_revision = SPEC_VERSION_DRIFT.
+        rev = int(rec.get("material_revision", 0)) + 1
+        rec["material_revision"] = rev
+        hist = {"at": _now(), "event": "material_change",
+                "note": note[:200], "generation": generation,
+                "prior_maturity": rec["maturity"],
+                "material_revision": rev}
+        self.doc["history"].append({"capability_id": rec["capability_id"],
+                                    **hist})
+        rec["updated_at"] = _now()
+        if generation:
+            rec["generation"] = generation
+        if source_locations is not None:
+            rec["source_locations"] = list(source_locations)
+        return rec
+
+    def move(self, capability_id: str, new_locations: list[str]) -> dict:
+        """File move: STABLE IDENTITY SURVIVES (rule 19S-6); the move is
+        recorded, history preserved."""
+        rec = self.get(capability_id)
+        if rec is None:
+            raise ValueError(f"unknown capability {capability_id!r}")
+        old = list(rec["source_locations"])
+        rec["source_locations"] = list(new_locations)
+        rec["updated_at"] = _now()
+        self.doc["history"].append({
+            "capability_id": rec["capability_id"], "at": _now(),
+            "event": "moved", "from": old, "to": list(new_locations)})
+        return rec
+
+    def supersede(self, capability_id: str, by: str, *,
+                  note: str = "") -> dict:
+        rec = self.get(capability_id)
+        if rec is None:
+            raise ValueError(f"unknown capability {capability_id!r}")
+        rec["maturity"] = "superseded"
+        rec["superseded_by"] = (by if ":" in by else f"capability:{by}")
+        rec["updated_at"] = _now()
+        self.doc["history"].append({
+            "capability_id": rec["capability_id"], "at": _now(),
+            "event": "superseded", "by": rec["superseded_by"],
+            "note": note[:160]})
+        return rec
+
+    def retire(self, capability_id: str, *, note: str = "") -> dict:
+        rec = self.get(capability_id)
+        if rec is None:
+            raise ValueError(f"unknown capability {capability_id!r}")
+        rec["maturity"] = "retired"
+        rec["updated_at"] = _now()
+        self.doc["history"].append({
+            "capability_id": rec["capability_id"], "at": _now(),
+            "event": "retired", "note": note[:160]})
+        return rec
+
+    # -- queries (surgical context, 19E) -----------------------------------
+
+    def active(self) -> list[dict]:
+        return [r for r in self.doc["capabilities"].values()
+                if r["maturity"] not in ("retired", "superseded")]
+
+    def by_component(self, component: str) -> list[dict]:
+        return [r for r in self.active() if r["component"] == component]
+
+    def context_packet(self, capability_id: str) -> dict:
+        """The LOCAL context packet (19E): everything a fresh builder
+        needs for one capability — bounded, structured, no reasoning."""
+        rec = self.get(capability_id)
+        if rec is None:
+            return {"error": f"unknown capability {capability_id!r}"}
+        return {
+            "schema_version": LEDGER_SCHEMA_VERSION,
+            "capability_id": rec["capability_id"],
+            "name": rec["name"],
+            "purpose": rec["purpose"],
+            "component": rec["component"],
+            "source_locations": rec["source_locations"],
+            "contract": rec["contract"],
+            "state_touched": rec["state_touched"],
+            "dependencies": rec["dependencies"],
+            "consumers": rec["consumers"],
+            "invariants": rec["invariants"],
+            "behaviour_specs": rec["behaviour_specs"],
+            "maturity": rec["maturity"],
+            "constraints": rec["constraints"],
+            "generation": rec["generation"],
+            "limitations": rec["limitations"],
+        }
+
+
+# ── T21 §21C/21D: LIVING CAPABILITY SPECIFICATIONS + STRICT SEMVER ──────
+#
+# A Living Capability Spec is the WHY+WHAT-SHOULD-EXIST layer; the
+# capability record remains the WHAT-EXISTS+WHERE layer. Both live in the
+# SAME ledger document (rule C: no second authority) — the spec is a
+# versioned field-set on the capability, not a separate subsystem.
+#
+# STRICT SEMVER (§21D): only meaningful units get versions.
+#   PATCH: bug fix / internal change, contract compatible
+#   MINOR: backwards-compatible new capability/behaviour
+#   MAJOR: breaking contract/interface/state-semantics change
+# Baselines for existing capabilities: explicit
+# governance_version_started_at + baseline_reason; no fake history.
+
+SPEC_SCHEMA_VERSION = 1
+
+import re as _re
+
+SEMVER_RE = _re.compile(r"^(\d+)\.(\d+)\.(\d+)$")
+
+
+class SpecError(ValueError):
+    """Invalid spec operation (material change without bump, unjustified
+    bump, bad version...)."""
+
+
+def classify_version_bump(*, contract_changed: bool,
+                          behaviour_added: bool,
+                          consumer_break: bool,
+                          implementation_only: bool) -> str:
+    """§21D-3: derive PATCH/MINOR/MAJOR from the change shape. The model
+    may propose; the machinery verifies."""
+    if consumer_break:
+        return "MAJOR"
+    if contract_changed:
+        # contract changed but no consumer break: interface tightened —
+        # still breaking for consumers relying on the old shape
+        return "MAJOR"
+    if behaviour_added:
+        return "MINOR"
+    if implementation_only:
+        return "PATCH"
+    raise SpecError("no version-worthy change: metadata churn does not bump")
+
+
+def bump_version(version: str, kind: str) -> str:
+    m = SEMVER_RE.match(version)
+    if not m:
+        raise SpecError(f"invalid semver {version!r}")
+    major, minor, patch = (int(x) for x in m.groups())
+    if kind == "MAJOR":
+        return f"{major + 1}.0.0"
+    if kind == "MINOR":
+        return f"{major}.{minor + 1}.0"
+    if kind == "PATCH":
+        return f"{major}.{minor}.{patch + 1}"
+    raise SpecError(f"unknown bump kind {kind!r}")
+
+
+def attach_spec(ledger: "CapabilityLedger", capability_id: str, *,
+                requirements: list[str],
+                expected_behaviours: list[str],
+                non_goals: list[str],
+                acceptance_criteria: list[str],
+                baseline_version: str = "0.1.0",
+                baseline_reason: str = "initial governed baseline",
+                governance_version_started_at: str | None = None) -> dict:
+    """Attach a Living Capability Specification to an existing capability
+    record (rule 21C: extend, don't duplicate). Baseline policy (§21D-1):
+    the starting version is explicit — 1.0.0 only for stable contracts
+    with sufficient evidence; otherwise 0.x with a reason."""
+    from rune.capability_ledger import LEDGER_SCHEMA_VERSION  # noqa: F401
+    rec = ledger.get(capability_id)
+    if rec is None:
+        raise ValueError(f"unknown capability {capability_id!r}")
+    if not SEMVER_RE.match(baseline_version):
+        raise SpecError(f"invalid baseline semver {baseline_version!r}")
+    spec = {
+        "schema_version": SPEC_SCHEMA_VERSION,
+        "version": baseline_version,
+        "governance_version_started_at": governance_version_started_at
+        or _now(),
+        "baseline_reason": baseline_reason,
+        "requirements": requirements,
+        "expected_behaviours": expected_behaviours,
+        "non_goals": non_goals,
+        "acceptance_criteria": acceptance_criteria,
+        "version_history": [{
+            "at": _now(), "version": baseline_version, "kind": "BASELINE",
+            "reason": baseline_reason, "changed_sections": [],
+            "affected_consumers": [], "validation_consequences": [],
+        }],
+    }
+    rec["spec"] = spec
+    rec["updated_at"] = _now()
+    return spec
+
+
+def revise_spec(ledger: "CapabilityLedger", capability_id: str, *,
+                contract_changed: bool = False,
+                behaviour_added: bool = False,
+                consumer_break: bool = False,
+                implementation_only: bool = False,
+                reason: str,
+                changed_sections: list[str] | None = None,
+                affected_consumers: list[str] | None = None,
+                validation_consequences: list[str] | None = None,
+                proposed_kind: str | None = None) -> dict:
+    """§21D-3/21E: revise a spec with an AUTOMATIC version decision.
+    The model proposes; the machinery classifies and records. Material
+    change without a bump is impossible (every revise computes one);
+    unjustified bumps are impossible (no-change revise raises)."""
+    rec = ledger.get(capability_id)
+    if rec is None:
+        raise ValueError(f"unknown capability {capability_id!r}")
+    spec = rec.get("spec")
+    if spec is None:
+        raise SpecError(f"{capability_id} has no spec — attach one first")
+    kind = (proposed_kind or classify_version_bump(
+        contract_changed=contract_changed,
+        behaviour_added=behaviour_added,
+        consumer_break=consumer_break,
+        implementation_only=implementation_only))
+    # verify a PROPOSED kind against the change shape (model cannot
+    # upgrade PATCH to MINOR without the corresponding change shape)
+    actual = classify_version_bump(
+        contract_changed=contract_changed,
+        behaviour_added=behaviour_added,
+        consumer_break=consumer_break,
+        implementation_only=implementation_only)
+    order = {"PATCH": 0, "MINOR": 1, "MAJOR": 2}
+    if proposed_kind and order[proposed_kind] > order[actual]:
+        raise SpecError(
+            f"proposed {proposed_kind} exceeds change shape {actual} — "
+            f"unjustified bump")
+    old = spec["version"]
+    new = bump_version(old, kind)
+    spec["version"] = new
+    spec["version_history"].append({
+        "at": _now(), "version": new, "kind": kind,
+        "reason": reason[:200],
+        "changed_sections": list(changed_sections or []),
+        "affected_consumers": list(affected_consumers or []),
+        "validation_consequences": list(validation_consequences or []),
+        # T21.1 §21.1D: the governed decision COVERS all material
+        # revisions up to and including the current one — drift closes
+        # deterministically, not by timestamp comparison
+        "covers_material_revision": int(rec.get("material_revision", 0)),
+    })
+    rec["updated_at"] = _now()
+    return {"old": old, "new": new, "kind": kind, "reason": reason}
+
+
+def spec_drift_status(ledger: "CapabilityLedger",
+                      capability_id: str) -> dict:
+    """T21.1 §21.1D: the callable drift detector / consistency gate.
+    Deterministic and revision-based (never wall-clock):
+
+      material_revision > last governed covers_material_revision
+        → SPEC_VERSION_DRIFT + durable obligation
+
+    Returns {"drift": bool, "obligation": str|None, "detail": ...}."""
+    rec = ledger.get(capability_id)
+    if rec is None:
+        return {"drift": False, "obligation": None,
+                "detail": f"unknown capability {capability_id!r}"}
+    spec = rec.get("spec")
+    if spec is None:
+        return {"drift": False, "obligation": None,
+                "detail": "no spec attached — nothing to drift from"}
+    hist = spec.get("version_history") or []
+    covered = hist[-1].get("covers_material_revision", 0) if hist else 0
+    current = int(rec.get("material_revision", 0))
+    if current > covered:
+        return {
+            "drift": True,
+            "obligation": (
+                f"SPEC_VERSION_DRIFT: {capability_id} has "
+                f"{current - covered} uncovered material revision(s) "
+                f"(material rev {current} > governed rev {covered}) — "
+                f"run revise_spec (PATCH/MINOR/MAJOR) to close"),
+            "detail": {"material_revision": current,
+                       "governed_revision": covered,
+                       "spec_version": spec.get("version")},
+        }
+    return {"drift": False, "obligation": None,
+            "detail": {"material_revision": current,
+                       "governed_revision": covered,
+                       "spec_version": spec.get("version")}}
+
+
+def spec_context(rec: dict) -> dict:
+    """The spec half of a surgical context packet (§21M): WHY + WHAT
+    SHOULD EXIST, compact."""
+    s = rec.get("spec") or {}
+    return {
+        "capability_id": rec["capability_id"],
+        "version": s.get("version"),
+        "purpose": rec.get("purpose"),
+        "requirements": s.get("requirements", []),
+        "expected_behaviours": s.get("expected_behaviours", []),
+        "non_goals": s.get("non_goals", []),
+        "acceptance_criteria": s.get("acceptance_criteria", []),
+        "contract": rec.get("contract"),
+        "invariants": rec.get("invariants", []),
+        "behaviour_specs": rec.get("behaviour_specs", []),
+        "last_bump": (s.get("version_history") or [{}])[-1],
+    }
@@ -1,602 +0,0 @@
-#!/usr/bin/env python3
-"""complexity_scorer.py — Auto-detection of task complexity for build mode.
-
-Scores user requests to determine if they need:
- Task mode (0-3): direct execution
- Task mode with planning (4-7): add planning step
- Build mode (8+): full pipeline
-
-Research basis:
- No existing tool does auto-detection (SWE-agent, Aider, Devin, Cursor, OpenCode)
- Simple keyword scoring: zero cost for 90% of requests, LLM fallback for 10%
- Agent reliability degrades 96% → 34% from 8K to 128K tokens (LOCA-bench, ICML 2026)
- Single-file changes: ~80% success vs <50% for 4-6 files (SWE-bench)
-
-Usage:
-    complexity_scorer.py score "add user auth with JWT, update API routes, add tests"
-    complexity_scorer.py mode "fix the typo in README"
-"""
-
-import json
-import os
-import re
-import sys
-from dataclasses import dataclass, field, asdict
-from typing import List, Dict, Any, Optional
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# SIGNAL DEFINITIONS
-# ═══════════════════════════════════════════════════════════════════════════════
-
-# Request analysis signals (regex, zero cost)
-CONJUNCTIONS = ["and", "then", "also", "additionally", "after that", "plus", "as well as"]
-CROSS_SYSTEM_REFS = [
-    "api",
-    "frontend",
-    "backend",
-    "database",
-    "tests",
-    "config",
-    "ci",
-    "deploy",
-    "ui",
-    "cli",
-    "server",
-    "client",
-    "model",
-    "view",
-]
-ARCH_KEYWORDS = [
-    "refactor",
-    "migrate",
-    "redesign",
-    "pipeline",
-    "system",
-    "architecture",
-    "overhaul",
-    "restructure",
-    "reorganize",
-    "consolidate",
-]
-NEW_ABSTRACTION_KEYWORDS = [
-    "create a new",
-    "build a new",
-    "implement a new",
-    "add a new",
-    "new module",
-    "new system",
-    "new feature",
-    "new component",
-    "new service",
-    "new layer",
-]
-SCOPE_MODIFIERS = ["entire", "all", "every", "throughout", "comprehensive", "complete", "full", "whole", "across"]
-EXT_DEPENDENCY_KEYWORDS = [
-    "integrate",
-    "third-party",
-    "library",
-    "package",
-    "sdk",
-    "api",
-    "service",
-    "webhook",
-    "external",
-]
-COMPLEXITY_MARKERS = [
-    "complex",
-    "big change",
-    "major",
-    "significant",
-    "extensive",
-    "substantial",
-    "non-trivial",
-    "involved",
-]
-SIMPLE_FIX_INDICATORS = [
-    "fix typo",
-    "rename",
-    "update comment",
-    "change color",
-    "add import",
-    "remove import",
-    "fix whitespace",
-    "format",
-    "lint",
-    "bump version",
-    "update readme",
-    "fix comment",
-]
-
-# Build mode trigger keywords (high confidence)
-BUILD_TRIGGERS = [
-    "implement",
-    "build",
-    "create feature",
-    "refactor",
-    "migrate",
-    "across multiple files",
-    "new system",
-    "new module",
-    "overhaul",
-    "redesign",
-    "restructure",
-    "full pipeline",
-    "end-to-end",
-]
-
-# Task mode trigger keywords
-TASK_TRIGGERS = [
-    "fix",
-    "change",
-    "update",
-    "rename",
-    "bug",
-    "typo",
-    "error",
-    "crash",
-    "broken",
-    "not working",
-    "issue",
-    "patch",
-    "hotfix",
-]
-
-# Research mode trigger keywords
-RESEARCH_TRIGGERS = [
-    "how does",
-    "what is",
-    "explain",
-    "investigate",
-    "explore",
-    "research",
-    "understand",
-    "analyze",
-    "study",
-    "review",
-]
-
-# Plan mode trigger keywords
-PLAN_TRIGGERS = [
-    "plan",
-    "design",
-    "architecture",
-    "before we start",
-    "approach",
-    "strategy",
-    "roadmap",
-    "blueprint",
-    "spec",
-]
-
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# SCORING FUNCTIONS
-# ═══════════════════════════════════════════════════════════════════════════════
-
-
-@dataclass
-class ScoreBreakdown:
-    """Detailed breakdown of complexity score."""
-
-    request_analysis: int = 0
-    codebase_impact: int = 0
-    historical_patterns: int = 0
-    risk_assessment: int = 0
-    total: int = 0
-    signals: List[str] = field(default_factory=list)
-    mode: str = "task"
-    confidence: str = "high"  # high, medium, low
-
-
-def analyze_request_text(request: str) -> tuple[int, List[str]]:
-    """Analyze request text for complexity signals. Zero cost, <1ms."""
-    score = 0
-    signals = []
-    request_lower = request.lower()
-
-    # Multiple verbs/conjunctions
-    conjunction_matches = [c for c in CONJUNCTIONS if c in request_lower]
-    if len(conjunction_matches) >= 2:
-        score += 2
-        signals.append(f"Multiple conjunctions: {', '.join(conjunction_matches)}")
-    elif len(conjunction_matches) == 1:
-        score += 1
-        signals.append(f"Conjunction: {conjunction_matches[0]}")
-
-    # Cross-system references
-    system_matches = [s for s in CROSS_SYSTEM_REFS if s in request_lower]
-    if len(system_matches) >= 2:
-        score += 3
-        signals.append(f"Cross-system: {', '.join(system_matches)}")
-    elif len(system_matches) == 1:
-        score += 1
-        signals.append(f"System ref: {system_matches[0]}")
-
-    # Architecture keywords
-    arch_matches = [k for k in ARCH_KEYWORDS if k in request_lower]
-    if arch_matches:
-        score += 2
-        signals.append(f"Architecture: {', '.join(arch_matches)}")
-
-    # New abstraction keywords
-    new_matches = [k for k in NEW_ABSTRACTION_KEYWORDS if k in request_lower]
-    if new_matches:
-        score += 2
-        signals.append(f"New abstraction: {', '.join(new_matches)}")
-
-    # Scope modifiers
-    scope_matches = [s for s in SCOPE_MODIFIERS if s in request_lower]
-    if scope_matches:
-        score += 1
-        signals.append(f"Scope: {', '.join(scope_matches)}")
-
-    # External dependency mentions
-    ext_matches = [k for k in EXT_DEPENDENCY_KEYWORDS if k in request_lower]
-    if ext_matches:
-        score += 1
-        signals.append(f"External dep: {', '.join(ext_matches)}")
-
-    # Explicit complexity markers
-    complex_matches = [k for k in COMPLEXITY_MARKERS if k in request_lower]
-    if complex_matches:
-        score += 3
-        signals.append(f"Complexity marker: {', '.join(complex_matches)}")
-
-    # Simple fix indicators (negative score)
-    simple_matches = [k for k in SIMPLE_FIX_INDICATORS if k in request_lower]
-    if simple_matches:
-        score -= 3
-        signals.append(f"Simple fix: {', '.join(simple_matches)}")
-
-    return score, signals
-
-
-def estimate_codebase_impact(request: str, repo_path: str = "/mnt/workdrive/rune") -> tuple[int, List[str]]:
-    """Estimate codebase impact using grep/glob. ~100ms."""
-    score = 0
-    signals = []
-
-    # Extract potential module/function names from request
-    words = re.findall(r"\b[a-zA-Z_][a-zA-Z0-9_]*\b", request)
-    # Filter out common words
-    stop_words = {
-        "the",
-        "a",
-        "an",
-        "and",
-        "or",
-        "but",
-        "in",
-        "on",
-        "at",
-        "to",
-        "for",
-        "of",
-        "with",
-        "by",
-        "from",
-        "is",
-        "are",
-        "was",
-        "were",
-        "be",
-        "been",
-        "being",
-        "have",
-        "has",
-        "had",
-        "do",
-        "does",
-        "did",
-        "will",
-        "would",
-        "could",
-        "should",
-        "may",
-        "might",
-        "can",
-        "shall",
-        "that",
-        "which",
-        "who",
-        "whom",
-        "this",
-        "these",
-        "those",
-        "it",
-        "its",
-        "my",
-        "your",
-        "his",
-        "her",
-        "our",
-        "their",
-        "me",
-        "you",
-        "him",
-        "us",
-        "them",
-        "i",
-        "we",
-        "he",
-        "she",
-        "they",
-        "what",
-        "where",
-        "when",
-        "how",
-        "why",
-        "if",
-        "then",
-        "else",
-        "so",
-        "just",
-        "also",
-        "very",
-        "too",
-        "much",
-        "many",
-        "some",
-        "any",
-        "all",
-        "each",
-        "every",
-        "both",
-        "few",
-        "more",
-        "most",
-        "other",
-        "another",
-        "such",
-        "no",
-        "not",
-        "only",
-        "own",
-        "same",
-        "than",
-        "too",
-        "now",
-        "here",
-        "there",
-        "about",
-        "up",
-        "out",
-        "off",
-        "over",
-        "under",
-        "again",
-        "further",
-        "then",
-        "once",
-        "new",
-        "add",
-        "update",
-        "fix",
-        "change",
-        "make",
-        "create",
-        "build",
-        "implement",
-    }
-    potential_modules = [w for w in words if w.lower() not in stop_words and len(w) > 3]
-
-    # Check if any potential modules exist in codebase
-    if potential_modules:
-        try:
-            import subprocess
-
-            # Quick grep for potential module names
-            grep_pattern = "|".join(potential_modules[:5])  # Limit to 5 to avoid slow grep
-            result = subprocess.run(
-                ["grep", "-r", "-l", "-E", grep_pattern, os.path.join(repo_path, "src")],
-                capture_output=True,
-                text=True,
-                timeout=5,
-            )
-            if result.returncode == 0:
-                files = [f for f in result.stdout.strip().split("\n") if f]
-                if len(files) > 5:
-                    score += 3
-                    signals.append(f"Files likely affected: {len(files)}")
-                elif len(files) > 2:
-                    score += 1
-                    signals.append(f"Files possibly affected: {len(files)}")
-        except Exception:
-            pass
-
-    # Check for cross-module dependencies
-    if "src/rune" in repo_path:
-        try:
-            import subprocess
-
-            result = subprocess.run(
-                ["grep", "-r", "-c", "import", os.path.join(repo_path, "src/rune")],
-                capture_output=True,
-                text=True,
-                timeout=5,
-            )
-            if result.returncode == 0:
-                import_count = sum(
-                    int(line.split(":")[-1])
-                    for line in result.stdout.strip().split("\n")
-                    if ":" in line and line.split(":")[-1].strip().isdigit()
-                )
-                if import_count > 50:
-                    score += 2
-                    signals.append(f"High import density: {import_count} imports")
-        except Exception:
-            pass
-
-    return score, signals
-
-
-def check_historical_patterns(request: str) -> tuple[int, List[str]]:
-    """Check memory for similar past tasks. Returns score and signals."""
-    score = 0
-    signals = []
-
-    # Check if similar tasks needed build mode
-    try:
-        import sys
-
-        sys.path.insert(0, os.path.expanduser("~/.config/opencode/scripts"))
-        import memory_store
-
-        db = memory_store.connect()
-        try:
-            results = memory_store.cached_recall(db, f"build mode {request}", budget=3)
-            if results:
-                for r in results:
-                    text = r.get("text", "").lower()
-                    if "build mode" in text or "multi-step" in text or "complex" in text:
-                        score += 2
-                        signals.append("Similar past task needed build mode")
-                        break
-        finally:
-            db.close()
-    except Exception:
-        pass
-
-    return score, signals
-
-
-def assess_risk(request: str, repo_path: str = "/mnt/workdrive/rune") -> tuple[int, List[str]]:
-    """Assess risk based on file paths and content."""
-    score = 0
-    signals = []
-    request_lower = request.lower()
-
-    # Auth/security code
-    auth_keywords = ["auth", "security", "token", "credential", "password", "login", "session"]
-    if any(k in request_lower for k in auth_keywords):
-        score += 2
-        signals.append("Touches auth/security code")
-
-    # Data layer
-    data_keywords = ["database", "migration", "schema", "sql", "sqlite", "model", "orm"]
-    if any(k in request_lower for k in data_keywords):
-        score += 1
-        signals.append("Touches data layer")
-
-    # Public API
-    api_keywords = ["api", "endpoint", "route", "export", "interface", "sdk", "public"]
-    if any(k in request_lower for k in api_keywords):
-        score += 2
-        signals.append("Touches public API")
-
-    # Reversibility concern
-    destructive_keywords = ["delete", "remove", "drop", "destroy", "clear", "reset", "wipe"]
-    if any(k in request_lower for k in destructive_keywords):
-        score += 1
-        signals.append("Reversibility concern")
-
-    return score, signals
-
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# MAIN SCORING
-# ═══════════════════════════════════════════════════════════════════════════════
-
-
-def score_request(request: str, repo_path: str = "/mnt/workdrive/rune") -> ScoreBreakdown:
-    """Score a request for complexity. Returns ScoreBreakdown with total and mode."""
-    breakdown = ScoreBreakdown()
-
-    # Signal Category 1: Request Analysis (zero cost, <1ms)
-    req_score, req_signals = analyze_request_text(request)
-    breakdown.request_analysis = req_score
-    breakdown.signals.extend(req_signals)
-
-    # Signal Category 2: Codebase Impact (~100ms)
-    impact_score, impact_signals = estimate_codebase_impact(request, repo_path)
-    breakdown.codebase_impact = impact_score
-    breakdown.signals.extend(impact_signals)
-
-    # Signal Category 3: Historical Patterns
-    hist_score, hist_signals = check_historical_patterns(request)
-    breakdown.historical_patterns = hist_score
-    breakdown.signals.extend(hist_signals)
-
-    # Signal Category 4: Risk Assessment
-    risk_score, risk_signals = assess_risk(request, repo_path)
-    breakdown.risk_assessment = risk_score
-    breakdown.signals.extend(risk_signals)
-
-    # Calculate total
-    breakdown.total = (
-        breakdown.request_analysis
-        + breakdown.codebase_impact
-        + breakdown.historical_patterns
-        + breakdown.risk_assessment
-    )
-
-    # Determine mode
-    if breakdown.total >= 8:
-        breakdown.mode = "build"
-        breakdown.confidence = "high"
-    elif breakdown.total >= 4:
-        breakdown.mode = "task_with_planning"
-        breakdown.confidence = "medium"
-    else:
-        breakdown.mode = "task"
-        breakdown.confidence = "high"
-
-    # Check for explicit mode triggers
-    request_lower = request.lower()
-    if any(t in request_lower for t in BUILD_TRIGGERS):
-        if breakdown.mode != "build":
-            breakdown.mode = "build"
-            breakdown.confidence = "high"
-            breakdown.signals.append("Explicit build trigger detected")
-    elif any(t in request_lower for t in RESEARCH_TRIGGERS):
-        breakdown.mode = "research"
-        breakdown.confidence = "high"
-        breakdown.signals.append("Research trigger detected")
-    elif any(t in request_lower for t in PLAN_TRIGGERS):
-        breakdown.mode = "plan"
-        breakdown.confidence = "high"
-        breakdown.signals.append("Plan trigger detected")
-
-    return breakdown
-
-
-def get_mode(request: str, repo_path: str = "/mnt/workdrive/rune") -> str:
-    """Get the recommended mode for a request. Returns mode string."""
-    breakdown = score_request(request, repo_path)
-    return breakdown.mode
-
-
-def get_score(request: str, repo_path: str = "/mnt/workdrive/rune") -> Dict[str, Any]:
-    """Get detailed score breakdown. Returns dict."""
-    breakdown = score_request(request, repo_path)
-    return asdict(breakdown)
-
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# CLI
-# ═══════════════════════════════════════════════════════════════════════════════
-
-
-def main():
-    if len(sys.argv) < 3:
-        print("Usage: complexity_scorer.py [score|mode] <request>")
-        sys.exit(1)
-
-    cmd = sys.argv[1]
-    request = " ".join(sys.argv[2:])
-
-    if cmd == "score":
-        result = get_score(request)
-        print(json.dumps(result, indent=2))
-
-    elif cmd == "mode":
-        mode = get_mode(request)
-        print(mode)
-
-    else:
-        print(f"Unknown command: {cmd}")
-        sys.exit(1)
-
-
-if __name__ == "__main__":
-    main()
@@ -0,0 +1,237 @@
+"""construction_activation.py — the activation dispatcher.
+
+SCAFFOLD-TRANCHE-0011. §3: NOT another planner/router/engine. Given an
+evidence-backed ConstructionMode, start/resume the correct Scaffolding
+lifecycle. ONE activation authority; consumers: Rune native backend path,
+Construction Interface (MiMo/other builders), manual operator override.
+
+Mode dispatch (tranche brief §4/§7/§8/§9/§10):
+
+    NOT_CONSTRUCTION   → no Scaffolding state touched (§10)
+    LOCAL_CHANGE       → bounded governance via the existing project's
+                         authority; Construction Contract only if policy
+                         requires; no Project Scaffold bootstrap (§7)
+    EXISTING_PROJECT_CHANGE
+                       → attach to the existing project authority (Rune
+                         Spine for Rune; bound Project Scaffold otherwise);
+                         full proportional lifecycle (§8) — never a new
+                         project (§1 existence evidence governs)
+    PROJECT_BUILD      → actually bootstrap: provisional Project Scaffold
+                         initialized with UNKNOWN fields explicit (§6/§7) —
+                         repository: undecided, toolchain: unresolved —
+                         progressive refinement, nothing manufactured
+    UNKNOWN            → NO activation; return evidence gaps + probe
+                         questions (§9)
+
+Persistence discipline (§21): every state mutation goes through
+Scaffold.commit() (save + readback verify) BEFORE any accounting.
+"""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+from rune import structural_store as ss
+from rune import scaffold as sc
+from rune import construction_mode as cm
+
+__all__ = ["activate"]
+
+
+def _scaffold_for(project: str, *, create: bool, parent_scaffold: sc.Scaffold | None = None) -> sc.Scaffold | None:
+    """Resolve the Project Scaffold store for a project (§1 evidence:
+    existence comes from a bound store on disk, not from the name).
+    Location derives from the PARENT scaffold's store directory — no
+    hardcoded relative path (T11 fix: activation under a tmp/fixture parent
+    must write the fixture root, not the repo's store)."""
+    parent_dir = parent_scaffold.current.parent if parent_scaffold else Path("store/scaffold")
+    base = parent_dir / "projects" / project
+    current = base / "current.json"
+    if current.is_file():
+        return sc.Scaffold(base, project)
+    if create:
+        return sc.Scaffold(base, project)
+    return None
+
+
+def activate(mode_decision, *, need, scaffold: sc.Scaffold, builder: str = "rune") -> dict:
+    """Dispatch on the evidence-backed mode; return an activation record
+    (mode, project authority, actions taken, context ref). The NEED must
+    already be classified — UNKNOWN never reaches here without explicit
+    operator override carried in mode_decision."""
+    mode = mode_decision.mode
+    rec: dict = {
+        "mode": mode,
+        "builder": builder,
+        "target_project": mode_decision.target_project,
+        "evidence": list(mode_decision.evidence),
+    }
+
+    if mode == "NOT_CONSTRUCTION":
+        # §10: no Scaffolding state touched
+        rec["activated"] = False
+        rec["reason"] = "non-construction intent — no construction state"
+        return rec
+
+    if mode == "UNKNOWN":
+        # §9: no activation; probe questions surface
+        rec["activated"] = False
+        rec["reason"] = "insufficient evidence — clarification required; no Project Scaffold initialized"
+        rec["probe_questions"] = mode_decision.probe_questions
+        rec["clarification_required"] = True
+        return rec
+
+    if mode == "LOCAL_CHANGE":
+        # §7: governance without ceremony — reuse the existing project's
+        # authority; the lightest contract path (caller decides policy);
+        # NO new Project Scaffold bootstrap
+        rec["activated"] = True
+        rec["lifecycle"] = "bounded"
+        rec["project_authority"] = mode_decision.project_authority or "rune_internal"
+        rec["note"] = "bounded governance via existing authority — no Project Scaffold bootstrap"
+        return rec
+
+    if mode == "EXISTING_PROJECT_CHANGE":
+        # §8: attach to the existing authority — never a new project.
+        # Existence was already proven structurally (T11 §1) before this
+        # mode was selected.
+        rec["activated"] = True
+        rec["lifecycle"] = "proportional"
+        rec["project_authority"] = mode_decision.project_authority or "rune_internal"
+        rec["note"] = (
+            "attached to existing project authority; Need/completeness/patterns/contract flow governs proportionally"
+        )
+        return rec
+
+    if mode == "PROJECT_BUILD":
+        # §6/§7: actually bootstrap. Provisional Project Scaffold with
+        # explicit UNKNOWN fields; identity before repository (§7).
+        project = mode_decision.target_project or need.project
+        if not project:
+            # Rune-originated Needs may carry no project identity (§18):
+            # derive a stable provisional id from the objective — the
+            # bootstrap is provisional and renameable before any repo
+            # exists (§7 identity before repository).
+            import re as _re
+
+            slug = _re.sub(r"[^a-z0-9]+", "-", need.objective.lower()).strip("-")[:48]
+            project = slug or "unnamed-project"
+        ps = _scaffold_for(project, create=True, parent_scaffold=scaffold)
+        assert ps is not None
+        pdoc = ps.load()
+        if ss.entity(pdoc, f"project:{project}") is None:
+            ps.init(purpose=need.objective)  # init persists its own doc
+            pdoc = ps.load()  # §21: reload after persistence
+        structural = ss.entity(pdoc, f"project:{project}")
+        assert structural is not None
+        # explicit UNKNOWN fields — nothing manufactured (§6/§10)
+        structural.setdefault("ext", {}).update(
+            {
+                "bootstrap": "initialized",
+                "repository": "undecided",
+                "toolchain": "unresolved",
+                "deployment": "unknown",
+                "requirement_gaps": list(mode_decision.probe_questions)
+                or ["target platform unknown", "requirements not yet refined"],
+                "bootstrap_started_at": ss.now(),
+                "origin_need": need.need_id,
+                "origin_builder": builder,
+            }
+        )
+        ps.commit(pdoc, what=f"PROJECT_BUILD bootstrap: {project}")
+        awareness = _auto_structural_awareness(project, ps, scaffold)
+        rec.update(
+            {
+                "activated": True,
+                "lifecycle": "project_build",
+                "project": project,
+                "project_scaffold": str(ps.current),
+                "structural_awareness": awareness,
+                "bootstrap": "initialized",
+                "repository": "undecided",
+                "toolchain": "unresolved",
+                "deployment": "unknown",
+                "requirement_gaps": structural["ext"]["requirement_gaps"],
+                "next_action": ("refine requirements; resolve toolchain/repo decisions via evidence (approval-gated)"),
+            }
+        )
+        return rec
+
+    rec["activated"] = False
+    rec["reason"] = f"unhandled mode {mode}"
+    return rec
+
+
+def _auto_structural_awareness(project: str, project_scaffold: sc.Scaffold,
+                               parent_scaffold: sc.Scaffold) -> dict:
+    """T21.5 §9 (rule 19Q family): governed adoption/construction expands
+    Rune's persistent structural world-model AUTOMATICALLY — no manual
+    Registry/Spine bookkeeping step. Boundaries preserved:
+      - Registry = discovery/index (that the project exists, where its
+        canonical Scaffold lives) — never project internals;
+      - Spine = Rune's relationship TO the project (one edge, builds/
+        manages vocabulary) — project modules stay in the Project
+        Scaffold (never duplicated into Spine);
+      - Project Scaffold remains canonical for the project's internals.
+    Failure-tolerant: awareness must not break the bootstrap (recorded
+    in the return, not raised)."""
+    result: dict = {"registry": "skipped", "spine": "skipped"}
+    try:
+        from rune.project_registry import ProjectRegistry
+        store_dir = parent_scaffold.current.parent
+        registry = ProjectRegistry(store_dir)
+        if registry.get(project) is None:
+            registry.register(
+                project_id=project,
+                scaffold_location=str(project_scaffold.current.parent),
+                repo_identity="",   # undecided at bootstrap (identity
+                                    # before repository — never invented)
+            )
+            registry.save()
+            result["registry"] = "registered"
+    except (OSError, ValueError) as exc:
+        result["registry"] = f"error: {exc!s:.120}"
+    try:
+        from rune import spine as spine_mod
+        spine_path = parent_scaffold.current.parent.parent / "spine" / "current.json"
+        sdoc = spine_mod.load(spine_path)
+        if sdoc is not None:
+            spine_mod.add_relationship(
+                sdoc, type="manages",
+                from_id="project:rune", to_id=f"project:{project}",
+                evidence_refs=[f"project-build-bootstrap:{project}"])
+            spine_mod.save(sdoc, spine_path)
+            result["spine"] = "recorded"
+    except (OSError, ValueError) as exc:
+        result["spine"] = f"error: {exc!s:.120}"
+    return result
+
+
+def dispatch_from_decision(trace: dict, *, need, scaffold: sc.Scaffold, builder: str = "rune") -> dict:
+    """§15: read the mode from a make_decision trace and dispatch. The
+    shared Construction Context then derives from the SAME canonical state
+    — no duplicate machines."""
+    mode_decision = trace.get("mode_decision")
+    if mode_decision is None:
+        # T21.5 fix: activate() expects the mode_decision object (with
+        # .mode/.target_project/.evidence); the trace carries its dict
+        # form, so reconstruct a ConstructionMode from it.
+        from rune.construction_mode import ConstructionMode
+        md = trace.get("construction_mode", {}) or {}
+        if md.get("mode", "UNKNOWN") == "UNKNOWN":
+            md = dict(md)
+            md["mode"] = "UNKNOWN"
+        mode_decision = ConstructionMode(
+            mode=md.get("mode", "UNKNOWN"),
+            evidence=list(md.get("evidence", [])),
+            target_project=md.get("target_project"),
+            project_authority=md.get("project_authority"),
+            confidence=md.get("confidence", 0.0),
+            clarification_required=md.get("clarification_required", False),
+            probe_questions=list(md.get("probe_questions", [])),
+            bootstrap_required=md.get("bootstrap_required", False),
+            override=md.get("override"),
+            proposed_by_model=md.get("proposed_by_model"),
+        )
+    return activate(mode_decision, need=need, scaffold=scaffold, builder=builder)
@@ -0,0 +1,173 @@
+"""construction_commands.py — T16 §16J: bounded validated inbound
+command interface for future Substrate integration.
+
+The future Substrate ASKS; Scaffolding DECIDES. Every command:
+
+  - carries identity (command_id, correlation_id, construction_id)
+  - is validated by Scaffolding against the existing lifecycle
+    (LEGAL_TRANSITIONS — no new lifecycle invented)
+  - cannot directly mutate storage (commands return validated intents;
+    the Construction Interface executes them through its own governed
+    path — the caller never writes the store)
+  - is idempotent where applicable (repeat delivery of the same
+    command_id returns the same recorded outcome)
+
+Only commands that fit EXISTING semantics safely are exposed. PAUSE and
+CANCEL are intentionally NOT exposed: the lifecycle has no paused phase
+and abandonment needs user provenance — a future Substrate asking for
+them gets a refusal, not a half-measure.
+"""
+from __future__ import annotations
+
+import time
+import uuid
+
+from rune.scaffold import LEGAL_TRANSITIONS
+
+COMMANDS = (
+    "REQUEST_STATE", "START", "RESUME", "ACKNOWLEDGE_BLOCKER",
+    "PROVIDE_APPROVAL", "PROVIDE_CONTEXT",
+)
+
+# Command → the interface operation it asks for (the interface still
+# decides: validation failures return refused, never raise into storage).
+_COMMAND_TO_OP = {
+    "REQUEST_STATE": "resume",       # read-only state projection
+    "START": "transition",           # draft → approved (then building)
+    "RESUME": "resume",              # re-read canonical state
+    "ACKNOWLEDGE_BLOCKER": "transition",  # blocked → building (retry path)
+    "PROVIDE_APPROVAL": "transition",     # draft → approved with provenance
+    "PROVIDE_CONTEXT": "amend_scope",     # context arrives as scope amendment
+}
+
+# In-memory idempotency ledger: command_id → recorded outcome. The
+# Scaffolding lifecycle is the authority; this ledger only makes repeat
+# delivery return the same answer (no double transition).
+_outcomes: dict[str, dict] = {}
+
+
+def command(command_type: str, *, construction_id: str,
+            correlation_id: str = "", command_id: str = "",
+            payload: dict | None = None,
+            scaffold=None) -> dict:
+    """Validate an inbound command. Returns a validated intent envelope:
+
+      {"accepted": True, "op": <interface op>, "args": {...},
+       "command_id": ..., "idempotent_replay": bool}
+
+    or {"accepted": False, "reason": ...}. The caller executes the intent
+    through the Construction Interface — it never touches storage."""
+    payload = payload or {}
+    command_id = command_id or f"cmd-{uuid.uuid4().hex[:8]}"
+    if command_id in _outcomes:
+        prior = dict(_outcomes[command_id])
+        prior["idempotent_replay"] = True
+        return prior
+    if command_type not in COMMANDS:
+        return _refuse(command_id, construction_id, correlation_id,
+                       f"unknown command {command_type!r} "
+                       f"(supported: {', '.join(COMMANDS)})")
+    if not construction_id:
+        return _refuse(command_id, construction_id, correlation_id,
+                       "construction_id is required")
+    intent = _validate(command_type, construction_id,
+                       correlation_id, command_id, payload, scaffold)
+    _outcomes[command_id] = intent
+    return intent
+
+
+def _refuse(command_id: str, construction_id: str,
+            correlation_id: str, reason: str) -> dict:
+    return {"accepted": False, "reason": reason,
+            "command_id": command_id, "construction_id": construction_id,
+            "correlation_id": correlation_id,
+            "recorded_at": time.strftime("%Y-%m-%dT%H:%M:%SZ",
+                                         time.gmtime()),
+            "idempotent_replay": False}
+
+
+def _validate(command_type: str, construction_id: str,
+              correlation_id: str, command_id: str,
+              payload: dict, scaffold) -> dict:
+    """Validate against the CURRENT canonical state — the lifecycle
+    decides, not the requester. The scaffold (canonical store handle) is
+    supplied by the serving layer; the client only ever sees the returned
+    envelope, never the store."""
+    if scaffold is None:
+        return _refuse(command_id, construction_id, correlation_id,
+                       "no serving scaffold bound (server-side wiring)")
+    doc = scaffold.load()
+    contract = None
+    for e in doc.get("entities", []):
+        if e.get("id") == f"contract:{construction_id}":
+            contract = e
+            break
+    if contract is None:
+        return _refuse(command_id, construction_id, correlation_id,
+                       f"unknown construction {construction_id!r}")
+    phase = contract.get("phase", "")
+
+    if command_type == "REQUEST_STATE":
+        return _accept(command_id, construction_id, correlation_id,
+                       "resume", {"contract_id": construction_id})
+    if command_type == "RESUME":
+        if phase in ("closed", "abandoned"):
+            return _refuse(command_id, construction_id, correlation_id,
+                           f"construction is {phase} — resume is invalid")
+        return _accept(command_id, construction_id, correlation_id,
+                       "resume", {"contract_id": construction_id})
+    if command_type == "START":
+        if phase != "draft":
+            return _refuse(command_id, construction_id, correlation_id,
+                           f"START requires draft phase (current: {phase})")
+        return _accept(command_id, construction_id, correlation_id,
+                       "transition",
+                       {"contract_id": construction_id,
+                        "phase": "approved"})
+    if command_type == "PROVIDE_APPROVAL":
+        if phase != "draft":
+            return _refuse(command_id, construction_id, correlation_id,
+                           f"approval applies to draft (current: {phase})")
+        if payload.get("approved_by") != "user":
+            # lifecycle rule: approval provenance must be the user —
+            # Substrate cannot approve on the user's behalf
+            return _refuse(command_id, construction_id, correlation_id,
+                           "approval provenance must be user "
+                           "(Substrate cannot approve for the user)")
+        return _accept(command_id, construction_id, correlation_id,
+                       "transition",
+                       {"contract_id": construction_id,
+                        "phase": "approved",
+                        "approved_by": "user"})
+    if command_type == "ACKNOWLEDGE_BLOCKER":
+        if phase not in ("failed", "building", "verifying"):
+            return _refuse(command_id, construction_id, correlation_id,
+                           f"no blockable phase (current: {phase})")
+        return _accept(command_id, construction_id, correlation_id,
+                       "transition",
+                       {"contract_id": construction_id,
+                        "phase": "building"})
+    if command_type == "PROVIDE_CONTEXT":
+        paths = payload.get("add_paths") or []
+        if not paths:
+            return _refuse(command_id, construction_id, correlation_id,
+                           "PROVIDE_CONTEXT requires add_paths")
+        return _accept(command_id, construction_id, correlation_id,
+                       "amend_scope",
+                       {"contract_id": construction_id,
+                        "add_paths": paths,
+                        "reason": payload.get("reason", "substrate context"),
+                        "authority": "user"
+                        if payload.get("approved_by") == "user" else "substrate"})
+    return _refuse(command_id, construction_id, correlation_id,
+                   "unreachable validator branch")
+
+
+def _accept(command_id: str, construction_id: str, correlation_id: str,
+            op: str, args: dict) -> dict:
+    return {"accepted": True, "op": op, "args": args,
+            "command_id": command_id, "construction_id": construction_id,
+            "correlation_id": correlation_id,
+            "recorded_at": time.strftime("%Y-%m-%dT%H:%M:%SZ",
+                                         time.gmtime()),
+            "idempotent_replay": False}
@@ -0,0 +1,937 @@
+"""construction_decision.py — live Need → Decision → Construction Contract.
+
+SCAFFOLD-TRANCHE-004: the adoption layer that turns the Tranche 2/3
+mechanisms into a durable procedure Rune no longer has to remember.
+
+    NEED (references Task/Intent — never duplicates them)
+      ↓ normalize_need()                missing info is NOT invented
+      ↓ discover + fit per authority    (deterministic; Spine + Project Scaffold)
+      ↓ resolve per authority + merge   (capability_fit resolution surface)
+      ↓ resolve_toolchain()             (existing project toolchain wins)
+      ↓ layout obligations              (LOCKED authority inherited)
+      ↓ approval gate                   (material changes → user)
+      ↓ DecisionTrace (durable, model-independent)
+      ↓ generate_contract()             (populated from the trace)
+
+Boundaries (brief §3, §32, §33):
+    - NOT a task engine: Need carries a reference to the existing Task/Intent
+      authority; Scaffolding never becomes the "why" authority.
+    - NOT a workflow engine: one decision function, one trace object.
+    - NOT a research pipeline: insufficient evidence routes to a recorded
+      research path (seam) — the live ADAPT wiring is a recorded route, not
+      an executed pipeline call (honest seam, §6).
+    - NOT a structural authority: it reads Spine/Project Scaffold and
+      WRITES only Construction Contracts + decision traces.
+    - BUILD_NEW is never a fallback (§8): merge precedence and the
+      insufficient-evidence invariant of capability_fit enforce it.
+"""
+
+from __future__ import annotations
+
+from rune import capability_fit as cf
+from rune import toolchain as tc
+from rune.structural_store import relationships
+
+__all__ = ["Need", "make_decision", "generate_contract", "merge_resolutions"]
+
+
+# ── Need intake (§4) ───────────────────────────────────────────────────────
+
+
+class Need:
+    """Construction-decision input. References — does not copy — the
+    existing Task/Intent authority (`task_ref` is a pointer, not a copy).
+    Carries no task state of its own: it is WHY-adjacent input to Scaffolding,
+    not a backlog entry."""
+
+    def __init__(
+        self,
+        need_id: str,
+        objective: str,
+        *,
+        origin: str = "user",
+        project: str = "rune",
+        task_ref: str = "",
+        constraints: dict | None = None,
+        affected_surface: list[str] | None = None,
+        evidence_refs: list[str] | None = None,
+        approval_context: str = "",
+        requested_permanence: str = "",
+    ):
+        if not need_id:
+            raise ValueError("need_id required")
+        if not objective:
+            raise ValueError("objective (required effect) required — do not normalize a need we cannot state")
+        self.need_id = need_id
+        self.objective = objective
+        self.origin = origin
+        self.project = project
+        self.task_ref = task_ref
+        self.constraints = dict(constraints or {})
+        self.affected_surface = list(affected_surface or [])
+        self.evidence_refs = list(evidence_refs or [])
+        self.approval_context = approval_context
+        self.requested_permanence = requested_permanence
+
+    def to_dict(self) -> dict:
+        return {
+            "need_id": self.need_id,
+            "objective": self.objective,
+            "origin": self.origin,
+            "project": self.project,
+            "task_ref": self.task_ref,
+            "constraints": dict(self.constraints),
+            "affected_surface": list(self.affected_surface),
+            "evidence_refs": list(self.evidence_refs),
+            "approval_context": self.approval_context,
+            "requested_permanence": self.requested_permanence,
+        }
+
+
+# ── requirement normalization (§5) ─────────────────────────────────────────
+
+
+def normalize_need(need: Need) -> tuple[cf.CapabilityRequirement, list[str]]:
+    """Derive the CapabilityRequirement WITHOUT inventing facts. Contracts
+    may come from explicit constraints (`contracts` key) or from contract-
+    shaped entries on the affected surface (`contract:x` ids). Everything
+    not determinable lands in `missing` — the caller records it and the
+    resolution degrades to INSUFFICIENT_EVIDENCE rather than guessing."""
+    contracts = sorted(need.constraints.get("contracts", []))
+    contracts += sorted(a for a in need.affected_surface if a.startswith("contract:"))
+    missing = []
+    if need.constraints.get("environment"):
+        pass  # carried through
+    else:
+        missing.append("environment")
+    if not need.evidence_refs:
+        missing.append("evidence_refs")
+    req = cf.CapabilityRequirement(
+        need.objective,
+        contracts=contracts,
+        environment=need.constraints.get("environment", {}),
+        target_project=need.project,
+        source_need=need.need_id,
+        task_ref=need.task_ref,
+        requested_permanence=need.requested_permanence,
+        **{k: v for k, v in need.constraints.items() if k not in ("contracts", "environment")},
+    )
+    return req, missing
+
+
+# ── affected-surface classification + resolution (Tranche 5 §3-5) ──────────
+
+_SURFACE_UI = "ui:"
+_SURFACE_REGION = "region:"
+_SURFACE_CONTRACT = "contract:"
+_SURFACE_MODULE = "module:"
+
+
+def classify_surface(surface: str) -> str:
+    """Classify one affected-surface entry (open-world; Tranche 5 §3).
+    Kinds: structural_entity | source_path | workflow_config | ui_component |
+    layout_region | unresolved. Never drops an entry — unresolved is a
+    first-class kind."""
+    if surface.startswith((_SURFACE_UI,)):
+        return "ui_component"
+    if surface.startswith((_SURFACE_REGION,)):
+        return "layout_region"
+    if surface.startswith(
+        (
+            _SURFACE_CONTRACT,
+            _SURFACE_MODULE,
+            _SURFACE_CAPABILITY,
+        )
+    ):
+        return "structural_entity"
+    if surface.endswith(".yml") or surface.endswith(".yaml") or surface.endswith(".json") or surface.endswith(".toml"):
+        return "workflow_config"
+    if (
+        surface.endswith((".py", ".js", ".ts", ".jsx", ".tsx", ".rs",
+                          ".md", ".go", ".java", ".c", ".cpp", ".h",
+                          ".css", ".html", ".yaml", ".yml",
+                          ".mjs", ".cjs", ".mts", ".cts"))
+        or "/" in surface
+    ):
+        return "source_path"
+    return "unresolved"
+
+
+_SURFACE_CAPABILITY = "capability:"
+
+
+def resolve_surface(surface: str, kind: str, authorities: dict[str, dict]) -> dict:
+    """Deterministically resolve one surface against the structural
+    authorities (Tranche 5 §5). No Memory, no LLM archaeology: identity
+    lookups and recorded path->module mappings only.
+
+    Returns {surface, kind, resolved: [...], unresolved_reason: ""}.
+    resolved entries: {entity_id, authority, via} where via records HOW the
+    mapping was established (identity | path_attribute | provides)."""
+    out = {"surface": surface, "kind": kind, "resolved": [], "unresolved_reason": ""}
+    if kind == "structural_entity":
+        for name, doc in authorities.items():
+            e = ss_entity(doc, surface)
+            if e is not None:
+                out["resolved"].append({"entity_id": surface, "authority": name, "via": "identity"})
+                break
+        if not out["resolved"]:
+            out["unresolved_reason"] = "entity id not found in any authority"
+    elif kind == "source_path":
+        # path -> module via recorded identities.path in any authority
+        for name, doc in authorities.items():
+            for e in doc["entities"]:
+                p = e.get("identities", {}).get("path", "")
+                if p and (p == surface or p.endswith("/" + surface) or surface.endswith(p)):
+                    out["resolved"].append({"entity_id": e["id"], "authority": name, "via": "path_attribute"})
+        if not out["resolved"]:
+            out["unresolved_reason"] = "no authority records this path as an entity (path remains file-scope only)"
+    elif kind == "ui_component":
+        for name, doc in authorities.items():
+            cid = surface if surface.startswith("ui:") else f"ui:{surface}"
+            e = ss_entity(doc, f"layout:{cid}")
+            if e is not None:
+                out["resolved"].append({"entity_id": f"layout:{cid}", "authority": name, "via": "layout_contract"})
+        if not out["resolved"]:
+            out["unresolved_reason"] = "no Layout Contract bound for component"
+    elif kind == "layout_region":
+        for name, doc in authorities.items():
+            e = ss_entity(doc, surface if surface.startswith("region:") else f"region:{surface}")
+            if e is not None:
+                out["resolved"].append({"entity_id": e["id"], "authority": name, "via": "identity"})
+        if not out["resolved"]:
+            out["unresolved_reason"] = "region not defined in any authority"
+    else:
+        out["unresolved_reason"] = f"kind {kind} has no deterministic resolver"
+    return out
+
+
+def ss_entity(doc: dict, eid: str):
+    from rune.structural_store import entity as _e
+
+    return _e(doc, eid)
+
+
+def implemented_contracts(doc: dict, entity_id: str) -> list[str]:
+    """Contracts an entity implements/exposes/provides (structural fact)."""
+    out = []
+    for r in relationships(doc, from_id=entity_id):
+        if r["type"] in {"implements", "exposes", "provides", "serves"}:
+            out.append(r["to"])
+    return sorted(out)
+
+
+def resolve_affected_surface(need: Need, authorities: dict[str, dict]) -> dict:
+    """Classify + resolve the WHOLE affected surface (Tranche 5 §3-5, §13).
+    Produces the propagated representation the generated contract carries:
+        affected_existing_entities  structural ids (resolved)
+        affected_paths              file-scope entries (mechanical scope)
+        affected_ui_surfaces        ui components / regions found
+        unresolved_surfaces         retained, never dropped
+    File scope != structural impact: a path stays a path for scope checking
+    even when it also resolves to a module entity (no authority duplication)."""
+    result = {
+        "affected_existing_entities": [],
+        "affected_paths": [],
+        "affected_ui_surfaces": [],
+        "affected_regions": [],
+        "workflow_config_surfaces": [],
+        "unresolved_surfaces": [],
+        "resolutions": [],
+    }
+    for surface in need.affected_surface:
+        kind = classify_surface(surface)
+        res = resolve_surface(surface, kind, authorities)
+        result["resolutions"].append(res)
+        if kind == "source_path":
+            result["affected_paths"].append(surface)
+        elif kind == "workflow_config":
+            result["workflow_config_surfaces"].append(surface)
+        elif kind == "ui_component":
+            result["affected_ui_surfaces"].append(surface)
+        elif kind == "layout_region":
+            result["affected_regions"].append(surface)
+        for r in res["resolved"]:
+            eid = r["entity_id"]
+            if eid not in result["affected_existing_entities"] and not eid.startswith("layout:"):
+                result["affected_existing_entities"].append(eid)
+        if res["unresolved_reason"] and not res["resolved"]:
+            result["unresolved_surfaces"].append({"surface": surface, "kind": kind, "reason": res["unresolved_reason"]})
+    for k in (
+        "affected_existing_entities",
+        "affected_paths",
+        "affected_ui_surfaces",
+        "affected_regions",
+        "workflow_config_surfaces",
+    ):
+        result[k] = sorted(set(result[k]))
+
+    # §9/§17 DERIVED EXPECTATION (structural consequence, not invention): a
+    # Rune test path (tests/test_<name>.py) affected by the work implies the
+    # "test:<name>" surface participates in the build. Derived only when not
+    # already an existing entity nor explicitly declared; provenance records
+    # the derivation for closure accounting.
+    derived = []
+    for surface in need.affected_surface:
+        if surface.startswith("tests/test_") and surface.endswith(".py"):
+            stem = surface[len("tests/test_") : -len(".py")]
+            test_entity = f"test:{stem}"
+            exists = any(ss_entity(doc, test_entity) is not None for doc in authorities.values())
+            if test_entity not in need.constraints.get("expected_new_entities", []) and not exists:
+                derived.append(
+                    {
+                        "entity": test_entity,
+                        "derived_from": surface,
+                        "rule": "Rune test-path convention: tests/test_<name>.py implies "
+                        "the test:<name> surface participates in the build",
+                        "evidence": f"affected_surface entry {surface}",
+                    }
+                )
+    result["derived_expected_new"] = derived
+    return result
+
+
+# ── requirement completeness gate (Tranche 5 §6-9) ──────────────────────────
+
+COMPLETE_ENOUGH = "COMPLETE_ENOUGH"
+MATERIAL_GAP = "MATERIAL_GAP"
+COMPLETENESS_UNKNOWN = "UNKNOWN"
+
+
+def check_completeness(
+    req: cf.CapabilityRequirement, need: Need, authorities: dict[str, dict], surface_result: dict
+) -> dict:
+    """Bounded, structure-grounded completeness check run BEFORE fit.
+
+    Rule (deterministic, no model vibes): every affected-surface entry that
+    resolves to a structural entity contributes that entity's IMPLEMENTED
+    contracts to the requirement's material scope. If an affected entity
+    implements contracts NOT represented in the normalized requirement, the
+    requirement is materially incomplete — the need touches a component
+    whose obligations it has not stated. That is exactly the deploy-incident
+    shape: the affected promotion implementation carried an unrepresented
+    ordering contract while the requirement named only qualification.
+
+    Verdicts (Tranche 5 §8):
+        COMPLETE_ENOUGH  all affected entities' contracts are represented
+                         (or none implement contracts) — fit may proceed
+        MATERIAL_GAP     uncovered implemented contracts exist on affected
+                         entities — named with provenance, routed to probe
+        UNKNOWN          no surface resolved — completeness undeterminable;
+                         never silently COMPLETE (§8)
+
+    Derived requirements (§9) are structural consequences, not inventions:
+    each uncovered contract carries derived_from (the entity that
+    implements it) + evidence (the implements relationship) + rule.
+    """
+    gaps = []
+    resolved_any = False
+    for res in surface_result["resolutions"]:
+        for r in res["resolved"]:
+            resolved_any = True
+            for name, doc in authorities.items():
+                e = ss_entity(doc, r["entity_id"])
+                if e is None:
+                    continue
+                for contract in implemented_contracts(doc, r["entity_id"]):
+                    if contract not in req.contracts:
+                        gaps.append(
+                            {
+                                "missing_contract": contract,
+                                "derived_from": r["entity_id"],
+                                "surface": res["surface"],
+                                "authority": name,
+                                "rule": "affected-surface entity implements a "
+                                "contract not represented in the "
+                                "normalized requirement",
+                                "evidence": f"{r['entity_id']} implements {contract} (authority: {name})",
+                            }
+                        )
+            break  # entity found in one authority is enough
+    if gaps:
+        # dedupe by (missing_contract, derived_from)
+        seen = set()
+        unique = []
+        for g in gaps:
+            key = (g["missing_contract"], g["derived_from"])
+            if key not in seen:
+                seen.add(key)
+                unique.append(g)
+        return {
+            "verdict": MATERIAL_GAP,
+            "gaps": sorted(unique, key=lambda g: (g["missing_contract"], g["derived_from"])),
+            "probe_questions": [
+                f"does the need's stated failure implicate {g['derived_from']}'s {g['missing_contract']} obligation?"
+                for g in unique[:5]
+            ],
+            "note": "MATERIAL_GAP: affected surface carries obligations "
+            "the requirement does not represent — refine before "
+            "trusting an apparently-complete fit",
+        }
+    if resolved_any:
+        return {
+            "verdict": COMPLETE_ENOUGH,
+            "gaps": [],
+            "probe_questions": [],
+            "note": "all resolved affected entities' implemented contracts are represented in the requirement",
+        }
+    return {
+        "verdict": COMPLETENESS_UNKNOWN,
+        "gaps": [],
+        "probe_questions": [f"which structural entities does '{s}' map to?" for s in need.affected_surface[:5]],
+        "note": "no affected surface resolved — completeness undeterminable; UNKNOWN never silently becomes COMPLETE",
+    }
+
+
+def record_proposed_requirement(
+    contract_id: str, *, provider: str, model: str, evidence: list[str], confidence: float
+) -> dict:
+    """Record a MODEL-PROPOSED requirement addition (Tranche 5 §12). It is
+    evidence with a verification state — never structural authority by
+    itself. It becomes accepted only via existing authority, research
+    evidence, implementation evidence, or explicit user decision; until
+    then normalize_need excludes it from the requirement's contract set."""
+    if not provider or not model:
+        raise ValueError("model-proposed requirement requires provider+model")
+    if not 0.0 <= confidence <= 1.0:
+        raise ValueError("confidence must be in [0,1]")
+    return {
+        "kind": "model_proposed_requirement",
+        "contract": contract_id,
+        "provider": provider,
+        "model": model,
+        "evidence": sorted(evidence),
+        "confidence": confidence,
+        "verification_state": "PENDING",
+    }
+
+
+# ── research routing seam (§6 — honest: recorded route, not execution) ─────
+
+
+def research_route(req: cf.CapabilityRequirement, decision: dict) -> dict:
+    """Record which evidence path the decision needs. This is a SEAM: the
+    live ADAPT interface is not yet wired for direct invocation, so the
+    route is recorded on the trace for the orchestrating session to act on.
+    NOT claimed as executed research."""
+    if decision["resolution"] == "INSUFFICIENT_EVIDENCE":
+        return {
+            "route": "TARGETED_EVIDENCE_PROBE",
+            "question": f"what satisfies: {req.objective[:120]}",
+            "status": "RECORDED_NOT_EXECUTED",
+        }
+    if decision["resolution"] == "ACQUIRE_EXTERNAL":
+        return {
+            "route": "FULL_RESEARCH_PIPELINE",
+            "question": f"external acquisition options for: {req.objective[:120]}",
+            "status": "RECORDED_NOT_EXECUTED",
+        }
+    return {"route": "NONE", "status": "LOCAL_SOLVE"}
+
+
+# ── resolution merge across authorities ─────────────────────────────────────
+
+_PRECEDENCE = [
+    "SATISFIED",
+    "REUSE",
+    "COMPOSE",
+    "ADAPT_EXISTING",
+    "CONFIGURE",
+    "ACQUIRE_EXTERNAL",
+    "INSUFFICIENT_EVIDENCE",
+    "BUILD_NEW",
+]
+
+
+def merge_resolutions(results: list[dict]) -> dict:
+    """Merge per-authority resolution results. Precedence: a concrete
+    reuse/compose/adapt verdict from any authority wins over
+    INSUFFICIENT_EVIDENCE from another; BUILD_NEW can only result when NO
+    authority produced a better verdict AND positive gap evidence exists —
+    it is never a fallback across merges (§8)."""
+    if not results:
+        return {"resolution": "INSUFFICIENT_EVIDENCE", "rationale": "no authority consulted"}
+    best = min(results, key=lambda r: _PRECEDENCE.index(r["resolution"]))
+    merged = dict(best)
+    merged["per_authority"] = [{"authority": r.get("authority", ""), "resolution": r["resolution"]} for r in results]
+    if best["resolution"] == "BUILD_NEW" and len(results) > 1:
+        # only honest if every authority independently agrees
+        if any(r["resolution"] != "BUILD_NEW" for r in results):
+            weaker = [r for r in results if r["resolution"] == "INSUFFICIENT_EVIDENCE"]
+            if weaker:
+                merged = dict(weaker[0])
+                merged["per_authority"] = [
+                    {"authority": r.get("authority", ""), "resolution": r["resolution"]} for r in results
+                ]
+                merged["rationale"] = (
+                    "BUILD_NEW rejected: at least one authority returned "
+                    "INSUFFICIENT_EVIDENCE — positive gap evidence required "
+                    "from all consulted authorities (§8)"
+                )
+    return merged
+
+
+# ── user-approval gate (§13) ────────────────────────────────────────────────
+
+
+def needs_user_approval(
+    resolution: str, toolchain_decision: tc.DecisionRecord, layout_obligations: list[dict], need: Need
+) -> dict:
+    """Classify whether decision→contract requires explicit user approval.
+    Small bounded work inside an already-approved contract does NOT re-ask
+    (need.approval_context covers it)."""
+    reasons = []
+    if toolchain_decision.user_approval_required:
+        reasons.append("toolchain change (CHANGE_MAJOR)")
+    if resolution == "BUILD_NEW" and need.requested_permanence != "temporary":
+        reasons.append("new capability construction")
+    if any(o["lock"] for o in layout_obligations):
+        reasons.append("USER-LOCKED layout surface affected — placement changes require user authorization")
+    if need.requested_permanence == "permanent":
+        reasons.append("permanent promotion requested (user-controlled)")
+    if need.constraints.get("new_repository") or need.constraints.get("new_application"):
+        reasons.append("new repository/application")
+    covered = bool(need.approval_context) and not reasons
+    if reasons and need.approval_context and need.constraints.get("approval_covers") == "all":
+        covered = True
+        reasons = [f"{r} — covered by: {need.approval_context}" for r in reasons]
+    return {
+        "required": bool(reasons) and not covered,
+        "reasons": reasons,
+        "covered_by": need.approval_context if covered else "",
+    }
+
+
+# ── the live decision path ──────────────────────────────────────────────────
+
+
+def _project_exists_evidence(project: str | None, authorities: dict[str, dict]) -> tuple[bool | None, list[str]]:
+    """T11 §1: project existence from STRUCTURAL EVIDENCE only.
+
+    Returns (exists, evidence) where exists is:
+        True   — a bound Project Scaffold or Rune Spine carries the project
+        False  — the project name is known but NO structural evidence for it
+        None   — project identity missing/unresolvable (UNKNOWN; classifier
+                 must not manufacture existence)
+
+    Rune itself exists (Rune Spine is its authority). Any other name must
+    appear as a project entity in a consulted authority; a bare string is
+    never proof (held-out: arbitrary "foo" → False, not silently True)."""
+    evidence: list[str] = []
+    if not project:
+        return None, ["no project identity supplied — existence UNKNOWN"]
+    if project in ("rune", "issue161-harness"):
+        evidence.append("Rune Spine is the Rune project authority (rune_internal mode)")
+        return True, evidence
+    for name, doc in authorities.items():
+        for e in doc.get("entities", []):
+            if e.get("kind") == "project" and e.get("id") == f"project:{project}":
+                evidence.append(f"bound Project Scaffold entity project:{project} in {name}")
+                return True, evidence
+    evidence.append(f"no structural evidence for project '{project}' in consulted authorities")
+    return False, evidence
+
+
+def make_decision(
+    need: Need,
+    *,
+    authorities: dict[str, dict],
+    toolchain_constraints: tc.ToolchainConstraints | None = None,
+    project_toolchain: dict | None = None,
+    prior_iterations: list[dict] | None = None,
+    extra_patterns: list | None = None,
+) -> dict:
+    """Run NEED → decision trace. `authorities` maps authority name → store
+    doc (e.g. {"rune_spine": spine_doc, "project_scaffold": scaffold_doc}).
+    `project_toolchain` seeds the Toolchain Resolver's existing-project
+    constraint (existing toolchain wins — §12).
+    `prior_iterations` (Tranche 5 §11): earlier decision iterations being
+    refined — retained in the trace with why-it-changed provenance; the
+    final resolution stays singular.
+
+    Order of operations (Tranche 5 §6): affected-surface resolution →
+    REQUIREMENT COMPLETENESS GATE → discovery/fit → merge. A MATERIAL_GAP
+    downgrades premature concrete verdicts (REUSE/SATISFIED/COMPOSE/
+    ADAPT_EXISTING) to INSUFFICIENT_EVIDENCE with the gap recorded — the
+    fit engine was right GIVEN its requirement; the requirement was too
+    narrow."""
+    req, missing = normalize_need(need)
+    trace: dict = {
+        "need": need.to_dict(),
+        "requirement": req.to_dict(),
+        "normalization_missing": missing,
+        "candidates": [],
+        "fits": [],
+        "per_authority_results": [],
+        "layout_obligations": [],
+        "research": {},
+        "toolchain": {},
+        "approval": {},
+        "resolution": "",
+        "rationale": "",
+        "iterations": list(prior_iterations or []),
+    }
+
+    # affected-surface classification + resolution (Tranche 5 §3-5)
+    surface_result = resolve_affected_surface(need, authorities)
+    trace["affected_surfaces"] = surface_result
+
+    # CONTEXTUAL CONSTRUCTION MODE (Tranche 10 + T11 §1): typed mode
+    # classification with evidence — ONE canonical home (construction_mode).
+    # §1 CORRECTION: project existence derives from STRUCTURAL EVIDENCE
+    # (bound Project Scaffold / Rune Spine), never from a non-"rune" string.
+    from rune.construction_mode import classify_mode
+
+    existence, existence_evidence = _project_exists_evidence(need.project, authorities)
+    mode_decision = classify_mode(
+        need.objective, target_project=need.project, project_exists=existence, affected_surface=need.affected_surface
+    )
+    mode_decision.evidence.extend(existence_evidence)
+    trace["construction_mode"] = mode_decision.to_dict()
+    trace["project_existence"] = {"project": need.project, "exists": existence, "evidence": existence_evidence}
+
+    # REQUIREMENT COMPLETENESS GATE (Tranche 5 §6-8) — before fit
+    completeness = check_completeness(req, need, authorities, surface_result)
+    trace["completeness"] = completeness
+
+    # live discovery + fit + resolution per authority (§7-8)
+    results = []
+    all_fits = []
+    for name, doc in authorities.items():
+        candidates = cf.discover_candidates(doc, req)
+        fits = [cf.evaluate_fit(req, c) for c in candidates]
+        for f, c in zip(fits, candidates):
+            f["authority"] = name
+            c["authority"] = name
+        all_fits.extend(fits)
+        trace["candidates"].extend(
+            {k: c[k] for k in ("id", "kind", "maturity", "status", "authority", "provides", "consumers")}
+            for c in candidates
+        )
+        res = cf.decide(doc, req, fits)
+        res["authority"] = name
+        results.append(res)
+    trace["fits"] = all_fits
+    if results:
+        merged = merge_resolutions(results)
+    else:
+        merged = {"resolution": "INSUFFICIENT_EVIDENCE", "rationale": "no authorities provided"}
+    if missing and merged["resolution"] in ("BUILD_NEW",):
+        # normalization gaps block BUILD_NEW too (§5: do not invent)
+        merged = {
+            "resolution": "INSUFFICIENT_EVIDENCE",
+            "rationale": f"requirement normalization incomplete: {missing}; refusing to build on guesses",
+        }
+    if completeness["verdict"] == MATERIAL_GAP and merged["resolution"] in (
+        "SATISFIED",
+        "REUSE",
+        "COMPOSE",
+        "ADAPT_EXISTING",
+        "CONFIGURE",
+    ):
+        # PREMATURE-FIT GUARD (Tranche 5 §6-7): the fit engine was correct
+        # GIVEN its requirement — but the requirement is materially
+        # incomplete (affected surface carries unrepresented obligations).
+        # A concrete verdict on an incomplete requirement is not trustworthy.
+        merged = {
+            "resolution": "INSUFFICIENT_EVIDENCE",
+            "rationale": (
+                "premature fit blocked: requirement is materially incomplete "
+                f"({len(completeness['gaps'])} unrepresented obligation(s) on "
+                "affected surfaces); refine the requirement before trusting "
+                "an apparently-complete fit"
+            ),
+            "premature_fit_blocked": {
+                "would_have_been": merged["resolution"],
+                "gaps": completeness["gaps"],
+            },
+        }
+    trace["resolution"] = merged["resolution"]
+    trace["rationale"] = merged.get("rationale", "")
+    if missing and trace["resolution"] == "INSUFFICIENT_EVIDENCE":
+        trace["rationale"] += f"; normalization gaps recorded, not invented: {missing}"
+    trace["per_authority_results"] = merged.get("per_authority", results)
+    trace["_merged"] = merged
+
+    # toolchain (§12): existing project toolchain wins
+    constraints = toolchain_constraints or tc.ToolchainConstraints(**(project_toolchain or {}))
+    tdec = tc.resolve_toolchain(constraints)
+    trace["toolchain"] = tdec.to_dict()
+
+    # layout obligations (§19-20): affected surface + candidate impls
+    affected = list(need.affected_surface)
+    for c in trace["candidates"]:
+        affected.append(c["id"])
+    lay = []
+    for name, doc in authorities.items():
+        lay.extend(layout_obligations_for(doc, affected))
+    trace["layout_obligations"] = lay
+
+    # research routing (§6, honest seam)
+    trace["research"] = research_route(req, merged)
+
+    # PATTERN SELECTION (Tranche 6 §4-7): evidence-backed applicability over
+    # the builtin registry + caller-supplied patterns; obligations filtered
+    # per project evidence; conflicts surfaced, never silently resolved.
+    # Project evidence > pattern assumption (§3): a pattern that contradicts
+    # recorded project exclusions becomes DOES_NOT_APPLY.
+    from rune.construction_pattern import (  # noqa: E402 — local: pattern layer
+        combine_pattern_obligations,
+        derive_pattern_obligations,
+        evaluate_applicability,
+    )
+    from rune.construction_patterns_builtin import BUILTIN_PATTERNS  # noqa: E402
+
+    pattern_candidates = list(BUILTIN_PATTERNS) + list(extra_patterns or [])
+    patterns_considered = []
+    derivations = []
+    for pattern in pattern_candidates:
+        app = evaluate_applicability(pattern, need, surface_result, authorities)
+        patterns_considered.append(app)
+        # APPLIES and PARTIAL both reach obligation derivation: applicability
+        # gates whether the pattern is CONSULTED; the per-obligation `when`
+        # conditions do the precision filtering (Tranche 6 §7). The
+        # applicability verdict is recorded on every derived obligation so
+        # weaker evidence stays visible.
+        if app["verdict"] in ("APPLIES", "PARTIAL"):
+            d = derive_pattern_obligations(
+                pattern,
+                need,
+                surface_result,
+                authorities,
+                project_exceptions=need.constraints.get("pattern_exceptions", []),
+            )
+            for entry in d["accepted"] + d["rejected"] + d["unresolved"]:
+                entry["pattern_applicability"] = app["verdict"]
+            derivations.append(d)
+    combined = combine_pattern_obligations(derivations) if derivations else {"obligations": [], "conflicts": []}
+    trace["patterns_considered"] = patterns_considered
+    trace["pattern_obligations"] = combined
+
+    # approval gate (§13)
+    trace["approval"] = needs_user_approval(merged["resolution"], tdec, lay, need)
+    trace["obligations"] = obligations_from_trace(trace, req)
+    # pattern-derived obligations join the contract's verification set (§19):
+    # provenance stays PATTERN (distinct from STRUCTURAL derivations in
+    # obligations_from_trace — §8 provenance separation)
+    trace["obligations"]["pattern"] = combined["obligations"]
+    trace["obligations"]["pattern_conflicts"] = combined["conflicts"]
+    for po in combined["obligations"]:
+        trace["obligations"]["verification"].append(f"[pattern:{po['pattern']}@r{po['revision']}] {po['obligation']}")
+    return trace
+
+
+def layout_obligations_for(doc: dict, affected: list[str]) -> list[dict]:
+    """Layout obligations from a store doc without a Scaffold instance
+    (delegates to the same authority logic via a lightweight query)."""
+    obligations = []
+    affected_set = set(affected)
+    for e in doc["entities"]:
+        if e["kind"] != "layout_contract" or e.get("status") == "RETIRED":
+            continue
+        ident = e["identities"]
+        comp, region, impl = (ident.get("component", ""), ident.get("region", ""), ident.get("implementation", ""))
+        hit = (
+            comp in affected
+            or f"region:{region}" in affected
+            or (impl and any(a in impl for a in affected if a.endswith(".py")))
+        )
+        if not hit:
+            continue
+        locked = bool(ident.get("lock"))
+        obligations.append(
+            {
+                "component": comp,
+                "region": region,
+                "lock": locked,
+                "provenance": ident.get("provenance", "scaffold"),
+                "obligation": (
+                    "USER-LOCKED: preserve user placement in region "
+                    "'%s'; changes require explicit user authorization" % region
+                )
+                if locked
+                else (
+                    "preserve/verify %s placement in region '%s'; movable "
+                    "only via authorized layout amendment" % (comp, region)
+                ),
+                "verification": ("layout_contract_check must PASS; USER-LOCKED placement unchanged")
+                if locked
+                else "layout_contract_check must PASS",
+                "status": e.get("status", "ACTIVE"),
+            }
+        )
+    return obligations
+
+
+def obligations_from_trace(trace: dict, req: cf.CapabilityRequirement) -> dict:
+    """BUILD-BUT-DON'T-BREAK (§9) + composition/acquire obligations (§10-11)
+    derived automatically from the trace — the builder does not memorize the
+    rule; the path produces it."""
+    resolution = trace["resolution"]
+    obligations: dict = {
+        "new_behavior": [],
+        "regression": [],
+        "composition": [],
+        "acquisition": [],
+        "layout": trace["layout_obligations"],
+        "verification": [],
+    }
+    merged = trace.get("_merged", {})
+    selected = merged.get("selected") or merged.get("components") or []
+    if isinstance(selected, str):
+        selected = [selected]
+    if resolution in ("ADAPT_EXISTING", "REUSE", "SATISFIED") and selected:
+        # regression obligations from existing consumers + contracts (§9)
+        for c in trace["candidates"]:
+            if c["id"] in selected and c.get("consumers"):
+                obligations["regression"].append(
+                    f"preserve existing behavior of {c['id']} for consumers: {', '.join(c['consumers'])}"
+                )
+        obligations["verification"].append("new behavior proven by tests before closure")
+    if resolution == "ADAPT_EXISTING":
+        obligations["verification"].append("regression proof: original capability's consumers still work")
+    if resolution == "COMPOSE":
+        obligations["composition"] = [
+            {
+                "components": merged.get("components", selected),
+                "joining_contracts": req.contracts,
+                "permanence": merged.get("permanence", "temporary (user-controlled)"),
+            }
+        ]
+        obligations["verification"].append("composition boundary behavior verified end-to-end")
+    if resolution == "ACQUIRE_EXTERNAL":
+        obligations["acquisition"] = merged.get("external", [])
+        obligations["verification"].append("external acquisition requires research/approval step before install")
+    if resolution == "BUILD_NEW":
+        obligations["verification"].append("positive gap evidence recorded in contract")
+    for lo in trace["layout_obligations"]:
+        obligations["verification"].append(lo["verification"])
+    return obligations
+
+
+# ── contract generation (§15) ────────────────────────────────────────────────
+
+
+def generate_contract(
+    s, doc: dict, trace: dict, *, contract_id: str, scope: list[str] | None = None, proposer_note: str = ""
+) -> dict:
+    """Populate a Construction Contract from the decision trace (Tranche 5
+    §13-17 IMPACT-AWARE generation). The Need's affected surface is never
+    lost — it is classified and propagated into distinct surfaces:
+
+        affected_entities        structural entity ids (existing)
+        expected_new_entities    planned new structural entities
+        ext.affected_paths       file-scope entries (mechanical scope check)
+        ext.affected_ui_surfaces ui components (layout obligations follow)
+        ext.unresolved_surfaces  retained, never dropped
+        ext.expected_runtime_effects  operational effects (deployment:current
+                                 advances etc.) — accounting metadata, NOT a
+                                 workflow engine; closure requires independent
+                                 observation before marking them observed.
+
+    File scope: generated from affected paths + selected capability
+    implementation paths + explicit expected-new paths — NOT unrestricted
+    repo scope. `scope` param may still extend it; generated paths are
+    recorded as such (provenance: affected-surface).
+    """
+    merged = trace["_merged"]
+    req = trace["requirement"]
+    surfaces = trace.get("affected_surfaces", {})
+    selected = merged.get("selected") or []
+    if isinstance(selected, str):
+        selected = [selected]
+
+    # structural affected: selected candidates + requirement contracts +
+    # resolved surface entities — the union the old generator partially lost
+    affected = {c["id"] for c in trace["candidates"] if c["id"] in selected}
+    affected |= set(trace["requirement"].get("contracts", []))
+    affected |= set(surfaces.get("affected_existing_entities", []))
+
+    # file scope: generated (affected paths + selected impl paths + expected
+    # new paths) UNION caller-supplied additions, provenance-tracked
+    affected_paths = list(surfaces.get("affected_paths", []))
+    impl_paths = []
+    for c in trace["candidates"]:
+        if c["id"] in selected:
+            p = c.get("identities", {}).get("path", "")
+            if p:
+                impl_paths.append(p)
+    expected_new_paths = trace["need"]["constraints"].get("expected_new_source_paths", [])
+    generated_scope = sorted(
+        set(affected_paths)
+        | set(surfaces.get("workflow_config_surfaces", []))
+        | set(impl_paths)
+        | set(expected_new_paths)
+    )
+    caller_scope = sorted(set(scope or []) - set(generated_scope))
+    final_scope = sorted(set(generated_scope) | set(caller_scope))
+
+    expected_runtime = sorted(trace["need"]["constraints"].get("expected_runtime_effects", []))
+
+    obl = trace["obligations"]
+    s.new_contract(
+        doc,
+        contract_id=contract_id,
+        need=trace["need"]["objective"],
+        origin=trace["need"]["origin"],
+        scope=final_scope,
+        requirement=req.get("objective", ""),
+        gap_evidence=merged.get("rationale", ""),
+        research_refs=(
+            trace["need"]["evidence_refs"]
+            + ([trace["research"]["route"]] if trace["research"]["route"] != "NONE" else [])
+        ),
+        resolution=trace["resolution"],
+        toolchain=trace["toolchain"],
+        affected_entities=sorted(affected),
+        expected_new_entities=sorted(
+            set(trace["need"]["constraints"].get("expected_new_entities", []))
+            | {d["entity"] for d in surfaces.get("derived_expected_new", [])}
+        ),
+        expected_consumers=[],
+        regression_obligations=obl["regression"],
+        verification_obligations=sorted(set(obl["verification"])),
+        rollback_ref=trace["need"]["constraints"].get("rollback_ref", "git revert"),
+        acceptance=[f"resolution {trace['resolution']} verified end-to-end"],
+        non_goals=trace["need"]["constraints"].get("non_goals", []),
+    )
+    e = s.contract(doc, contract_id)
+    e.setdefault("ext", {}).update(
+        {
+            "decision_trace": {
+                k: trace[k]
+                for k in (
+                    "need",
+                    "requirement",
+                    "candidates",
+                    "fits",
+                    "resolution",
+                    "rationale",
+                    "research",
+                    "toolchain",
+                    "approval",
+                    "obligations",
+                    "layout_obligations",
+                    "normalization_missing",
+                    "completeness",
+                    "affected_surfaces",
+                    "iterations",
+                    "patterns_considered",
+                    "pattern_obligations",
+                    "construction_mode",
+                )
+            },
+            # distinct impact surfaces (Tranche 5 §13) — file scope != structural
+            # impact != expected new != runtime effect
+            "affected_paths": affected_paths,
+            "affected_ui_surfaces": surfaces.get("affected_ui_surfaces", []),
+            "affected_regions": surfaces.get("affected_regions", []),
+            "workflow_config_surfaces": surfaces.get("workflow_config_surfaces", []),
+            "unresolved_surfaces": surfaces.get("unresolved_surfaces", []),
+            "expected_runtime_effects": expected_runtime,
+            "scope_provenance": {"generated_from_affected_surface": generated_scope, "caller_added": caller_scope},
+        }
+    )
+    if proposer_note:
+        e["ext"]["decision_trace"]["proposer_note"] = proposer_note
+    return e
@@ -0,0 +1,138 @@
+"""construction_depth.py — proportional Scaffolding depth selection.
+
+SCAFFOLD-TRANCHE-0012 §12E. ONE construction discipline, variable depth.
+"Small" never means "bypass governance"; a one-line typo never creates
+six meaningless artefacts.
+
+Depth is selected from EVIDENCE, not from keywords alone. The evidence
+inputs (§12E): construction mode, novelty, persistence, risk, affected
+surface, architecture impact, migration, deployment impact, user-locked
+structure, uncertainty.
+
+Levels:
+    LIGHT     classification + authority/scope + minimal obligations +
+              focused verification. For tiny bounded local fixes.
+    STANDARD  + requirements/completeness + capability/toolchain check
+              + broader verification. For features/repairs/migrations.
+    FULL      + project bootstrap + architecture/toolchain decisions +
+              patterns + deployment/rollback. For PROJECT_BUILD.
+
+Reuses existing semantics: construction mode (T10/T11) already encodes
+most of the signal — depth is DERIVED from mode + modifiers, not a
+parallel classification.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+__all__ = ["Depth", "select_depth", "LEVELS"]
+
+LEVELS = ("NONE", "UNRESOLVED", "LIGHT", "STANDARD", "FULL")
+
+
+@dataclass
+class Depth:
+    level: str
+    reason: str
+    evidence: list[str] = field(default_factory=list)
+    ceremony: list[str] = field(default_factory=list)  # what depth requires
+    bypassed: list[str] = field(default_factory=list)  # what is skipped and why
+
+    def to_dict(self) -> dict:
+        return {"level": self.level, "reason": self.reason,
+                "evidence": self.evidence, "ceremony": self.ceremony,
+                "bypassed": self.bypassed}
+
+
+def select_depth(mode: str, *, objective: str = "",
+                 affected_surface: list[str] | None = None,
+                 has_migration: bool = False,
+                 breaking: bool = False,
+                 user_locked_layout: bool = False,
+                 external_service: bool = False,
+                 uncertainty: str = "",  # "" | "partial" | "high"
+                 project_exists: bool | None = None,
+                 ) -> Depth:
+    """Select proportional depth from construction mode + evidence."""
+    ev: list[str] = [f"mode: {mode}"]
+    ceremony: list[str] = []
+    bypassed: list[str] = []
+
+    if mode == "NOT_CONSTRUCTION":
+        # §13C: NOT_CONSTRUCTION has NO construction depth — not even LIGHT
+        return Depth(level="NONE", reason="not construction — no construction "
+                     "depth applies", evidence=ev, ceremony=[],
+                     bypassed=["all"])
+
+    if mode == "UNKNOWN":
+        # §13C: UNKNOWN means construction hasn't started — no lifecycle yet
+        return Depth(level="UNRESOLVED",
+                     reason="UNKNOWN — probe/clarify before any construction "
+                            "lifecycle begins",
+                     evidence=ev, ceremony=["probe/clarify"],
+                     bypassed=["contract", "patterns", "bootstrap",
+                               "depth selection"])
+
+    if mode == "LOCAL_CHANGE":
+        ev.append("LOCAL_CHANGE: bounded scope, minimal ceremony")
+        ceremony.extend(["classification", "authority/scope",
+                         "focused verification", "durable result"])
+        bypassed.extend(["requirements/completeness", "patterns",
+                         "capability/toolchain check", "project bootstrap",
+                         "deployment/rollback"])
+        if user_locked_layout:
+            ev.append("USER-LOCKED layout surface — preservation obligation "
+                      "retained even at LIGHT depth")
+            ceremony.append("layout obligation check")
+        return Depth(level="LIGHT",
+                     reason="bounded local fix — governance without ceremony",
+                     evidence=ev, ceremony=ceremony, bypassed=bypassed)
+
+    if mode == "PROJECT_BUILD":
+        ev.append("PROJECT_BUILD: full lifecycle required")
+        ceremony.extend(["project bootstrap", "requirements/completeness",
+                         "capability/toolchain decisions",
+                         "patterns/obligations", "Construction Contract",
+                         "implementation", "verification",
+                         "deployment/rollback strategy",
+                         "maintenance lifecycle"])
+        return Depth(level="FULL",
+                     reason="genuinely new project — full Scaffolding "
+                            "lifecycle",
+                     evidence=ev, ceremony=ceremony)
+
+    # EXISTING_PROJECT_CHANGE: base STANDARD, escalate on modifiers
+    ceremony.extend(["classification", "requirements/completeness",
+                     "capability fit", "Construction Contract", "scope",
+                     "obligations", "verification"])
+    level = "STANDARD"
+    reason = "existing-project change — proportional governance"
+
+    if has_migration:
+        ev.append("migration: old/new compatibility + rollback obligations")
+        ceremony.extend(["migration proof", "rollback preparation"])
+    if breaking:
+        ev.append("breaking change: stronger approval + regression "
+                  "obligations")
+        ceremony.append("consumer migration completeness")
+    if user_locked_layout:
+        ev.append("USER-LOCKED layout surface affected")
+        ceremony.append("layout obligation check")
+    if external_service:
+        ev.append("external service consequence")
+        ceremony.append("deployment verification")
+
+    if uncertainty == "high":
+        ev.append("high uncertainty: probe before deeper ceremony")
+        ceremony.insert(0, "evidence probe")
+    surface_list = affected_surface or []
+    if len(surface_list) > 5:
+        ev.append(f"wide affected surface: {len(surface_list)} entries")
+        level = "FULL"
+        reason = "wide surface — full depth warranted"
+    if breaking and has_migration:
+        level = "FULL"
+        reason = "breaking + migration — full depth"
+
+    return Depth(level=level, reason=reason, evidence=ev, ceremony=ceremony,
+                 bypassed=bypassed)
@@ -0,0 +1,555 @@
+"""construction_intent.py — T17 §17J-§17N: the conversational
+interpretation layer of contextual construction.
+
+AUTHORITY BOUNDARY (§17O):
+
+    THIS LAYER owns: meaning, referent resolution, target resolution,
+    action-vs-discussion, conversational authorization, user constraints.
+
+    SCAFFOLDING (construction_mode.classify_mode) owns: construction
+    classification, depth, contracts, scope, lifecycle.
+
+This module NEVER deepens or reclassifies — it resolves what the user
+MEANT and whether they AUTHORIZED it, then hands a structured
+ConstructionIntent to Scaffolding, which runs the ONE canonical
+classifier (classify_mode). No second classifier lives here.
+
+Not a phrase table: explicit demands are recognized by imperative
+grammar + construction verbs; questions are recognized by
+interrogative structure; referents are resolved against the live
+conversation context (pending proposals, recency, topic drift).
+Paraphrases resolve because the criteria are structural, not lexical
+incantations.
+"""
+from __future__ import annotations
+
+import re
+import time
+import uuid
+from dataclasses import dataclass, field
+
+# ── conversation context (the layer's memory, NOT Scaffolding state) ────
+
+@dataclass
+class Proposal:
+    """A pending construction proposal awaiting user authorization.
+    Rune suggesting a change is NOT authorization to mutate (§17L)."""
+    proposal_id: str
+    summary: str
+    created_turn: int
+    superseded: bool = False
+
+
+@dataclass
+class ConversationContext:
+    """Active conversational/project context for referent resolution.
+    Recency-aware, proposal-aware, stale-context-resistant (§17M)."""
+    turn: int = 0
+    recent_turns: list = field(default_factory=list)  # [{role, text}]
+    pending_proposals: list = field(default_factory=list)  # [Proposal]
+    active_project: str | None = None
+    active_task_id: str | None = None
+
+    def next_turn(self) -> int:
+        self.turn += 1
+        return self.turn
+
+    def record_turn(self, role: str, text: str) -> None:
+        self.recent_turns.append({"role": role, "text": text,
+                                  "turn": self.turn})
+        self.turn += 1  # each recorded exchange advances the context
+
+    def add_proposal(self, summary: str) -> Proposal:
+        p = Proposal(proposal_id=f"prop-{uuid.uuid4().hex[:8]}",
+                     summary=summary,
+                     created_turn=self.turn)
+        self.pending_proposals.append(p)
+        return p
+
+    def supersede_proposal(self, proposal_id: str) -> None:
+        for p in self.pending_proposals:
+            if p.proposal_id == proposal_id:
+                p.superseded = True
+
+    def live_proposals(self) -> list:
+        return [p for p in self.pending_proposals if not p.superseded]
+
+
+# ── the structured intent (§17N) ─────────────────────────────────────────
+
+APPROVAL_STATES = ("DISCUSSION", "PROPOSAL", "USER_APPROVED",
+                   "CONSTRUCTION_STARTED")
+
+# approval-state lifecycle legality (§17L): authorization may only be
+# granted from PROPOSAL or DISCUSSION (direct demand); CONSTRUCTION_STARTED
+# and mutations are Scaffolding's side of the boundary.
+LEGAL_APPROVAL_TRANSITIONS = {
+    "DISCUSSION": {"PROPOSAL", "USER_APPROVED"},
+    "PROPOSAL": {"USER_APPROVED"},
+    "USER_APPROVED": {"CONSTRUCTION_STARTED"},
+    "CONSTRUCTION_STARTED": set(),
+}
+
+
+@dataclass
+class ConstructionIntent:
+    """Structured meaning-resolution output (§17N). Scaffolding decides
+    mode/depth from this; this layer never does."""
+    intent_id: str
+    objective: str
+    requested_effect: str
+    resolved_target: str | None
+    resolved_referent: str | None
+    origin_context_id: str | None
+    action_intent: str  # ACTION | DISCUSSION
+    approval_state: str  # DISCUSSION | PROPOSAL | USER_APPROVED
+    confidence: float
+    ambiguity: str  # NONE | REFERENT | TARGET | SCOPE
+    consequence_level: str  # LOW | MEDIUM | HIGH
+    relevant_constraints: list = field(default_factory=list)
+    classification_evidence: list = field(default_factory=list)
+    probe_questions: list = field(default_factory=list)
+
+    def to_dict(self) -> dict:
+        return {
+            "intent_id": self.intent_id,
+            "objective": self.objective,
+            "requested_effect": self.requested_effect,
+            "resolved_target": self.resolved_target,
+            "resolved_referent": self.resolved_referent,
+            "origin_context_id": self.origin_context_id,
+            "action_intent": self.action_intent,
+            "approval_state": self.approval_state,
+            "confidence": self.confidence,
+            "ambiguity": self.ambiguity,
+            "consequence_level": self.consequence_level,
+            "relevant_constraints": self.relevant_constraints,
+            "classification_evidence": self.classification_evidence,
+            "probe_questions": self.probe_questions,
+        }
+
+
+# ── structural lexical criteria (grammar classes, not incantations) ─────
+
+# §17M stale-context resistance: a proposal older than this many turns
+# is stale unless the user's text explicitly references its content.
+STALE_GAP_TURNS = 2
+
+_NEW_ARTICLE = re.compile(r"\b(me a|me an|a|an|another)\b")
+_CREATION_VERBS = {"build", "create", "make", "write", "develop"}
+
+_CONSTRUCTION_VERBS = {
+    "build", "implement", "add", "replace", "fix", "create", "make",
+    "write", "port", "migrate", "refactor", "integrate", "remove",
+    "extend", "wire", "ship",
+}
+_DEMONSTRATIVES = {"that", "this", "it", "those", "these"}
+_OPTION_REFERENTS = re.compile(
+    r"\b(first|second|third|other|latter|former)\s+(option|approach|one)\b")
+_QUESTION_MARKERS = re.compile(
+    r"^\s*(how|what|why|when|where|who|which|should|could|would|can|"
+    r"is|are|did|do you|do we|do they|does it|does this|does that)\b|"
+    r"\?\s*$", re.IGNORECASE)
+_ANALYSIS_MARKERS = re.compile(
+    r"\b(think|opinion|view|recommend|suggest|assess|evaluate|review|"
+    r"compare|consider|worth it|worth it\?|needs? improvement|any "
+    r"ideas?)\b", re.IGNORECASE)
+# relative-'that' detection by PRECEDING context (T18 18N fix): 'that'
+# is a relative pronoun when the word before it is an article/noun
+# ("the proposal that you suggested") — demonstrative when
+# sentence-initial or after a verb/punctuation ("build that" /
+# "that works for me"). The old following-word heuristic misclassified
+# "that works for me" as a relative clause.
+_ARTICLES = {"the", "a", "an", "this", "these", "those", "one"}
+
+
+def _is_relative_that(low: str, dem_tokens: set) -> bool:
+    """True when every 'that' token reads as a relative pronoun by its
+    preceding context; demonstrative otherwise."""
+    words = re.findall(r"[a-z']+", low)
+    seen_that = False
+    for i, w in enumerate(words):
+        if w != "that":
+            continue
+        seen_that = True
+        prev = words[i - 1] if i > 0 else None
+        if prev is None or prev in _CONSTRUCTION_VERBS:
+            return False  # sentence-initial or post-verb → demonstrative
+        if prev not in _ARTICLES:
+            # preceding word is neither article nor verb: treat as
+            # demonstrative unless it is clearly a noun modifier —
+            # conservative default is demonstrative (referent-seeking)
+            return False
+    return seen_that and "that" in dem_tokens
+
+_GO_AHEAD = re.compile(
+    r"^\s*(yes|yeah|yep|ok|okay|sure|go ahead|please do|do it|do that|"
+    r"carry on|proceed|approved|lgtm|ship it)\b[\s.,!]*$", re.IGNORECASE)
+_AFFIRMATION_OPENERS = re.compile(
+    r"^\s*(yes|yeah|yep|yup|ok|okay|sure|agreed|absolutely|indeed|"
+    r"alright|right)\b", re.IGNORECASE)
+_APPROVAL_IDIOM = re.compile(
+    r"\b(go ahead|do it|do that|carry on|build that|implement that|"
+    r"make it so|make it happen|make that happen|that's the one|go with|"
+    r"let's go|let's do|let's have it|proceed|good with that|"
+    r"fine with that|happy with that|on board|approved|ship it)\b",
+    re.IGNORECASE)
+
+
+def _tokens(text: str) -> set[str]:
+    return set(re.split(r"[^a-z0-9]+", text.lower()))
+
+
+_NEGATION = re.compile(
+    r"^\s*(don't|do not|dont|never|not yet|stop|hold off|no need|"
+    r"skip it|forget it)\b", re.IGNORECASE)
+_IDIOM_PREFIX = re.compile(
+    r"^\s*(go ahead and|go and|please|kindly|then|now)\s+", re.IGNORECASE)
+
+
+def _is_question(text: str) -> bool:
+    """Interrogative STRUCTURE, not vocabulary: leading wh/auxiliary or
+    terminal '?'. 'How would you build this?' is a question regardless
+    of the construction verb inside it (§17K)."""
+    if "?" in text and not re.match(r"^\s*(build|implement|add|fix|make|"
+                                    r"create|write|replace|remove)\b",
+                                    text, re.IGNORECASE):
+        return True
+    return bool(_QUESTION_MARKERS.match(text))
+
+
+def _imperative_verb(text: str) -> str | None:
+    """Leading imperative construction verb — the strong/direct demand
+    signal (§17K). 'Build me a program...' → 'build'. Idiom prefixes
+    ('go ahead and build X', 'please implement X') are stripped so the
+    real verb surfaces."""
+    t = text
+    for _ in range(3):
+        t2 = _IDIOM_PREFIX.sub("", t)
+        if t2 == t:
+            break
+        t = t2
+    m = re.match(r"^\s*(\w+)", t, re.IGNORECASE)
+    if m and m.group(1).lower() in _CONSTRUCTION_VERBS:
+        return m.group(1).lower()
+    return None
+
+
+def resolve_construction_intent(
+    text: str,
+    context: ConversationContext,
+    *,
+    origin_context_id: str | None = None,
+) -> ConstructionIntent:
+    """Resolve one user turn into a structured ConstructionIntent.
+
+    Criteria (§17J), in precedence order:
+      1. interrogative structure → DISCUSSION (never mutation); analysis
+         of an improvement question may RECORD a proposal for later.
+      2. bare approval ("go ahead") or demonstrative authorization
+         ("build that") → resolve referent against LIVE proposals,
+         recency-first; multiple equally plausible → clarification;
+         none live → clarification (stale-context resistant).
+      3. explicit imperative with resolved target → USER_APPROVED
+         direct demand — no unnecessary re-asking (§17K).
+      4. everything else → DISCUSSION; construction language without
+         resolved target → ambiguity flagged, never guessed (§17M).
+    """
+    context.next_turn()
+    context.record_turn("user", text)
+    t = text.strip()
+    low = t.lower()
+    ev: list[str] = []
+    probes: list[str] = []
+
+    # 0. NEGATION (§18N): "don't build that yet" blocks authorization —
+    # construction language inside a negation frame is NOT a demand.
+    if _NEGATION.match(t):
+        ev.append("negation frame — explicitly blocks construction")
+        return _intent(t, "DISCUSSION", "DISCUSSION", 0.85, "NONE",
+                       ev, probes, context, origin_context_id)
+
+    # 1a. REQUEST MODALS: "can you fix that?" is an imperative in
+    # interrogative clothing — a state-changing action request, NOT an
+    # information question. Information openers (how/what/why/should we)
+    # stay discussion; request modals authorize when the target resolves.
+    # information openers outrank modals: "how would you build X?" asks
+    # about METHOD; "can you fix that?" requests ACTION
+    information_opener = bool(re.match(
+        r"^\s*(how|what|why|when|where|which)\b", low))
+    request_modal = (not information_opener) and bool(re.search(
+        r"\b(can you|could you|would you|will you|please)\b", low))
+    if request_modal:
+        # modal + a construction verb somewhere in the turn
+        request_modal = bool(_tokens(low) & _CONSTRUCTION_VERBS)
+    if request_modal:
+        ev.append("request modal (can/could/would you, please) + "
+                  "construction verb → action request, not discussion")
+        # a modal + demonstrative + live proposal authorizes THAT
+        # proposal (referent resolution outranks target resolution)
+        dem_tokens_m = _tokens(low) & _DEMONSTRATIVES
+        if dem_tokens_m and context.live_proposals():
+            referent = _resolve_referent(low, context)
+            if referent and not referent.get("ambiguous"):
+                tp = referent["proposal"]
+                ev.append(f"modal authorization resolves proposal "
+                          f"{tp.proposal_id}")
+                return ConstructionIntent(
+                    intent_id=f"intent-{uuid.uuid4().hex[:8]}",
+                    objective=tp.summary,
+                    requested_effect=f"construct: {tp.summary}",
+                    resolved_target=context.active_project,
+                    resolved_referent=tp.proposal_id,
+                    origin_context_id=origin_context_id,
+                    action_intent="ACTION", approval_state="USER_APPROVED",
+                    confidence=0.85, ambiguity="NONE",
+                    consequence_level=_consequence(t, context),
+                    relevant_constraints=_constraints(context),
+                    classification_evidence=ev)
+        target = context.active_project
+        if target:
+            ev.append(f"target resolved from active context: {target}")
+            # demonstrative referent: "fix that" with no live proposal —
+            # the antecedent is the defect described in THIS turn
+            return ConstructionIntent(
+                intent_id=f"intent-{uuid.uuid4().hex[:8]}",
+                objective=t, requested_effect=f"fix/act: {t}",
+                resolved_target=target,
+                resolved_referent="this-turn-topic",
+                origin_context_id=origin_context_id,
+                action_intent="ACTION", approval_state="USER_APPROVED",
+                confidence=0.85, ambiguity="NONE",
+                consequence_level=_consequence(t, context),
+                relevant_constraints=_constraints(context),
+                classification_evidence=ev)
+        ev.append("request modal but no resolvable target")
+        probes.append("where should that change land — which project?")
+        return _intent(t, "ACTION", "DISCUSSION", 0.5, "TARGET",
+                       ev, probes, context, origin_context_id)
+
+    # 1. questions are discussion — even with construction verbs inside
+    if _is_question(t):
+        ev.append("interrogative structure → discussion, not authorization")
+        analysis = bool(_ANALYSIS_MARKERS.search(t))
+        if analysis:
+            ev.append("analysis/advice request — a proposal may be recorded "
+                      "for later authorization")
+        return ConstructionIntent(
+            intent_id=f"intent-{uuid.uuid4().hex[:8]}",
+            objective=t, requested_effect="analysis/discussion",
+            resolved_target=context.active_project,
+            resolved_referent=None, origin_context_id=origin_context_id,
+            action_intent="DISCUSSION", approval_state="DISCUSSION",
+            confidence=0.85, ambiguity="NONE",
+            consequence_level="LOW", classification_evidence=ev)
+
+    # 2. authorization of a prior proposal (bare approval or demonstrative)
+    bare_approval = bool(_GO_AHEAD.match(t))
+    imperative = _imperative_verb(t)
+    dem_tokens = _tokens(low) & _DEMONSTRATIVES
+    if "that" in dem_tokens and _is_relative_that(low, dem_tokens):
+        dem_tokens.discard("that")  # relative clause, not a demonstrative
+        ev.append("'that' is a relative pronoun here, not a referent")
+    demonstrative = bool(dem_tokens) or \
+        bool(_OPTION_REFERENTS.search(low))
+    option_ref = bool(_OPTION_REFERENTS.search(low))
+    approval_idiom = bool(_APPROVAL_IDIOM.search(low))
+    affirmation = bool(_AFFIRMATION_OPENERS.match(t))
+    authorized = (bare_approval or approval_idiom or
+                  (affirmation and (demonstrative or option_ref)) or
+                  (imperative and (demonstrative or option_ref)) or
+                  (option_ref and imperative is not None))
+
+    if authorized:
+        referent = _resolve_referent(low, context)
+        if referent is None:
+            # "go ahead and build X": idiom + explicit imperative — fall
+            # through to the demand path (target resolution) rather than
+            # dead-ending in referent ambiguity
+            if imperative is None:
+                ev.append("authorization language but NO live resolvable "
+                          "referent — never guessed")
+                probes.append("what should that refer to? I don't have an "
+                              "active proposal matching it.")
+                return _intent(t, "ACTION", "DISCUSSION", 0.4, "REFERENT",
+                               ev, probes, context, origin_context_id)
+            ev.append("no live proposal matches; treating as direct "
+                      "demand (explicit imperative object present)")
+        if referent is None:
+            # live proposals exist but are ALL STALE (topic moved on) —
+            # never approve a stale referent; fall to the demand path
+            ev.append("proposals stale for this turn — not selected")
+            if imperative is None:
+                probes.append("what should that refer to? The earlier "
+                              "proposal no longer seems active.")
+                return _intent(t, "ACTION", "DISCUSSION", 0.4, "REFERENT",
+                               ev, probes, context, origin_context_id)
+            ev.append("stale proposals ignored; explicit imperative → "
+                      "demand path")
+            referent = {"__fallthrough__": True}
+        if referent.get("ambiguous"):
+            ev.append("multiple equally plausible referents — asking, "
+                      "not guessing (§17M)")
+            probes.append("which one — " + " or ".join(
+                p.summary[:40] for p in referent["candidates"]) + "?")
+            return _intent(t, "ACTION", "DISCUSSION", 0.4, "REFERENT",
+                           ev, probes, context, origin_context_id)
+        if referent.get("__fallthrough__"):
+            # demand path: target from context or clarify (never stale)
+            target = context.active_project
+            if not target and imperative in _CREATION_VERBS and \
+                    _NEW_ARTICLE.search(low):
+                # new standalone artifact demand — no target needed
+                ev.append("creation verb + indefinite article → new "
+                          "standalone artifact; no existing target required")
+                return _intent(t, "ACTION", "USER_APPROVED", 0.85, "NONE",
+                               ev, probes, context, origin_context_id,
+                               requested_effect=f"construct new: {t}",
+                               consequence_level=_consequence(t, context))
+            if target:
+                ev.append(f"target resolved from active context: {target}")
+                return _intent(t, "ACTION", "USER_APPROVED", 0.8, "NONE",
+                               ev, probes, context, origin_context_id,
+                               requested_effect=f"construct: {t}",
+                               consequence_level=_consequence(t, context))
+            probes.append("build where — which project/system?")
+            return _intent(t, "ACTION", "DISCUSSION", 0.5, "TARGET",
+                           ev, probes, context, origin_context_id)
+        target_prop = referent["proposal"]
+        ev.append(f"referent resolved: {target_prop.proposal_id} "
+                  f"(recency-first among live proposals)")
+        ev.append("prior proposal + explicit authorization → USER_APPROVED")
+        return ConstructionIntent(
+            intent_id=f"intent-{uuid.uuid4().hex[:8]}",
+            objective=target_prop.summary,
+            requested_effect=f"construct: {target_prop.summary}",
+            resolved_target=context.active_project,
+            resolved_referent=target_prop.proposal_id,
+            origin_context_id=origin_context_id,
+            action_intent="ACTION", approval_state="USER_APPROVED",
+            confidence=0.9, ambiguity="NONE",
+            consequence_level=_consequence(t, context),
+            relevant_constraints=_constraints(context),
+            classification_evidence=ev)
+
+    # 3. explicit imperative construction demand with resolved target
+    if imperative:
+        # non-constructible object check is Scaffolding's classifier's
+        # job (§17O) — this layer records the demand structure only
+        ev.append(f"imperative construction verb: {imperative} — strong/"
+                  f"direct demand signal (§17K)")
+        target = context.active_project
+        if target:
+            ev.append(f"target resolved from active context: {target}")
+            return _intent(t, "ACTION", "USER_APPROVED", 0.9, "NONE",
+                           ev, probes, context, origin_context_id,
+                           requested_effect=f"construct: {t}",
+                           consequence_level=_consequence(t, context))
+        # no active project: a CREATION verb + indefinite article signals
+        # a NEW standalone artifact — the demand is complete without an
+        # existing target ("build me a program that...")
+        if imperative in _CREATION_VERBS and _NEW_ARTICLE.search(low):
+            ev.append("creation verb + indefinite article → new standalone "
+                      "artifact demand; no existing target required")
+            return _intent(t, "ACTION", "USER_APPROVED", 0.85, "NONE",
+                           ev, probes, context, origin_context_id,
+                           requested_effect=f"construct new: {t}")
+        ev.append("imperative demand but no resolvable target — "
+                  "clarification, never guessing")
+        probes.append("build where — which project/system?")
+        return _intent(t, "ACTION", "DISCUSSION", 0.5, "TARGET",
+                       ev, probes, context, origin_context_id)
+
+    # 4. default: discussion / context statement
+    ev.append("no authorization structure, no imperative demand → "
+              "discussion")
+    return _intent(t, "DISCUSSION", "DISCUSSION", 0.7, "NONE",
+                   ev, probes, context, origin_context_id)
+
+
+def _resolve_referent(low: str, context: ConversationContext) -> dict | None:
+    """Recency-first referent resolution over LIVE proposals (§17M).
+    Stale-context-resistant: proposals older than STALE_GAP_TURNS are
+    ignored unless the user's text explicitly references their content.
+    Returns {"proposal": Proposal} or {"ambiguous": True,
+    "candidates": [...]} or None."""
+    live = context.live_proposals()
+    if not live:
+        return None
+    fresh = [p for p in live
+             if context.turn - p.created_turn <= STALE_GAP_TURNS
+             or _references(p.summary, low)]
+    if not fresh:
+        return None  # all live proposals are stale for this turn
+    live = fresh
+    # option-selectors pick among live proposals by ordinal
+    m = _OPTION_REFERENTS.search(low)
+    if m:
+        ordinals = {"first": 0, "second": 1, "third": 2, "other": 1,
+                    "latter": len(live) - 1, "former": 0}
+        idx = ordinals.get(m.group(1).lower())
+        if idx is not None and idx < len(live):
+            return {"proposal": live[idx]}
+        if len(live) > 1:
+            return {"ambiguous": True, "candidates": live}
+    # multiple live proposals + no ordinal → ambiguous unless recency
+    # is decisive (only one proposal younger than the last topic shift)
+    if len(live) > 1:
+        # recency heuristic: if the newest proposal is at least 2 turns
+        # newer than the runner-up, recency decides; otherwise ask
+        newest, runner = live[-1], live[-2]
+        if newest.created_turn - runner.created_turn >= 2:
+            return {"proposal": newest}
+        return {"ambiguous": True, "candidates": live}
+    return {"proposal": live[0]}
+
+
+def _references(summary: str, low: str) -> bool:
+    """Does the user's text explicitly reference the proposal's content?
+    Distinctive content words (len>4) appearing in the text revive a
+    stale proposal (explicit > recency)."""
+    words = {w for w in _tokens(summary) if len(w) > 4}
+    return bool(words & set(_tokens(low)))
+
+
+def _consequence(text: str, context: ConversationContext) -> str:
+    low = text.lower()
+    if any(w in low for w in ("replace", "migrate", "rewrite", "delete",
+                              "remove all", "subsystem")):
+        return "HIGH"
+    if any(w in low for w in ("add", "extend", "port", "integrate")):
+        return "MEDIUM"
+    return "LOW"
+
+
+def _constraints(context: ConversationContext) -> list:
+    """User/project constraints recorded in recent turns (light capture;
+    the constraint STORE is future Substrate's — §17O boundary)."""
+    out = []
+    for turn in context.recent_turns[-6:]:
+        txt = turn.get("text", "").lower()
+        if "must not" in txt or "don't" in txt or "never" in txt or \
+                "no hardcoded" in txt:
+            out.append(turn["text"][:120])
+    return out
+
+
+def _intent(objective: str, action_intent: str, approval_state: str,
+            confidence: float, ambiguity: str, ev: list, probes: list,
+            context: ConversationContext, origin_context_id: str | None,
+            requested_effect: str = "",
+            consequence_level: str | None = None) -> ConstructionIntent:
+    return ConstructionIntent(
+        intent_id=f"intent-{uuid.uuid4().hex[:8]}",
+        objective=objective,
+        requested_effect=requested_effect or objective,
+        resolved_target=context.active_project,
+        resolved_referent=None,
+        origin_context_id=origin_context_id,
+        action_intent=action_intent,
+        approval_state=approval_state,
+        confidence=confidence, ambiguity=ambiguity,
+        consequence_level=consequence_level
+        or ("LOW" if action_intent == "DISCUSSION" else "MEDIUM"),
+        relevant_constraints=_constraints(context),
+        classification_evidence=ev, probe_questions=probes)
@@ -0,0 +1,395 @@
+"""construction_mode.py — contextual construction classification.
+
+SCAFFOLD-TRANCHE-0010. ONE canonical home for the construction-mode
+decision (§2/§24: no competing router). Consumers — Rune's native backend
+path, the Construction Interface (MiMo/other builders), manual operator
+override — all call THIS.
+
+The distinction (§3-4):
+
+    NOT_CONSTRUCTION          not a build request at all
+    LOCAL_CHANGE              tiny fix / bounded patch — governance without
+                              ceremony (lightest path)
+    EXISTING_PROJECT_CHANGE   feature/repair/refactor/migration INSIDE an
+                              existing project — attaches to the existing
+                              Project Scaffold / Rune Spine
+    PROJECT_BUILD             genuinely NEW app/tool/service/project —
+                              fuller lifecycle (identity, repo decision,
+                              PRD state, bootstrap)
+    UNKNOWN                   insufficient evidence — ask/probe, NEVER
+                              silently choose
+
+KEY SEMANTIC DISTINCTION (§3): capability resolution BUILD_NEW does NOT
+imply PROJECT_BUILD. Building a new parser module inside Rune is
+EXISTING_PROJECT_CHANGE. Conversely "build me a standalone scheduler app"
+is PROJECT_BUILD before any capability decision exists. Tested explicitly.
+
+NOT keyword-authoritative (§4): "build confidence" / "make a list" do not
+become PROJECT_BUILD. Deterministic evidence comes from:
+    - explicit target project identity (constraint or recorded binding)
+    - whether a Project Scaffold exists for the target
+    - standalone/own-lifecycle language ("standalone", "new app",
+      "new service", "own repository")
+    - scoping words ("this", "existing", "the module") pointing at a
+      known target
+Deterministic evidence is insufficient → UNKNOWN with probe questions.
+Model interpretation may be PROPOSED (recorded, PENDING — never authority,
+§5/§12 of the broader architecture).
+"""
+
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass, field
+
+__all__ = ["ConstructionMode", "classify_mode", "MODES"]
+
+MODES = ("NOT_CONSTRUCTION", "LOCAL_CHANGE", "EXISTING_PROJECT_CHANGE", "PROJECT_BUILD", "UNKNOWN")
+
+# deterministic lexical evidence — presence contributes, absence never
+# decides alone; every rule requires corroboration (§4)
+_STANDALONE_WORDS = {
+    "standalone",
+    "new app",
+    "new application",
+    "new service",
+    "new tool",
+    "new project",
+    "new repository",
+    "own repo",
+    "from scratch",
+    "separate app",
+    "separate service",
+    "greenfield",
+    "scaffold a new",
+    "new web",
+}
+# T15 §15J: extensionless filenames that bind a scoped target when
+# mentioned explicitly (dotted/path tokens go through classify_surface)
+_KNOWN_FILES = {"readme", "makefile", "dockerfile", "changelog", "license",
+                "gemfile", "procfile", "jenkinsfile", "justfile"}
+_LOCAL_WORDS = {
+    "typo",
+    "small fix",
+    "tiny fix",
+    "quick fix",
+    "patch",
+    "minor",
+    "one-liner",
+    "cosmetic",
+    "comment fix",
+    "small correction",
+}
+# "this" alone is deixis without a referent — only NAMED-scope phrases
+# ("the module", "this project", ...) or an explicit target_project bind
+# scope. Bare "this" contributes nothing (Tranche 10 §4).
+_SCOPING_WORDS = {
+    "existing",
+    "the module",
+    "the project",
+    "the app",
+    "the service",
+    "current",
+    "that module",
+    "the codebase",
+    "this module",
+    "this project",
+    "this app",
+}
+_CONSTRUCTION_WORDS = {
+    "build",
+    "create",
+    "make",
+    "implement",
+    "add",
+    "write",
+    "develop",
+    "construct",
+    "fix",
+    "repair",
+    "refactor",
+    "migrate",
+    "change",
+    "extend",
+    "replace",
+    "feature",
+    "export",
+    "module",
+    "service",
+    "app",
+    "tool",
+    "component",
+    "schema",
+    "api",
+}
+_NOT_CONSTRUCTION_HINTS = {
+    "explain",
+    "what is",
+    "how does",
+    "why does",
+    "list",
+    "confidence",
+    "improvisation",
+    "summarize",
+    "describe",
+    "teach me",
+}
+
+
+@dataclass
+class ConstructionMode:
+    """Typed classification decision with full evidence provenance."""
+
+    mode: str
+    evidence: list[str] = field(default_factory=list)
+    target_project: str | None = None
+    project_authority: str | None = None  # e.g. "rune_internal" | scaffold id
+    confidence: float = 0.0
+    clarification_required: bool = False
+    probe_questions: list[str] = field(default_factory=list)
+    bootstrap_required: bool = False
+    override: dict | None = None  # provenance if manually forced
+    proposed_by_model: dict | None = None  # recorded, never authority
+
+    def to_dict(self) -> dict:
+        return {
+            "mode": self.mode,
+            "evidence": self.evidence,
+            "target_project": self.target_project,
+            "project_authority": self.project_authority,
+            "confidence": self.confidence,
+            "clarification_required": self.clarification_required,
+            "probe_questions": self.probe_questions,
+            "bootstrap_required": self.bootstrap_required,
+            "override": self.override,
+            "proposed_by_model": self.proposed_by_model,
+        }
+
+    @classmethod
+    def from_dict(cls, d: dict) -> "ConstructionMode":
+        known = {
+            k: d[k]
+            for k in (
+                "mode",
+                "evidence",
+                "target_project",
+                "project_authority",
+                "confidence",
+                "clarification_required",
+                "probe_questions",
+                "bootstrap_required",
+                "override",
+                "proposed_by_model",
+            )
+            if k in d
+        }
+        return cls(**known)
+
+
+def _tokens(text: str) -> set[str]:
+    return set(re.split(r"[^a-z0-9]+", text.lower()))
+
+
+def _contains_any(text: str, phrases) -> list[str]:
+    t = text.lower()
+    return [p for p in phrases if p in t]
+
+
+def classify_mode(
+    objective: str,
+    *,
+    target_project: str | None = None,
+    project_exists: bool | None = None,
+    affected_surface: list[str] | None = None,
+    explicit_override: str | None = None,
+    override_authority: str | None = None,
+    model_proposal: dict | None = None,
+) -> ConstructionMode:
+    """Classify the construction mode for one intent (§4-6).
+
+    `project_exists` — structural context: whether a Project Scaffold /
+    project binding already exists for the target. Project existence
+    MATERIALLY affects classification (§5/§20): the same request resolves
+    differently with and without an existing project.
+
+    `explicit_override` — operator/builder authority feed (§11/§23):
+    validated against known modes; provenance recorded. Override feeds the
+    SAME machinery — it never bypasses the evidence record.
+    """
+    ev: list[str] = []
+    obj = objective or ""
+    mode = None
+    conf = 0.0
+    bootstrap = False
+    clarification = False
+    probes: list[str] = []
+
+    # 1. explicit override (§23): authority-validated, provenance-carrying
+    if explicit_override is not None:
+        if explicit_override not in MODES:
+            raise ValueError(f"override must be one of {MODES}, got {explicit_override!r}")
+        ev.append(f"explicit override by {override_authority or 'operator'}")
+        m = ConstructionMode(
+            mode=explicit_override,
+            evidence=ev,
+            target_project=target_project,
+            confidence=1.0,
+            bootstrap_required=(explicit_override == "PROJECT_BUILD"),
+            override={"authority": override_authority or "operator", "from_mode": None},
+        )
+        if explicit_override == "UNKNOWN":
+            m.clarification_required = True
+        return m
+
+    # 2. NOT_CONSTRUCTION: question/list/learning intents with no build verbs
+    notc = _contains_any(obj.lower(), _NOT_CONSTRUCTION_HINTS)
+    has_verb = bool(_tokens(obj) & _CONSTRUCTION_WORDS)
+    if notc:
+        # verb-object check: a construction verb whose object is a
+        # non-constructible noun (list/confidence/explanation) is not a
+        # construction request (§4: "build confidence", "make a list")
+        non_constructible = {"list", "confidence", "explanation", "summary", "chord", "note"}
+        obj_tokens = _tokens(obj)
+        standalone_evidence = bool(_contains_any(obj.lower(), _STANDALONE_WORDS))
+        if not has_verb or (obj_tokens & non_constructible and not standalone_evidence):
+            ev.append(f"non-construction intent markers: {notc[:3]}")
+            ev.append(
+                "construction-verb object is not a buildable artifact" if has_verb else "no construction verb present"
+            )
+            return ConstructionMode(mode="NOT_CONSTRUCTION", evidence=ev, confidence=0.85)
+
+    if not obj.strip():
+        ev.append("empty objective")
+        return ConstructionMode(
+            mode="UNKNOWN",
+            evidence=ev,
+            confidence=0.0,
+            clarification_required=True,
+            probe_questions=["what should be built?"],
+        )
+
+    standalone = _contains_any(obj.lower(), _STANDALONE_WORDS)
+    local = _contains_any(obj.lower(), _LOCAL_WORDS)
+    scoped = _contains_any(obj.lower(), _SCOPING_WORDS)
+    # T15 §15J: an explicit FILE mention binds a scoped target ("fix the
+    # typo in the README", "update scheduler.js"). Deterministic lexical
+    # evidence in the same corroboration style as the other lexicons;
+    # dotted/path tokens delegated to the ONE surface authority
+    # (construction_decision.classify_surface) — lazily imported (cycle:
+    # construction_decision imports this module's classify_mode).
+    file_scoped = False
+    for w in re.split(r"[^a-z0-9_./-]+", obj.lower()):
+        if not w or len(w) < 3:
+            continue
+        if w in _KNOWN_FILES or "." in w or "/" in w:
+            try:
+                from rune.construction_decision import classify_surface
+                if classify_surface(w) in ("source_path", "workflow_config"):
+                    file_scoped = True
+                    break
+            except Exception:  # noqa: BLE001,S110 — authority unavailable:
+                pass           # lexical fallback decides; never fatal
+            if w in _KNOWN_FILES:
+                file_scoped = True
+                break
+
+    # T17 §17Q: indefinite-article new-artifact evidence — "a small CLI
+    # that renames photos", "build me a program that..." — is
+    # new-lifecycle language even without explicit standalone keywords.
+    # Guarded: requires a CREATION verb or artifact noun, and yields to
+    # local/scope/file evidence (fix a bug / add a flag never counts).
+    _CREATION_VERBS = {"build", "create", "make", "write", "develop"}
+    _ARTIFACT_NOUNS = {"program", "cli", "tool", "app", "script", "service",
+                       "library", "package", "module", "dashboard",
+                       "website", "bot", "generator", "tracker", "game",
+                       "application", "command"}
+    article_standalone = False
+    if re.search(r"\b(a|an|another)\s+[a-z]", obj.lower()):
+        creation_verb = bool(_tokens(obj.lower()) & _CREATION_VERBS)
+        artifact_noun = bool(_tokens(obj.lower()) & _ARTIFACT_NOUNS)
+        # structural evidence outranks lexical newness (§5): a known
+        # existing target means article language describes a CHANGE
+        # ("add a flag to the deploy script"), not a new project
+        if (creation_verb or artifact_noun) and not local and not scoped \
+                and not file_scoped and not target_project \
+                and project_exists is not True:
+            article_standalone = True
+            ev.append("indefinite-article new-artifact language")
+
+    # 3. PROJECT_BUILD: standalone/new-lifecycle language AND no existing
+    #    project for the target (project existence materially matters, §5)
+    if standalone or article_standalone:
+        ev.append(f"standalone/new-lifecycle language: {standalone[:3]}")
+        if project_exists is True:
+            # §5: existing project preferred unless evidence says otherwise —
+            # contradictory evidence → UNKNOWN, never silent choice
+            ev.append("BUT target project already exists (structural evidence contradicts standalone language)")
+            mode, conf, clarification = "UNKNOWN", 0.3, True
+            probes.append("should this extend the existing project or start a genuinely separate one?")
+        elif project_exists is None:
+            ev.append("no project-existence evidence for the target")
+            mode, conf, clarification = "UNKNOWN", 0.4, True
+            probes.append("does a project for this already exist?")
+        else:
+            mode, conf = "PROJECT_BUILD", 0.9
+            bootstrap = True
+    # 4. LOCAL_CHANGE: explicit smallness + scoped target + no standalone
+    elif local and (scoped or file_scoped or target_project or project_exists):
+        ev.append(f"smallness markers: {local[:3]}")
+        if scoped:
+            ev.append(f"scoping words bind to known target: {scoped[:3]}")
+        if file_scoped:
+            ev.append("file mention binds scoped target")
+        mode, conf = "LOCAL_CHANGE", 0.8
+    # 5. EXISTING_PROJECT_CHANGE: scoping to a known/existing target.
+    #    T11 §1: a change scoped to a project with PROVEN non-existence must
+    #    NOT silently claim existing-project semantics — clarify instead.
+    elif (scoped or target_project or project_exists is True) and has_verb:
+        ev.append("change scoped to an existing project/target")
+        if target_project:
+            ev.append(f"target project: {target_project}")
+        if project_exists is False:
+            ev.append(
+                "BUT structural evidence proves the target project "
+                "does not exist — refusing to manufacture existing-"
+                "project semantics"
+            )
+            mode, conf, clarification = "UNKNOWN", 0.3, True
+            probes.append(
+                f"project '{target_project or 'the target'}' has "
+                "no structural record — is it new (project "
+                "build) or does it live somewhere not bound?"
+            )
+        else:
+            mode, conf = "EXISTING_PROJECT_CHANGE", 0.8
+    # 6. construction verb but no scope and no standalone language → UNKNOWN
+    elif has_verb:
+        ev.append("construction verb present but no target-scope evidence and no standalone/new-lifecycle language")
+        mode, conf, clarification = "UNKNOWN", 0.35, True
+        probes.append("is this a change to an existing project or a new standalone one?")
+        probes.append("which project/module is the target?")
+    else:
+        ev.append("insufficient lexical evidence either way")
+        mode, conf, clarification = "UNKNOWN", 0.2, True
+        probes.append("is this a construction request?")
+
+    if affected_surface and mode in ("LOCAL_CHANGE", "EXISTING_PROJECT_CHANGE", "UNKNOWN"):
+        ev.append(f"affected surface known: {affected_surface[:2]}")
+        if mode == "UNKNOWN":
+            mode, conf = "EXISTING_PROJECT_CHANGE", 0.6
+            clarification = False
+
+    # §5 model proposal: recorded, never authority
+    proposed = model_proposal if model_proposal else None
+
+    return ConstructionMode(
+        mode=mode,
+        evidence=ev,
+        target_project=target_project,
+        project_authority=("rune_internal" if target_project in (None, "rune", "issue161-harness") else target_project),
+        confidence=conf,
+        clarification_required=clarification,
+        probe_questions=probes,
+        bootstrap_required=(mode == "PROJECT_BUILD"),
+        proposed_by_model=proposed,
+    )
@@ -0,0 +1,496 @@
+"""construction_pattern.py — reusable evidence about HOW a class of thing
+is normally built.
+
+SCAFFOLD-TRANCHE-006. Canonical distinction (tranche brief):
+
+    Construction Pattern  = reusable knowledge for a CLASS of construction
+    Project Scaffold      = structural truth for THIS project
+    Construction Contract = obligations for THIS change
+
+A pattern is GUIDANCE, not authority (§3): project evidence always wins.
+A pattern must never overwrite contradictory project evidence — if the
+pattern assumes HTTP services and the project evidence says message-queue-
+only, the pattern adapts or becomes non-applicable.
+
+Applicability (§4-5) is EVIDENCE-BACKED and deterministic: signals are
+checked against the affected surface + structural authorities. Verdicts:
+APPLIES / PARTIAL / DOES_NOT_APPLY / UNKNOWN. UNKNOWN never becomes APPLIES
+because a name sounds similar; name similarity establishes nothing. Model
+reasoning may PROPOSE a pattern (recorded, PENDING) — it may not certify
+applicability.
+
+Obligation derivation (§6-7) is FILTERED by this project: every pattern
+obligation carries a `when` condition evaluated against project evidence;
+verdicts REQUIRED / APPLICABLE / NOT_APPLICABLE / UNKNOWN with the reason.
+Patterns REDUCE false-positive obligation expansion, never amplify it.
+
+Provenance (§16): every rule answers "why is this here" with real Rune
+evidence — incidents, contracts, tests, user requirements. Patterns
+accumulate verified engineering knowledge, not model folklore.
+
+Revision (§17): lightweight revision identity; contracts record the
+revision consumed so "which builds used the old rule" stays answerable.
+"""
+
+from __future__ import annotations
+
+from rune import structural_store as ss
+
+__all__ = [
+    "ConstructionPattern",
+    "evaluate_applicability",
+    "derive_pattern_obligations",
+    "record_pattern_proposal",
+    "APPLIES",
+    "PARTIAL",
+    "DOES_NOT_APPLY",
+    "UNKNOWN",
+    "OBL_REQUIRED",
+    "OBL_APPLICABLE",
+    "OBL_NOT_APPLICABLE",
+    "OBL_UNKNOWN",
+]
+
+APPLIES = "APPLIES"
+PARTIAL = "PARTIAL"
+DOES_NOT_APPLY = "DOES_NOT_APPLY"
+UNKNOWN = "UNKNOWN"
+
+OBL_REQUIRED = "REQUIRED"
+OBL_APPLICABLE = "APPLICABLE"
+OBL_NOT_APPLICABLE = "NOT_APPLICABLE"
+OBL_UNKNOWN = "UNKNOWN"
+
+
+# ── pattern object (open-world) ─────────────────────────────────────────────
+
+
+class ConstructionPattern:
+    """One reusable construction pattern. Open-world: `ext` carries anything
+    the schema above does not name. NOT structural truth — guidance."""
+
+    def __init__(
+        self,
+        pattern_id: str,
+        purpose: str,
+        *,
+        revision: int = 1,
+        maturity: str = "CANDIDATE",
+        signals: dict | None = None,
+        obligations: list[dict] | None = None,
+        provenance: list[str] | None = None,
+        exclusions: list[str] | None = None,
+        toolchain_notes: list[str] | None = None,
+        runtime_effects: list[str] | None = None,
+        **ext,
+    ):
+        if not pattern_id:
+            raise ValueError("pattern_id required")
+        if not purpose:
+            raise ValueError("purpose required")
+        for o in obligations or []:
+            for key in ("id", "obligation", "when"):
+                if key not in o:
+                    raise ValueError(f"obligation missing {key!r}: {o}")
+        self.pattern_id = pattern_id
+        self.purpose = purpose
+        self.revision = revision
+        self.maturity = maturity
+        self.signals = dict(signals or {})
+        self.obligations = list(obligations or [])
+        self.provenance = sorted(provenance or [])
+        self.exclusions = sorted(exclusions or [])
+        self.toolchain_notes = sorted(toolchain_notes or [])
+        self.runtime_effects = sorted(runtime_effects or [])
+        self.ext = dict(ext)
+
+    def to_dict(self) -> dict:
+        return {
+            "pattern_id": self.pattern_id,
+            "purpose": self.purpose,
+            "revision": self.revision,
+            "maturity": self.maturity,
+            "signals": dict(self.signals),
+            "obligations": [dict(o) for o in self.obligations],
+            "provenance": list(self.provenance),
+            "exclusions": list(self.exclusions),
+            "toolchain_notes": list(self.toolchain_notes),
+            "runtime_effects": list(self.runtime_effects),
+            **self.ext,
+        }
+
+    @classmethod
+    def from_dict(cls, d: dict) -> "ConstructionPattern":
+        src = dict(d)  # never mutate the caller's dict
+        known = {
+            k: src.pop(k)
+            for k in (
+                "pattern_id",
+                "purpose",
+                "revision",
+                "maturity",
+                "signals",
+                "obligations",
+                "provenance",
+                "exclusions",
+                "toolchain_notes",
+                "runtime_effects",
+            )
+            if k in src
+        }
+        return cls(**known, **src)
+
+    def revise(self, *, reason: str, evidence: list[str], changes: dict | None = None) -> "ConstructionPattern":
+        """Revision with provenance (§17): prior meaning not invisibly
+        mutated — returns a NEW pattern at revision+1 recording the change."""
+        d = self.to_dict()
+        d.pop("pattern_id")
+        rev = {**d, **(changes or {})}
+        rev["revision"] = self.revision + 1
+        rev.setdefault("revision_history", []).append(
+            {
+                "prior_revision": self.revision,
+                "reason": reason[:200],
+                "evidence": sorted(evidence),
+                "changed": sorted((changes or {}).keys()),
+            }
+        )
+        return ConstructionPattern(self.pattern_id, **rev)
+
+
+# ── model proposal seam (§5 — recorded, never authoritative) ────────────────
+
+
+def record_pattern_proposal(
+    pattern_id: str, *, provider: str, model: str, evidence: list[str], confidence: float
+) -> dict:
+    """A model may PROPOSE a pattern; it may not certify applicability.
+    Recorded with verification state PENDING until deterministic project
+    evidence, research evidence, proven construction evidence, or an
+    explicit user decision accepts it."""
+    if not provider or not model:
+        raise ValueError("pattern proposal requires provider+model")
+    if not 0.0 <= confidence <= 1.0:
+        raise ValueError("confidence must be in [0,1]")
+    return {
+        "kind": "pattern_proposal",
+        "pattern_id": pattern_id,
+        "provider": provider,
+        "model": model,
+        "evidence": sorted(evidence),
+        "confidence": confidence,
+        "verification_state": "PENDING",
+    }
+
+
+# ── applicability (§4 — deterministic signals only) ─────────────────────────
+
+
+def _surface_text(need, surface_result: dict) -> set[str]:
+    """Everything the affected surface offers as evidence: paths, entities,
+    ui surfaces, regions, workflow configs, the objective text tokens."""
+    tokens = set()
+    for k in (
+        "affected_paths",
+        "affected_existing_entities",
+        "affected_ui_surfaces",
+        "affected_regions",
+        "workflow_config_surfaces",
+    ):
+        tokens.update(surface_result.get(k, []))
+    tokens.update(need.affected_surface)
+    # objective word tokens (deterministic split — not semantics)
+    tokens.update(w.lower() for w in need.objective.replace("/", " ").replace("-", " ").split())
+    return tokens
+
+
+# ── shared boundary-aware matchers (Tranche 7 §0) ──────────────────────────
+
+def _path_matches(signal: str, path: str) -> bool:
+    """Boundary-aware path-signal matching: short (<4 char) signals match
+    path TOKENS only (the 'ui'-in-'builtin' false-positive class); longer
+    signals may use containment."""
+    import re as _re
+
+    s = signal.lower()
+    if len(s) >= 4:
+        return s in path.lower()
+    return s in set(_re.split(r"[^a-z0-9]+", path.lower()))
+
+
+def _entity_matches(signal: str, entity_id: str) -> bool:
+    """Same discipline for entity ids. Prefix signals ending in ':' (e.g.
+    'ui:', 'region:') match the id's namespace prefix — structural kinds
+    carry meaning in their prefix, and prefix equality is exact, not a
+    substring accident."""
+    import re as _re
+
+    s = signal.lower()
+    if s.endswith(":"):
+        return entity_id.lower().startswith(s)
+    if len(s) >= 4:
+        return s in entity_id.lower()
+    return s in set(_re.split(r"[^a-z0-9:]+", entity_id.lower()))
+
+
+def evaluate_applicability(
+    pattern: ConstructionPattern, need, surface_result: dict, authorities: dict[str, dict]
+) -> dict:
+    """Deterministic applicability verdict from the pattern's declared
+    signals (§4). Signals (all optional; evidence-checked against the
+    affected surface and structural authorities):
+        surface_kinds     subset of {source_path, workflow_config,
+                                     ui_component, layout_region,
+                                     structural_entity}
+        path_contains     any-of substrings expected in affected paths
+        entities_match    any-of entity-id prefixes/substrings
+        objective_any     any-of lowercase words expected in the objective
+        contracts_match   any-of contract ids implemented by affected entities
+
+    Verdict: count how many DECLARED signal groups produce evidence.
+        all declared groups hit  -> APPLIES
+        some hit                 -> PARTIAL (or DOES_NOT_APPLY when the
+                                    pattern's `strong` groups all miss)
+        none / no signals        -> UNKNOWN (never APPLIES by name)
+    Project evidence overrides (§3): exclusions evaluated first — if the
+    need explicitly declares the pattern non-applicable, DOES_NOT_APPLY
+    with the exclusion recorded.
+    """
+    # §3/§18: PROJECT evidence overrides the pattern. Exclusions may be
+    # declared by the project/need (pattern_exclusions constraint) or by the
+    # pattern itself (non-applicability domains).
+    declared_exclusions = list(pattern.exclusions) + list(need.constraints.get("pattern_exclusions", []))
+    for ex in declared_exclusions:
+        if ex.lower() in need.objective.lower() or ex in need.constraints.get("pattern_exclusions", []):
+            return {
+                "pattern": pattern.pattern_id,
+                "verdict": DOES_NOT_APPLY,
+                "reason": f"declared exclusion matched: {ex}",
+                "evidence": {"exclusion": ex, "source": "project/need declaration"},
+            }
+
+    signals = pattern.signals
+    if not signals:
+        return {
+            "pattern": pattern.pattern_id,
+            "verdict": UNKNOWN,
+            "reason": "pattern declares no applicability signals — nothing deterministic to check",
+            "evidence": {},
+        }
+    tokens = _surface_text(need, surface_result)
+
+    groups_hit, groups_declared = [], 0
+    evidence: dict = {}
+
+    def _check(group: str, candidates: list[str], present) -> None:
+        nonlocal groups_declared
+        if not candidates:
+            return
+        groups_declared += 1
+        hits = [c for c in candidates if present(c)]
+        if hits:
+            groups_hit.append(group)
+            evidence[group] = hits[:5]
+
+    _check(
+        "surface_kinds",
+        signals.get("surface_kinds", []),
+        lambda c: (
+            bool(
+                surface_result.get(
+                    {
+                        "source_path": "affected_paths",
+                        "workflow_config": "workflow_config_surfaces",
+                        "ui_component": "affected_ui_surfaces",
+                        "layout_region": "affected_regions",
+                    }.get(c, ""),
+                    [],
+                )
+            )
+            or any(r["kind"] == c for r in surface_result.get("resolutions", []))
+        ),
+    )
+    _check(
+        "path_contains",
+        [p for p in signals.get("path_contains", [])],
+        lambda c: any(
+            _path_matches(c, p)
+            for p in surface_result.get("affected_paths", []) + surface_result.get("workflow_config_surfaces", [])
+        ),
+    )
+    _check(
+        "entities_match",
+        [e for e in signals.get("entities_match", [])],
+        lambda c: any(
+            _entity_matches(c, e) for e in surface_result.get("affected_existing_entities", [])
+        ),
+    )
+    _check(
+        "objective_any",
+        [w.lower() for w in signals.get("objective_any", [])],
+        lambda c: c in {w.lower() for w in need.objective.replace("/", " ").split()},
+    )
+    contracts_match = signals.get("contracts_match", [])
+    if contracts_match:
+        groups_declared += 1
+        hits = []
+        for name, doc in authorities.items():
+            for e in surface_result.get("affected_existing_entities", []):
+                for r in ss.relationships(doc, from_id=e):
+                    if r["type"] in {"implements", "exposes", "provides", "serves"} and r["to"] in contracts_match:
+                        hits.append(f"{e}->{r['to']}")
+        if hits:
+            groups_hit.append("contracts_match")
+            evidence["contracts_match"] = hits[:5]
+
+    if groups_declared == 0:
+        return {
+            "pattern": pattern.pattern_id,
+            "verdict": UNKNOWN,
+            "reason": "no applicable signal groups declared",
+            "evidence": {},
+        }
+    if len(groups_hit) == groups_declared:
+        verdict = APPLIES
+    elif groups_hit:
+        verdict = PARTIAL
+    else:
+        verdict = DOES_NOT_APPLY
+    return {
+        "pattern": pattern.pattern_id,
+        "verdict": verdict,
+        "reason": f"{len(groups_hit)}/{groups_declared} signal groups "
+        f"produced evidence: {sorted(groups_hit) or 'none'}",
+        "evidence": evidence,
+    }
+
+
+# ── obligation derivation (§6-7 — filtered by project evidence) ─────────────
+
+
+def _obligation_when_hit(obligation: dict, need, surface_result: dict, authorities: dict[str, dict]) -> tuple[str, str]:
+    """Evaluate one obligation's `when` condition against project evidence.
+    Returns (verdict, reason). `when` shapes (all deterministic):
+        path_contains: [substrings]   — any affected/config path matches
+        entities_match: [substrings]  — any affected entity matches
+        always: true                  — applies whenever the pattern applies
+        contracts_touched: [ids]      — contract in requirement or implemented
+                                        by affected entities
+    An obligation with no recognized `when` is UNKNOWN (never silently
+    required)."""
+    when = obligation.get("when")
+    paths = [p for p in surface_result.get("affected_paths", [])] + [
+        p for p in surface_result.get("workflow_config_surfaces", [])
+    ]
+    entities = (
+        [e for e in surface_result.get("affected_existing_entities", [])]
+        + [u for u in surface_result.get("affected_ui_surfaces", [])]
+        + [r for r in surface_result.get("affected_regions", [])]
+    )
+    if when is None:
+        return OBL_UNKNOWN, "obligation declares no applicability condition"
+    if when.get("always") is True:
+        return OBL_REQUIRED, "pattern-level obligation (always applies when the pattern applies)"
+    if "path_contains" in when:
+        hits = [p for sub in when["path_contains"] for p in paths if _path_matches(sub, p)]
+        if hits:
+            return OBL_APPLICABLE, f"affected path(s): {hits[:3]}"
+        return OBL_NOT_APPLICABLE, ("no affected path matches " + str(when["path_contains"]))
+    if "entities_match" in when:
+        hits = [e for sub in when["entities_match"] for e in entities if _entity_matches(sub, e)]
+        if hits:
+            return OBL_APPLICABLE, f"affected entity/entities: {hits[:3]}"
+        return OBL_NOT_APPLICABLE, ("no affected entity matches " + str(when["entities_match"]))
+    if "contracts_touched" in when:
+        touched = set(need.constraints.get("contracts", []))
+        for name, doc in authorities.items():
+            for e in entities:
+                for r in ss.relationships(doc, from_id=e):
+                    if r["type"] in {"implements", "exposes", "provides", "serves"}:
+                        touched.add(r["to"])
+        hits = [c for c in when["contracts_touched"] if c in touched]
+        if hits:
+            return OBL_APPLICABLE, f"contract(s) touched: {hits}"
+        return OBL_NOT_APPLICABLE, ("touched contracts do not include " + str(when["contracts_touched"]))
+    return OBL_UNKNOWN, f"unrecognized when-condition shape: {sorted(when)}"
+
+
+def derive_pattern_obligations(
+    pattern: ConstructionPattern,
+    need,
+    surface_result: dict,
+    authorities: dict[str, dict],
+    project_exceptions: list[dict] | None = None,
+) -> dict:
+    """Derive this pattern's obligations FILTERED by project evidence (§7).
+    Every returned obligation carries: pattern, revision, obligation text,
+    verdict, reason/evidence, and provenance kind PATTERN (distinct from
+    Tranche 5's STRUCTURAL derivation — §8 provenance separation).
+    Project exceptions (§18) suppress with provenance retained."""
+    project_exceptions = list(project_exceptions or [])
+    out = {
+        "pattern": pattern.pattern_id,
+        "revision": pattern.revision,
+        "accepted": [],
+        "rejected": [],
+        "unresolved": [],
+    }
+    for o in pattern.obligations:
+        entry = {
+            "id": o["id"],
+            "obligation": o["obligation"],
+            "pattern": pattern.pattern_id,
+            "revision": pattern.revision,
+            "provenance": "PATTERN",
+            "pattern_provenance": list(pattern.provenance),
+        }
+        # project exception check first (§18) — suppress with provenance
+        exc = next((x for x in project_exceptions if x.get("obligation") == o["id"]), None)
+        if exc:
+            entry.update(
+                {
+                    "verdict": "EXCEPTED",
+                    "reason": exc.get("reason", ""),
+                    "exception_evidence": exc.get("evidence", ""),
+                    "exception_authority": exc.get("authority", "user"),
+                }
+            )
+            out["rejected"].append(entry)
+            continue
+        verdict, reason = _obligation_when_hit(o, need, surface_result, authorities)
+        entry["verdict"] = verdict
+        entry["reason"] = reason
+        if verdict in (OBL_REQUIRED, OBL_APPLICABLE):
+            out["accepted"].append(entry)
+        elif verdict == OBL_NOT_APPLICABLE:
+            out["rejected"].append(entry)
+        else:
+            out["unresolved"].append(entry)
+    return out
+
+
+def combine_pattern_obligations(derivations: list[dict]) -> dict:
+    """Combine multiple patterns' derivations (§26-27): duplicates of the
+    same obligation id collapse (provenance lists merge); conflicting
+    obligations (same id, contradictory verdicts across patterns) surface
+    as conflicts requiring evidence/user decision — never silently chosen."""
+    merged: dict[str, dict] = {}
+    conflicts: list[dict] = []
+    for d in derivations:
+        for entry in d["accepted"]:
+            oid = entry["id"]
+            if oid in merged:
+                prev = merged[oid]
+                if prev["obligation"] != entry["obligation"]:
+                    conflicts.append(
+                        {
+                            "obligation_id": oid,
+                            "conflict": "same id, different text",
+                            "sources": [prev["pattern"], entry["pattern"]],
+                        }
+                    )
+                    continue
+                prev.setdefault("also_from", []).append(entry["pattern"])
+                continue
+            merged[oid] = dict(entry)
+    return {"obligations": sorted(merged.values(), key=lambda x: x["id"]), "conflicts": conflicts}
@@ -0,0 +1,249 @@
+"""construction_patterns_builtin.py — the initial evidence-grounded pattern
+set (SCAFFOLD-TRANCHE-006 §10-14: 4 patterns, each derived from REAL Rune
+evidence, each appropriately immature).
+
+This module is DATA, not a framework: pattern provenance answers "why is
+this rule here" with real incidents/contracts/tests. Maturity stays
+CANDIDATE/UNIT-PROVEN — existence is not trust (§15).
+"""
+
+from __future__ import annotations
+
+from rune.construction_pattern import ConstructionPattern
+
+__all__ = ["BUILTIN_PATTERNS", "get_pattern", "PATTERN_REGISTRY_REVISION"]
+
+# Bumped when any builtin pattern's obligations change meaningfully.
+PATTERN_REGISTRY_REVISION = 1
+
+CAPABILITY_ADAPTATION = ConstructionPattern(
+    "capability-module-adaptation",
+    "Adapting or extending an existing capability/module: prove the NEW "
+    "behavior, preserve the OLD behavior (BUILD BUT DON'T BREAK).",
+    maturity="UNIT-PROVEN",
+    signals={
+        "entities_match": ["module:", "capability:"],
+        "path_contains": ["src/"],
+        "objective_any": ["adapt", "extend", "modify", "preserve", "existing", "capability", "repair"],
+    },
+    obligations=[
+        {
+            "id": "adapt-new-behavior-proof",
+            "obligation": "new behavior proven by tests before closure",
+            "when": {"always": True},
+        },
+        {
+            "id": "adapt-old-behavior-preserved",
+            "obligation": "existing capability behavior preserved; original contracts remain valid",
+            "when": {"always": True},
+        },
+        {
+            "id": "adapt-consumers-verified",
+            "obligation": "existing consumers of the adapted capability verified against their contracts",
+            "when": {"entities_match": ["module:", "capability:"]},
+        },
+        {
+            "id": "adapt-migration-detected",
+            "obligation": "migration obligations detected and recorded if any consumer-visible interface changes",
+            "when": {"path_contains": ["src/"]},
+        },
+    ],
+    provenance=[
+        "SCAFFOLD-TRANCHE-001 (build-but-don't-break established)",
+        "SCAFFOLD-TRANCHE-003 (adaptation two-proof rule)",
+        "construction_decision.obligations_from_trace (auto-derived since T4)",
+        "LIVE-CONTRACT-QUALITY-001 (adaptation resolved + closed live)",
+    ],
+)
+
+RELEASE_DEPLOYMENT = ConstructionPattern(
+    "release-deployment-change",
+    "Changes to the release/deployment path: fail-closed qualification, "
+    "exact candidate identity, staging != promotion, truthful promotion "
+    "outcomes, deployed-identity verification, negative + forward proofs.",
+    maturity="UNIT-PROVEN",
+    signals={
+        "path_contains": [
+            "deploy",
+            "promotion",
+            "qualify",
+            "release.yml",
+            "ci.yml",
+            "deploy_qualify",
+            "deploy_stage",
+            "deploy_promotion",
+        ],
+        "objective_any": ["deploy", "release", "promot", "qualif", "candidate", "current", "monotonic"],
+        "workflow_config_signals": True,  # informational; surface kinds carry it
+    },
+    obligations=[
+        {
+            "id": "rel-fail-closed-qualification",
+            "obligation": "CI qualification remains fail-closed (exact SHA, "
+            "two-leg verdict); red candidates never deploy",
+            "when": {"path_contains": ["deploy", "release", "qualify", "ci.yml"]},
+        },
+        {
+            "id": "rel-exact-candidate-identity",
+            "obligation": "deployment addresses the EXACT pushed revision, never whatever HEAD is at deploy time",
+            "when": {"path_contains": ["deploy", "release"]},
+        },
+        {
+            "id": "rel-staging-promotion-distinction",
+            "obligation": "staged candidate != promoted candidate; promotion outcome must be independently verifiable",
+            "when": {"path_contains": ["deploy", "promotion", "stage"]},
+        },
+        {
+            "id": "rel-promotion-truthfulness",
+            "obligation": "promotion outcomes are truthful (PROMOTED / "
+            "NOOP_ALREADY_CURRENT / STALE_SUPERSEDED / REFUSED_*); "
+            "stale candidates never claim success",
+            "when": {"path_contains": ["promotion", "deploy_stage", "deploy_promotion"]},
+        },
+        {
+            "id": "rel-monotonic-promotion",
+            "obligation": "promotion ordering: an older candidate finishing "
+            "later cannot regress current; ancestry check + "
+            "fenced critical section",
+            "when": {"path_contains": ["deploy_promotion", "deploy_stage", "promotion"]},
+        },
+        {
+            "id": "rel-negative-candidate-proof",
+            "obligation": "negative proof: red/unqualified candidate is refused; stale candidate cannot promote",
+            "when": {"path_contains": ["deploy", "promotion", "qualify"]},
+        },
+        {
+            "id": "rel-forward-candidate-proof",
+            "obligation": "forward proof: a green newer candidate still promotes normally",
+            "when": {"path_contains": ["deploy", "promotion"]},
+        },
+        {
+            "id": "rel-deployed-identity-verification",
+            "obligation": "deployed identity verified from the deployment "
+            "authority (current symlink + build identity), not "
+            "from job artifacts alone",
+            "when": {"always": True},
+        },
+        {
+            "id": "rel-release-regression-verification",
+            "obligation": "existing release-safety guarantees re-proven (the "
+            "guarantees that already exist must not silently "
+            "weaken)",
+            "when": {"always": True},
+        },
+    ],
+    provenance=[
+        "maintenance:deploy-order-race (live incident: deploys 7355/7359)",
+        "DEPLOY-MONOTONIC-001 (monotonic promotion + truthful outcomes, live-proven)",
+        "release-safety repair 2026-09-13 (red-promotion incidents 0.66.28/29: required-checks aggregator lied)",
+        "scripts/deploy_promotion.py (ancestry + fenced critical section)",
+        "tests/test_deploy_stage_race.py (22-test matrix incl. two-process live race)",
+    ],
+)
+
+UI_TUI_STRUCTURAL = ConstructionPattern(
+    "ui-tui-structural-change",
+    "UI/TUI structural changes: component identity, Layout Contract "
+    "inheritance, USER-LOCKED preservation, deterministic verification.",
+    maturity="UNIT-PROVEN",
+    signals={
+        "surface_kinds": ["ui_component", "layout_region"],
+        "path_contains": ["tui", "ui", "widget"],
+        "entities_match": ["ui:", "region:", "layout:"],
+    },
+    obligations=[
+        {
+            "id": "ui-layout-contract-inherited",
+            "obligation": "affected UI components inherit their Layout Contract "
+            "obligations; layout_contract_check must PASS",
+            "when": {"entities_match": ["ui:", "region:", "layout:"]},
+        },
+        {
+            "id": "ui-user-locked-preserved",
+            "obligation": "USER-LOCKED placement is user authority: preserved "
+            "exactly; changes require explicit user authorization",
+            "when": {"entities_match": ["ui:", "region:"]},
+        },
+        {
+            "id": "ui-unknown-never-pass",
+            "obligation": "layout verification UNKNOWN (no adapter evidence) must never be reported as PASS",
+            "when": {"entities_match": ["ui:", "region:", "layout:"]},
+        },
+        {
+            "id": "ui-component-identity-stable",
+            "obligation": "ui:<component> identity survives implementation-path "
+            "changes; authority never keyed by filename alone",
+            "when": {"entities_match": ["ui:", "region:", "layout:"]},
+        },
+    ],
+    provenance=[
+        "issue #105 + plans/2026-09-03-tui-footer.md (version display)",
+        "Tranche 3 real TUI proof (gate PASS -> violation FAIL -> restore)",
+        "tests/test_layout_contract.py + tests/test_layout_contract_real_proof.py",
+        "maintenance:version-display-user-lock (USER-LOCK upgrade pending "
+        "user confirmation — provenance rule honored, not fabricated)",
+    ],
+)
+
+STORE_SCHEMA_EVOLUTION = ConstructionPattern(
+    "store-schema-evolution",
+    "Durable store/schema changes: backward readability, authority "
+    "preservation, deterministic roundtrip, migration only when necessary.",
+    maturity="CANDIDATE",
+    signals={
+        "path_contains": ["store", "schema", "structural_store", "scaffold", "spine"],
+        "entities_match": ["module:structural_store", "module:scaffold", "module:spine"],
+        "objective_any": ["store", "schema", "record", "persist", "field", "migration"],
+    },
+    obligations=[
+        {
+            "id": "store-backward-readability",
+            "obligation": "old records without new fields read as empty/unknown, never crash; legacy keys preserved",
+            "when": {"path_contains": ["store", "structural_store", "scaffold", "spine", "schema"]},
+        },
+        {
+            "id": "store-roundtrip",
+            "obligation": "deterministic save/load roundtrip incl. unknown entity/relation kinds and open-world fields",
+            "when": {"path_contains": ["store", "structural_store", "scaffold", "schema"]},
+        },
+        {
+            "id": "store-authority-preservation",
+            "obligation": "structural facts only change through their owning "
+            "projector/authority; canonical rebuild after "
+            "registration (no hand-edited structural facts)",
+            "when": {"path_contains": ["spine", "scaffold", "store/"]},
+        },
+        {
+            "id": "store-migration-only-if-needed",
+            "obligation": "no mandatory migration unless semantics truly change; "
+            "ext/open-world fields preferred over schema churn",
+            "when": {"path_contains": ["schema", "store", "structural_store"]},
+        },
+    ],
+    provenance=[
+        "Tranche 2 structural-store extraction (behavior-neutrality proof)",
+        "Tranche 4 contract-quality repair (legacy keys preserved, old contracts readable — test_h)",
+        "Tranche 5 expected-runtime-effects extension (old contracts close without new fields)",
+        "Tranche 2 CI lesson: hand-edited spine store diverged from canonical rebuild (deploy-identity edges)",
+    ],
+)
+
+BUILTIN_PATTERNS = [CAPABILITY_ADAPTATION, RELEASE_DEPLOYMENT, UI_TUI_STRUCTURAL, STORE_SCHEMA_EVOLUTION]
+
+_REGISTRY: dict[str, ConstructionPattern] = {p.pattern_id: p for p in BUILTIN_PATTERNS}
+
+
+def get_pattern(pattern_id: str, revision: int | None = None) -> ConstructionPattern | None:
+    """Fetch a builtin pattern; a specific revision raises if the registry
+    no longer carries it (contracts reference the revision they consumed —
+    §17)."""
+    p = _REGISTRY.get(pattern_id)
+    if p is None:
+        return None
+    if revision is not None and p.revision != revision:
+        raise ValueError(
+            f"pattern {pattern_id} revision {revision} not in registry "
+            f"(current: {p.revision}) — the consuming contract records which "
+            f"revision it used"
+        )
+    return p
@@ -0,0 +1,135 @@
+"""construction_projection.py — T16 §16I: stable versioned PUBLIC
+projection of construction state for future Substrate integration.
+
+This is a PROJECTION, never a second authority. All fields are derived
+(read-only) from the canonical scaffold store (Scaffold.load) plus the
+live wait bridge state (event_wait.operation_wait). A consumer (the
+future Substrate, the mock client in tests) must know NOTHING about
+current.json layout or internal entity arrays — only this schema.
+
+Schema versioning: SCHEMA_VERSION is bumped on any field change; consumers
+negotiate by reading schema_version first. Unknown future fields are
+ignored by readers (forward-tolerant); removed fields bump the major.
+"""
+from __future__ import annotations
+
+from typing import Any
+
+SCHEMA_VERSION = 1
+
+# The mandate's required projection fields (§16I). None may carry live
+# provider/account secrets or internal storage layout.
+FIELDS = (
+    "schema_version", "project_id", "originating_goal_id",
+    "construction_id", "contract_id",
+    "mode", "depth", "phase", "stage",
+    "operation_id", "waiting_state", "correlation_id",
+    "builder_state", "blockers", "approval_required",
+    "readiness_canary", "liveness_canary", "exit_canary",
+    "verification_state", "checkpoint", "evidence_refs",
+    "failure_class", "completion_state", "candidate_next_action",
+)
+
+
+def _contract_entity(doc: dict, contract_id: str) -> dict | None:
+    for e in doc.get("entities", []):
+        if e.get("id") == f"contract:{contract_id}":
+            return e
+    return None
+
+
+def project(
+    scaffold, contract_id: str, *,
+    originating_goal_id: str = "",
+    operation_id: str = "",
+    correlation_id: str = "",
+    stage: str = "",
+    builder_state: str = "",
+    verification_state: str = "",
+    checkpoint: str = "",
+    failure_class: str = "",
+    completion_state: str = "",
+) -> dict:
+    """Build the stable public projection for one contract. `scaffold` is
+    a Rune Scaffold instance (the canonical store); all live fields are
+    derived from it. Optional live fields (waiting_state etc.) are filled
+    from the event bridge when operation_id is supplied."""
+    doc = scaffold.load()
+    c = _contract_entity(doc, contract_id) or {}
+    ids = c.get("identities", {})
+    ext = c.get("ext", {})
+    trace = ext.get("decision_trace", {})
+
+    waiting_state = ""
+    live_correlation = correlation_id
+    if operation_id:
+        try:
+            from rune import event_wait as ew
+            rec = ew.operation_wait(operation_id)
+            if rec is not None:
+                waiting_state = rec.get("status", "")
+                live_correlation = correlation_id or rec.get("wait_id", "")
+        except OSError:  # bridge data root unreadable → degrade, never lie
+            waiting_state = "BRIDGE_UNREACHABLE"
+        except RuntimeError:  # waitstore not resolvable in this context
+            waiting_state = "BRIDGE_UNREACHABLE"
+
+    checkpoints = c.get("checkpoints", [])
+    evidence_refs = []
+    for cp in checkpoints:
+        ev = cp.get("evidence", "")
+        if ev:
+            evidence_refs.append(ev[:120])
+
+    approval = trace.get("approval", {})
+    approval_required = bool(
+        approval.get("required") or approval.get("state") == "REQUIRED"
+        or c.get("phase") == "awaiting_approval")
+
+    blockers: list[str] = []
+    blk = ids.get("blocker")
+    if isinstance(blk, dict):
+        blockers.append(f"{blk.get('class', 'BLOCKED')}: "
+                        f"{blk.get('evidence', '')[:120]}")
+    elif isinstance(blk, str) and blk:
+        blockers.append(blk)
+
+    return {
+        "schema_version": SCHEMA_VERSION,
+        "project_id": getattr(scaffold, "project_id", ""),
+        "originating_goal_id": originating_goal_id,
+        "construction_id": contract_id,
+        "contract_id": contract_id,
+        "mode": ids.get("construction_mode", "") or "",
+        "depth": ids.get("depth", "") or "",
+        "phase": c.get("phase", ""),
+        "stage": stage,
+        "operation_id": operation_id,
+        "waiting_state": waiting_state,
+        "correlation_id": live_correlation,
+        "builder_state": builder_state,
+        "blockers": blockers,
+        "approval_required": approval_required,
+        "readiness_canary": ext.get("readiness_canary", ""),
+        "liveness_canary": ext.get("liveness_canary", ""),
+        "exit_canary": ext.get("exit_canary", ""),
+        "verification_state": verification_state,
+        "checkpoint": checkpoint,
+        "evidence_refs": evidence_refs,
+        "failure_class": failure_class,
+        "completion_state": completion_state,
+        "candidate_next_action": ids.get("next_action", ""),
+    }
+
+
+def validate(projection: dict) -> list[str]:
+    """Check a projection against the schema. Returns a list of errors
+    (empty = valid). Consumers use this before trusting a projection."""
+    errors = []
+    if projection.get("schema_version") != SCHEMA_VERSION:
+        errors.append(f"schema_version: expected {SCHEMA_VERSION}, got "
+                      f"{projection.get('schema_version')!r}")
+    for f in FIELDS:
+        if f not in projection:
+            errors.append(f"missing field: {f}")
+    return errors
@@ -187,6 +187,9 @@ class DriftCache:
 # ═══════════════════════════════════════════════════════════════════════════════


+import os
+
+SCRIPTS_DIR = os.environ.get("RUNE_SCRIPTS_DIR", os.path.join(os.path.dirname(__file__), "scripts"))
 def main():
    import argparse

@@ -350,63 +350,6 @@ def pre_commit_gate_entry(entry: dict) -> dict:
        "critics": critics,
        "reason": "all critics passed" if all_pass else "one or more critics failed",
    }
-    row = db.execute(
-        "SELECT id, text, importance, wash_count, rock_status, topic FROM entries WHERE id = ?", (entry_id,)
-    ).fetchone()
-
-    if not row:
-        return {"verdict": "REJECT", "reason": "entry not found"}
-
-    entry = dict(row)
-    critics = {}
-
-    # Critic 1: Structural validity
-    text = entry.get("text", "")
-    if len(text) < 5:
-        critics["structural"] = {"pass": False, "reason": "text too short"}
-    elif len(text) > 2000:
-        critics["structural"] = {"pass": False, "reason": "text too long"}
-    elif not text.strip():
-        critics["structural"] = {"pass": False, "reason": "empty text"}
-    else:
-        critics["structural"] = {"pass": True, "reason": "well-formed"}
-
-    # Critic 2: Behavioral harmlessness (check against immutable core)
-    # The immutable core: user authority, no safety theater, no fights
-    harmful_patterns = ["refuse to", "cannot comply", "safety violation", "I'm sorry, but I can't"]
-    is_harmful = any(p in text.lower() for p in harmful_patterns)
-    critics["behavioral"] = {
-        "pass": not is_harmful,
-        "reason": "contains refusal language" if is_harmful else "aligned with core",
-    }
-
-    # Critic 3: Semantic consistency (check for contradictions with existing rocks)
-    rocks = db.execute("SELECT text FROM entries WHERE rock_status = 'rock' AND status = 'active' LIMIT 50").fetchall()
-    contradiction_found = False
-    for rock in rocks:
-        rock_text = rock[0].lower()
-        entry_text = text.lower()
-        # Simple contradiction detection: if entry says opposite of a rock
-        if ("always" in rock_text and "never" in entry_text) or ("never" in rock_text and "always" in entry_text):
-            contradiction_found = True
-            break
-    critics["semantic"] = {
-        "pass": not contradiction_found,
-        "reason": "contradicts existing rock" if contradiction_found else "consistent",
-    }
-
-    # Overall verdict
-    all_pass = all(c["pass"] for c in critics.values())
-    verdict = "PASS" if all_pass else "REJECT"
-
-    return {
-        "verdict": verdict,
-        "entry_id": entry_id,
-        "critics": critics,
-        "reason": "all critics passed" if all_pass else "one or more critics failed",
-    }
-
-
 # ═══════════════════════════════════════════════════════════════════════════════
 # CLI
 # ═══════════════════════════════════════════════════════════════════════════════
@@ -0,0 +1,118 @@
+"""drift_resume.py — T16 §16P: drift-aware resume for months-later
+project re-entry.
+
+Months-later resume must NOT blindly trust the last checkpoint. The
+16O benchmark proved both cold and scaffold-assisted paths converge on
+the same principle: checkpoint claims NEVER substitute for reality.
+
+Flow:
+
+    load last verified scaffold state (expectations)
+        ↓
+    inspect current repo identity/state (observations)
+        ↓
+    compare expected vs actual  →  DriftRecord list
+        ↓
+    unchanged:  resume
+    drift:      classify EXPECTED / BENIGN / MATERIAL / CONFLICTING /
+                UNKNOWN — MATERIAL/UNKNOWN can block dependent work
+
+Reality wins unless the claim carries explicit user-locked authority
+(the one scaffold exception, already enforced by layout_contract).
+"""
+from __future__ import annotations
+
+from typing import Callable
+
+EXPECTED = "EXPECTED"      # checkpoint older than reality; benign aging
+BENIGN = "BENIGN"          # difference with no bearing on the claims
+MATERIAL = "MATERIAL"      # reality contradicts a scaffold claim
+CONFLICTING = "CONFLICTING"  # two authorities disagree; needs user
+UNKNOWN = "UNKNOWN"        # cannot decide — treated as blocking
+
+BLOCKING_CLASSES = (MATERIAL, CONFLICTING, UNKNOWN)
+
+
+def classify_drift(claim: dict, observation: dict) -> dict:
+    """Classify one claim-vs-observation pair. Deterministic:
+    verification evidence beats prose; explicit beats inferred.
+
+    claim:       {"claim": str, "kind": str, "verified_at": str,
+                  "user_locked": bool}
+    observation: {"actual": str|None, "checked_by": str,
+                  "matches": bool|None}
+    """
+    if observation.get("matches") is True:
+        return {"claim": claim.get("claim", ""),
+                "class": EXPECTED if claim.get("stale_ok") else "CONFIRMED",
+                "evidence": observation.get("checked_by", "")}
+    if observation.get("matches") is None:
+        return {"claim": claim.get("claim", ""), "class": UNKNOWN,
+                "evidence": f"unverifiable via {observation.get('checked_by', '?')}"}
+    if claim.get("user_locked"):
+        return {"claim": claim.get("claim", ""), "class": CONFLICTING,
+                "evidence": ("reality differs from a USER-LOCKED claim "
+                             "— user arbitration required")}
+    return {"claim": claim.get("claim", ""), "class": MATERIAL,
+            "evidence": (f"reality differs: {observation.get('actual', '')}"
+                         f" (checked by {observation.get('checked_by', '?')})")}
+
+
+def reconcile(claims: list[dict], observer: Callable[[dict], dict]) -> dict:
+    """Reconcile a set of checkpoint claims against reality.
+
+    observer(claim) → observation dict (runs the actual check: test
+    command, file read, git query — never trusts the claim text).
+
+    Returns {"resume": bool, "drifts": [...], "blocking": [...]}. The
+    resume decision: drift-aware, not blind — ANY blocking-class drift
+    stops dependent work until reconciled."""
+    drifts = [classify_drift(c, observer(c)) for c in claims]
+    blocking = [d for d in drifts if d["class"] in BLOCKING_CLASSES]
+    return {"resume": not blocking, "drifts": drifts,
+            "blocking": blocking}
+
+
+# ── T18 §18H: structural drift detection (extends drift_resume; no
+# duplicate mechanism). Detects observable drift between a structural
+# model's claims and repository reality. REALITY WINS unless user-locked.
+
+def detect_structural_drift(model, repo_root) -> list[dict]:
+    """Compare structural-model component claims against source reality.
+    Observable cases: source file moved/missing (component identities
+    carry 'source'), entities for files that no longer exist. Returns
+    drift records in the classify_drift vocabulary; reality wins unless
+    the claim is user-locked (authority == 'user' on the entity's
+    evidence or identities.user_locked)."""
+    import os
+    from pathlib import Path as _P
+    root = _P(repo_root)
+    drifts = []
+    for e in model.doc["entities"]:
+        if e.get("kind") != "component":
+            continue
+        src = ((e.get("ext") or {}).get("source")
+               or (e.get("identities") or {}).get("source"))
+        if not src:
+            continue
+        locked = (e.get("identities") or {}).get("user_locked") is True
+        expected = root / src
+        if not expected.exists():
+            # reality: file gone/moved — search one level for a moved twin
+            moved = None
+            if expected.name:
+                for cand in root.rglob(expected.name):
+                    if cand.is_file():
+                        moved = str(cand.relative_to(root))
+                        break
+            drifts.append({
+                "claim": f"{e['id']} source at {src}",
+                "class": "CONFLICTING" if locked else "MATERIAL",
+                "evidence": (f"source missing; moved to {moved}"
+                             if moved else "source file absent from repo"),
+                "reality_wins": not locked,
+                "reconciliation": (f"update source to {moved}" if moved
+                                   else "entity requires re-derivation or "
+                                        "user-locked preservation"),
+            })
+    return drifts
@@ -0,0 +1,517 @@
+"""event_wait.py — T15 §15D: event-driven wait as the CANONICAL DEFAULT
+for long-running construction operations (builders, CI, deploys).
+
+GLOBAL WAITING INVARIANT (tranche brief):
+
+    EVENT DRIVES CONTINUATION.
+    TIMEOUT DETECTS FAILURE OF CONTINUATION.
+
+The W/R bridge (daemon.py + waitstore.py at RUNE_WR_ROOT, default
+~/.local/share/mimo-wait-resume) already provides: durable waits, event
+matching, per-session single-flight leases, process-tree ownership, and a
+resume driver. Until T15 its resume target was exclusively a MiMo
+session (global mimo-shaped resume_cmd), so Python controllers (stage
+drivers, builder sessions) could not be resumed by the bridge — which is
+exactly why fixed sleep/poll loops crept back into T12-T14.
+
+This adapter CLOSES that gap with the smallest possible surface:
+
+1. register_process_wait()  — a controller/operation registers a
+   PROCESS_EVENT wait (correlation: operation_id) whose per-wait
+   resume_cmd writes a resume token to a local FIFO. Registration is a
+   durable waitstore record: survives restarts, visible to reconciliation.
+
+2. run_builder_under_events() — launches the real builder command under
+   a notifier wrapper (separate session/process-group). The wrapper
+   observes the builder's actual exit and POSTs the completion event to
+   the daemon (/process-event). The wrapper is the event source; the
+   driver does NOT wait on the builder directly.
+
+3. await_resume_event() — the driver YIELDS: blocks in select() on the
+   FIFO (no polling loop, no sleep floor) with a watchdog deadline. The
+   daemon matches the event, advances the wait, and the per-wait
+   resume_cmd fires immediately — the driver resumes on the EVENT, not
+   after a configured delay.
+
+4. Watchdog semantics: on timeout the driver performs a ONE-SHOT
+   liveness inspection of the notifier process group (ALIVE → re-arm;
+   GONE → PROCESS_LOST recovery). The timeout is a failure detector,
+   never the scheduler.
+
+Idempotency (15E) is inherited from the bridge: match() applies only to
+WAITING_EXTERNAL waits, so duplicate/late events cannot double-resume;
+correlation by operation_id (run_ref) means a stale event from a prior
+operation matches nothing; the reconciler never verifies PROCESS_EVENT
+waits against Gitea (the event is the authority).
+"""
+
+from __future__ import annotations
+
+import importlib.util
+import json
+import os
+import select
+import signal
+import subprocess
+import sys
+import time
+from pathlib import Path
+from typing import Any
+
+DEFAULT_WR_ROOT = Path.home() / ".local" / "share" / "mimo-wait-resume"
+FALLBACK_WR_ROOT = Path("/home/nick/.local/share/mimo-wait-resume")
+DEFAULT_DAEMON_URL = "http://127.0.0.1:8791"
+
+# The notifier wrapper: runs the real builder command as its child,
+# observes the ACTUAL exit, and POSTs the completion event to the daemon.
+# It is the event SOURCE — deliberately independent of the waiting driver.
+_NOTIFIER_SRC = r"""
+import json, os, subprocess, sys, time, urllib.request
+op = sys.argv[1]; cmd = sys.argv[2:]
+start = time.time()
+proc = subprocess.run(cmd)
+completed_at = time.time()
+payload = {
+    "operation_id": op,
+    "exit_code": proc.returncode,
+    "completed_at": completed_at,
+    "started_at": start,
+}
+for attempt in range(3):
+    try:
+        req = urllib.request.Request(
+            os.environ["T15_DAEMON_URL"] + "/process-event",
+            data=json.dumps(payload).encode(),
+            headers={"Content-Type": "application/json"})
+        with urllib.request.urlopen(req, timeout=5) as r:
+            r.read()
+        break
+    except Exception as e:
+        if attempt == 2:
+            sys.stderr.write(f"event POST failed: {e}\n")
+            # best-effort: the watchdog is the safety net for lost events
+        time.sleep(1)
+"""
+
+
+def wr_root() -> Path:
+    """Resolve the bridge DATA root — DELEGATED (T21.5 §7) to the single
+    shared resolver (scripts/wr_root_resolver.py): explicit env →
+    canonical production root → legacy home; split-brain between live
+    stores fails loudly, never silently operates two stores."""
+    resolver_path = (Path(__file__).resolve().parent.parent.parent
+                     / "scripts" / "wr_root_resolver.py")
+    try:
+        spec = importlib.util.spec_from_file_location(
+            "wr_root_resolver", str(resolver_path))
+        if spec is None or spec.loader is None:
+            raise RuntimeError("no spec/loader")
+        mod = importlib.util.module_from_spec(spec)
+        spec.loader.exec_module(mod)
+        return mod.resolve_root()
+    except (OSError, RuntimeError) as exc:
+        # resolver unavailable (partial checkout): keep the documented
+        # policy rather than failing the wait path
+        sys.stderr.write(f"event_wait: resolver fallback: {exc}\n")
+        env = os.environ.get("RUNE_WR_ROOT")
+        if env:
+            return Path(env)
+        shared = Path("/mnt/workdrive/wr-bridge")
+        if shared.exists():
+            return shared
+        if DEFAULT_WR_ROOT.exists():
+            return DEFAULT_WR_ROOT
+        return FALLBACK_WR_ROOT
+
+
+# waitstore instances cached per data root (isolated tests use per-test
+# tmp roots; the daemon uses its install dir). Loading via importlib.util
+# keeps the dynamic import pyright-clean — no unresolvable-import warning.
+_waitstore_cache: dict[str, Any] = {}
+
+
+def _load_waitstore() -> Any:
+    """Load the bridge's OWN waitstore module (never a copy), bound to
+    the adapter's data root. waitstore.py lives in the DAEMON CODE dir;
+    in production WR_ROOT is that dir (code + data together); in isolated
+    test environments WR_ROOT is a data-only dir, so the code dir is
+    resolved separately: RUNE_WR_CODE env → WR_ROOT → known installs."""
+    root = wr_root()
+    # waitstore binds its DATA dir from WR_ROOT at import time. Pin
+    # WR_ROOT to the adapter's authority so adapter-side and daemon-side
+    # state agree — otherwise a mismatch silently splits the wait store
+    # from the event matcher and continuation dies as a watchdog timeout.
+    os.environ["WR_ROOT"] = str(root)
+    cached = _waitstore_cache.get(str(root))
+    if cached is not None:
+        return cached
+    code_dir = None
+    for cand in (os.environ.get("RUNE_WR_CODE"), str(root),
+                 str(FALLBACK_WR_ROOT)):
+        if cand and (Path(cand) / "waitstore.py").is_file():
+            code_dir = Path(cand)
+            break
+    if code_dir is None:
+        raise RuntimeError(
+            "waitstore.py not found in known W/R roots "
+            "(RUNE_WR_CODE, WR_ROOT, ~/.local/share/mimo-wait-resume)")
+    spec = importlib.util.spec_from_file_location(
+        "waitstore", str(code_dir / "waitstore.py"))
+    if spec is None or spec.loader is None:
+        raise RuntimeError(f"cannot load waitstore from {code_dir}")
+    mod = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(mod)
+    _waitstore_cache[str(root)] = mod
+    return mod
+
+
+def register_process_wait(operation_id: str, *, contract_id: str = "",
+                          project: str = "", objective: str = "",
+                          watched_for: str = "",
+                          provider: str = "builder",
+                          event_type: str = "builder_exit",
+                          sha: str = "",
+                          watchdog_secs: int | None = None,
+                          continuation_state: str = "WAITING_EXTERNAL",
+                          op_type: str = "",
+                          authority: str = "",
+                          task_id: str = "",
+                          resume_mode: str = "fifo") -> dict:
+    """E-PATCH §4: the PendingOperation IS the durable continuation object.
+
+    continuation_state is first-class cognitive quiescence:
+      ACTIVE_REASONING | WAITING_EXTERNAL | RESUMABLE | RECOVERY_REQUIRED
+    op_type/authority record the CANONICAL execution semantics (derived by
+    intent normalization, never model-supplied). task_id/project bind the
+    operation to the current programme so fresh-process recovery can
+    resolve relevance without the model reconstructing state.
+
+    resume_mode owns the SESSION-DELIVERY domain (REAL_ASYNC_BRIDGE §2/§3):
+      "fifo" — legacy delivery: the daemon writes a resume token to the
+               operation's FIFO on event match (T15 semantics; the waiter
+               blocks reading the FIFO). Default preserves T15 builder
+               behavior.
+      "self" — the waiter process itself blocks on the durable completion
+               file (inotify) and stamps its own terminal operation state
+               on authority verification. NO resume_cmd is registered: the
+               daemon must never dispatch a delivery worker for this wait,
+               and a dead reader cannot downgrade operation truth."""
+    """Register a durable PROCESS_EVENT wait for one operation. The
+    per-wait resume_cmd writes a resume token to the operation's FIFO —
+    the bridge can resume a Python controller, not just MiMo sessions.
+
+    provider/event_type are the adapter's event identity on the shared
+    bridge (16E: ONE wait primitive; builder / Gitea-CI / future adapters
+    differ ONLY in these identity fields + their liveness probe — the
+    registration, FIFO rendezvous, matching, idempotency, restart
+    recovery, and watchdog semantics are the same code for all).
+    Defaults preserve the T15 builder semantics."""
+    ws = _load_waitstore()
+    fifo = fifo_path(operation_id)
+    fifo.parent.mkdir(parents=True, exist_ok=True)
+    # Do NOT unlink+recreate an existing fifo: a caller that opened its
+    # read end before registering (open_fifo_reader, §16Y #10 race fix)
+    # would be orphaned onto the old inode while the daemon's resume
+    # writer opened the new one. The rendezvous is create-if-missing.
+    if not fifo.exists():
+        os.mkfifo(fifo)
+        os.chmod(fifo, 0o666)  # cross-user bridge (see open_fifo_reader)
+    writer = ("import sys,json,time,os;"
+              "fd=os.open(sys.argv[1], os.O_WRONLY | os.O_NONBLOCK);"
+              "os.write(fd, json.dumps({'wait_id': sys.argv[2],"
+              " 'resumed_at': time.time()}).encode());os.close(fd)")
+    if resume_mode == "self":
+        # REAL_ASYNC_BRIDGE §3: no session-delivery worker. The waiter
+        # self-resumes on the durable completion file; the daemon's only
+        # roles are event persistence and failure-detection sweeps.
+        resume_cmd = None
+    else:
+        resume_cmd = ["python3", "-c", writer, str(fifo), "{wait_id}"]
+    rec = ws.new_wait(
+        session_id=f"controller-{operation_id}",
+        provider=provider,
+        event_type=event_type,
+        sha=sha,
+        repo=project or "construction",
+        run_ref=operation_id,
+        expected="success",
+        wait_kind="PROCESS_EVENT",
+        target_kind="headless",   # owned continuation target: FIFO writer
+        objective=objective or f"builder operation {operation_id}",
+        waited_for=watched_for or "builder process exit event",
+        verify_when_woken="reload canonical state; exit canary; "
+                          "independent verification",
+        on_success="continue programme continuation canary",
+        on_failure="classify failure; recovery or block",
+        resume_cmd=resume_cmd,
+    )
+    rec["ext"] = {"contract_id": contract_id, "operation_id": operation_id,
+                  "fifo": str(fifo),
+                  "watchdog_secs": watchdog_secs
+                  or int(os.environ.get("WR_WATCH_SECS", "600")),
+                  "continuation_state": continuation_state,
+                  "op_type": op_type, "authority": authority,
+                  "task_id": task_id,
+                  # §3 delivery-domain marker: the daemon drive loop skips
+                  # waits marked "self" (the waiter owns its continuation);
+                  # records WITHOUT the marker remain eligible for the
+                  # env-template session-delivery path (WR_RESUME_CMD).
+                  "resume_mode": resume_mode}
+    ws.save(rec)
+    # cross-user bridge (T16 §16C live finding): the daemon reconciler may
+    # run under a different account than the registering controller — the
+    # durable record must be readable/writable on both sides
+    try:
+        os.chmod(ws.WAITS / f"{rec['wait_id']}.json", 0o666)
+    except OSError:
+        pass  # single-user deployment: nothing to relax
+    return rec
+
+
+def fifo_path(operation_id: str) -> Path:
+    return wr_root() / "resumes" / f"{operation_id}.fifo"
+
+
+def open_fifo_reader(operation_id: str) -> int:
+    """Open the operation's FIFO read end (O_NONBLOCK) BEFORE the wait is
+    registered. §16Y adversarial #10: an event that arrives after
+    registration but before the controller enters select() would find no
+    reader — the resume write fails (ENXIO) and the wait lands
+    RESUME_UNVERIFIED while the controller sleeps until the watchdog.
+    Holding the reader open across the registration window closes that
+    race for every adapter; the fd is passed to await_resume_event and
+    closed by the caller."""
+    fifo = fifo_path(operation_id)
+    fifo.parent.mkdir(parents=True, exist_ok=True)
+    if fifo.exists():
+        fifo.unlink()
+    os.mkfifo(fifo)
+    # CROSS-USER BRIDGE (T17 17C root-cause, re-landed): the T16 chmod was
+    # lost in a rebase resolution — FIFOs landed 644 (666 & umask 022),
+    # and the daemon running as a DIFFERENT user got EACCES on the resume
+    # write (rc=1 → RESUME_UNVERIFIED) while same-user tests passed.
+    # chmod AFTER mkfifo is the only reliable act — umask never applies.
+    os.chmod(fifo, 0o666)
+    return os.open(fifo, os.O_RDONLY | os.O_NONBLOCK)
+
+
+def run_builder_under_events(builder_cmd: list[str], operation_id: str, *,
+                             daemon_url: str = DEFAULT_DAEMON_URL):
+    """Launch the REAL builder under the event notifier (separate process
+    group). Returns the Popen for the notifier (pgid == its pid). The
+    driver must NOT wait on this directly — it yields via
+    await_resume_event; the notifier's exit event comes back through the
+    bridge."""
+    env = dict(os.environ, T15_DAEMON_URL=daemon_url)
+    log_file = wr_root() / "resumes" / f"{operation_id}.builder.log"
+    log_file.parent.mkdir(parents=True, exist_ok=True)
+    with open(log_file, "w") as lf:
+        return subprocess.Popen(
+            [sys.executable, "-c", _NOTIFIER_SRC, operation_id, *builder_cmd],
+            stdout=lf, stderr=subprocess.PIPE,
+            text=True, start_new_session=True, env=env)
+
+
+def _pgroup_alive(pgid: int) -> bool:
+    """Liveness = at least one NON-ZOMBIE member of the process group.
+    A killed-but-unreaped notifier is a zombie and does NOT count —
+    treating zombies as alive would misclassify PROCESS_LOST as alive."""
+    try:
+        os.killpg(pgid, 0)  # group exists at all?
+    except (ProcessLookupError, PermissionError):
+        return False
+    for proc in Path("/proc").iterdir():
+        if not proc.name.isdigit():
+            continue
+        try:
+            stat = (proc / "stat").read_text()
+            # field 5 is state; comm (field 2) may contain spaces/parens
+            state = stat.rsplit(")", 1)[1].split()[0]
+            pgid_field = int(stat.rsplit(")", 1)[1].split()[2])
+        except (OSError, ValueError, IndexError):
+            continue
+        if pgid_field == pgid and state != "Z":
+            return True
+    return False
+
+
+def _cmdline_alive(needle: str) -> bool:
+    """Liveness fallback: any live process whose cmdline contains the
+    needle. Real builders (node CLIs) may re-group their children, so
+    pgid alone can false-negative — T15Q observed a 65-minute real build
+    outliving the watchdog AND invisible to the pgid scan. The needle
+    should be operation-specific (the prompt marker), not generic."""
+    for proc in Path("/proc").iterdir():
+        if not proc.name.isdigit():
+            continue
+        try:
+            cmdline = (proc / "cmdline").read_bytes().decode(
+                errors="replace")
+            stat = (proc / "stat").read_text()
+            state = stat.rsplit(")", 1)[1].split()[0]
+        except (OSError, ValueError, IndexError):
+            continue
+        if needle in cmdline and state != "Z":
+            return True
+    return False
+
+
+def liveness(operation_id: str, pgid: int | None = None) -> bool:
+    """Watchdog helper: is the watched operation actually alive? pgid
+    membership first; cmdline evidence as the regroup-tolerant fallback."""
+    if pgid and _pgroup_alive(pgid):
+        return True
+    return _cmdline_alive(f"op:{operation_id}")
+
+
+def await_resume_event(operation_id: str, *, watchdog_secs: int,
+                       pgid: int | None = None,
+                       max_rearms: int = 20,
+                       fifo_fd: int | None = None) -> dict:
+    """YIELD until the bridge resumes this controller. Blocks in select()
+    on the FIFO — no sleep, no polling.
+
+    fifo_fd: a pre-opened read end from open_fifo_reader() (§16Y #10 race
+    fix — the fd must already exist when the wait becomes registered).
+    Caller-owned fds are never closed here; on EOF (writer gone, no
+    token) the call returns {"kind": "eof"} instead of busy-looping on an
+    always-readable fd — the caller falls back to the durable waitstore
+    record via resume_or_recover/operation_wait.
+
+    Watchdog semantics (§15D): the deadline is a FAILURE DETECTOR, never
+    the scheduler. On expiry, liveness is inspected ONCE (pgid membership,
+    then operation-specific cmdline evidence — real builders regroup their
+    children): ALIVE → re-arm (select again, up to max_rearms); GONE →
+    PROCESS_LOST. The T15Q live run proved why: a real 65-minute model
+    build outlived a single 900s deadline, and node regrouping hid the
+    processes from the pgid scan — neither is a failure of continuation.
+    """
+    owned_fd = fifo_fd is None
+    if owned_fd:
+        fifo = fifo_path(operation_id)
+        # idempotent re-arm: a prior await consumed+unlinked the fifo; the
+        # rendezvous point is (re)created so watchdog re-arm and recovery
+        # paths can wait again on the same operation
+        fifo.parent.mkdir(parents=True, exist_ok=True)
+        if not fifo.exists():
+            os.mkfifo(fifo)
+        fd = os.open(fifo, os.O_RDONLY | os.O_NONBLOCK)
+    else:
+        fd = fifo_fd
+    try:
+        for arm in range(max_rearms + 1):
+            deadline = time.monotonic() + watchdog_secs
+            while True:
+                remaining = deadline - time.monotonic()
+                if remaining <= 0:
+                    break
+                r, _, _ = select.select([fd], [], [], remaining)
+                if not r:
+                    continue
+                data = os.read(fd, 4096)
+                if not data:
+                    if owned_fd:
+                        continue
+                    # caller-owned fd at EOF: writer is gone without
+                    # delivering a token — never a busy loop; the durable
+                    # record decides what actually happened
+                    return {"kind": "eof"}
+                try:
+                    token = json.loads(data.decode())
+                except ValueError:
+                    break  # malformed token → treat as expiry
+                ws = _load_waitstore()
+                rec = ws.load(token.get("wait_id", ""))
+                return {"kind": "event", "wait": rec,
+                        "resumed_at": token.get("resumed_at")}
+            # deadline expired: liveness decides re-arm vs PROCESS_LOST
+            alive = liveness(operation_id, pgid)
+            if not alive:
+                return {"kind": "watchdog", "alive": False,
+                        "rearms": arm}
+        return {"kind": "watchdog", "alive": liveness(operation_id, pgid),
+                "rearms": max_rearms}
+    finally:
+        if owned_fd:
+            os.close(fd)
+
+
+def operation_wait(operation_id: str) -> dict | None:
+    ws = _load_waitstore()
+    for rec in ws.all_waits():
+        if rec.get("ext", {}).get("operation_id") == operation_id:
+            return rec
+    return None
+
+
+def resume_or_recover(operation_id: str) -> dict:
+    """15E case 10 — controller restart after the event was persisted but
+    before the controller handled it. The durable waitstore record is the
+    authority, not the transient FIFO token: if the wait already reached a
+    terminal/resumable state, return it directly (idempotent recovery, no
+    duplicate transition); if still WAITING_EXTERNAL, re-arm the normal
+    event wait."""
+    rec = operation_wait(operation_id)
+    if rec is None:
+        return {"kind": "unknown_operation"}
+    if rec.get("status") in ("COMPLETED", "FAILED", "READY_TO_RESUME",
+                             "READY_TO_RETRY", "RESUMING"):
+        # RESUMING counts as recovered: the resume delivery is in flight
+        # (or already consumed the rendezvous). The durable record is the
+        # truth; the controller restarts from it, never from the token.
+        return {"kind": "event", "wait": rec, "recovered": True}
+    if rec.get("status") == "RESUME_UNVERIFIED" and \
+            rec.get("process_event_evidence"):
+        # the completion EVENT was processed (verdict recorded durably);
+        # only the resume-token delivery was unverified (e.g. controller
+        # was down, so no reader consumed the rendezvous). Canonical
+        # state wins — this IS the restart-after-event case.
+        return {"kind": "event", "wait": rec, "recovered": True}
+    return {"kind": "still_waiting", "wait": rec}
+
+
+def find_matching_event(provider: str, event_type: str, run_ref: str) -> dict | None:
+    """Search the bridge's durable event log for an event matching this
+    identity. T16 §16C live finding: on a serialized single-runner
+    machine the deploy job ALWAYS starts after ci.yml completes — the
+    completion event lands before the waiter registers, and a pure
+    forward-only FIFO rendezvous can never deliver it (both production
+    cycles fell back to the 900s deadline). Events are durable precisely
+    so they can be replayed: a waiter that finds a matching event at
+    registration goes straight to the authority read instead of sleeping
+    for a wake that already happened."""
+    import json as _json
+    ws = _load_waitstore()
+    events_dir = ws.EVENTS
+    if not events_dir.is_dir():
+        return None
+    best = None
+    for f in sorted(events_dir.glob("*.json"),
+                    key=lambda p: p.stat().st_mtime, reverse=True):
+        try:
+            rec = _json.loads(f.read_text())
+        except (OSError, ValueError):
+            continue
+        ev = rec.get("event", {})
+        if (ev.get("provider") == provider
+                and ev.get("event_type") == event_type
+                and str(ev.get("run_ref", "")) == run_ref):
+            best = ev  # newest match wins
+            break
+    return best
+
+
+def post_process_event(operation_id: str, exit_code: int, *,
+                       daemon_url: str = DEFAULT_DAEMON_URL) -> dict:
+    """Direct event injection (test seam + manual reconciliation). The
+    normal path is the notifier wrapper; this exists so tests and
+    watchdog recovery can emit the same event shape."""
+    import urllib.request
+    payload = {"operation_id": operation_id, "exit_code": exit_code,
+               "completed_at": time.time()}
+    req = urllib.request.Request(
+        daemon_url + "/process-event",
+        data=json.dumps(payload).encode(),
+        headers={"Content-Type": "application/json"})
+    with urllib.request.urlopen(req, timeout=5) as r:
+        return json.loads(r.read())
@@ -192,7 +192,8 @@ def _default_runner(prompt: str, artifact_path: str) -> None:
    import subprocess

    mimo_bin = os.environ.get("MIMO_BIN", "/var/lib/gitea-desktop-runner/bin/mimo")
-    mimo_model = os.environ.get("MIMO_MODEL", "Verboo Code/glm-5.3-flash")
+    from rune.provider_context import require_model
+    mimo_model = require_model()  # T16 §16F: provider layer owns selection
    cmd = [
        mimo_bin,
        "run",
@@ -0,0 +1,108 @@
+"""failure_recovery.py — T15 §15T: failure classification + recovery
+policy.
+
+Before retrying, classify the failure:
+
+    TRANSIENT            bounded immediate retry appropriate
+    INTERMITTENT         bounded retry/backoff
+    KNOWN_PROCESS_DEFECT apply evidenced repair, then retry
+    PERMANENT / INVALID  do NOT retry blindly
+    APPROVAL_REQUIRED    block
+    UNKNOWN              bounded evidence probe, then block/escalate
+
+No infinite retry loops: every class carries a retry budget, and
+PERMANENT/APPROVAL_REQUIRED never auto-retry.
+"""
+
+from __future__ import annotations
+
+TRANSIENT = "TRANSIENT"
+INTERMITTENT = "INTERMITTENT"
+KNOWN_PROCESS_DEFECT = "KNOWN_PROCESS_DEFECT"
+PERMANENT = "PERMANENT"
+APPROVAL_REQUIRED = "APPROVAL_REQUIRED"
+PROVIDER_UNAVAILABLE = "PROVIDER_UNAVAILABLE"
+UNKNOWN = "UNKNOWN"
+
+CLASSES = (TRANSIENT, INTERMITTENT, KNOWN_PROCESS_DEFECT, PERMANENT,
+           APPROVAL_REQUIRED, PROVIDER_UNAVAILABLE, UNKNOWN)
+
+# recovery policy per class: (max_retries, action)
+POLICY = {
+    TRANSIENT: (2, "retry_immediately"),
+    INTERMITTENT: (3, "retry_with_backoff"),
+    KNOWN_PROCESS_DEFECT: (1, "apply_evidenced_repair_then_retry"),
+    PERMANENT: (0, "do_not_retry"),
+    APPROVAL_REQUIRED: (0, "block"),
+    # T16 §16F: provider failures NEVER auto-retry and NEVER fall back
+    # to another provider — preserve the real error, block the branch.
+    PROVIDER_UNAVAILABLE: (0, "block_branch_preserve_error"),
+    UNKNOWN: (1, "bounded_evidence_probe_then_block_or_escalate"),
+}
+
+# deterministic markers (substring, lowercase) — the classifier is
+# evidence-first: explicit signals win; absence defaults to UNKNOWN
+_TRANSIENT_MARKERS = (
+    "connection reset", "timeout", "timed out", "temporarily unavailable",
+    "rate limit", "429", "503", "econnreset", "socket hang up",
+    "service unavailable",
+)
+_INTERMITTENT_MARKERS = (
+    "flaky", "intermittent", "occasionally fails", "sometimes fails",
+    "retry succeeded",
+)
+_PERMANENT_MARKERS = (
+    "invalid input", "schema-invalid", "not found", "no such file",
+    "permission denied", "unsupported", "incompatible",
+    "module not found", "syntaxerror", "typeerror",
+)
+_APPROVAL_MARKERS = (
+    "approval required", "needs authorization", "user confirmation",
+    "destructive", "external service", "new repository",
+    "user-locked", "user locked",
+)
+_KNOWN_DEFECT_MARKERS = (
+    "stale document", "forgotten save", "wrong path", "guessed api",
+    "missing import", "no-op edit", "substring match", "wrong signature",
+)
+
+
+def classify_failure(error_text: str, *, context: dict | None = None) -> dict:
+    """Classify one failure into a recovery class. Deterministic: marker
+    evidence first; explicit context hints (e.g. known defect id) second;
+    UNKNOWN when nothing matches (never manufactured certainty)."""
+    text = (error_text or "").lower()
+    context = context or {}
+    known_defect = context.get("known_defect_id")
+    if known_defect:
+        return {"class": KNOWN_PROCESS_DEFECT,
+                "evidence": f"known defect: {known_defect}",
+                "policy": POLICY[KNOWN_PROCESS_DEFECT]}
+    # T16 §16F: an explicit provider-context failure wins over every
+    # generic marker — the real status/API code is preserved verbatim in
+    # the evidence, the branch blocks, and no fallback is attempted.
+    if context.get("provider_error"):
+        return {"class": PROVIDER_UNAVAILABLE,
+                "evidence": (f"provider error preserved: {error_text}"
+                             if error_text else "provider error"),
+                "policy": POLICY[PROVIDER_UNAVAILABLE]}
+    if any(m in text for m in _APPROVAL_MARKERS):
+        return {"class": APPROVAL_REQUIRED,
+                "evidence": "approval marker present",
+                "policy": POLICY[APPROVAL_REQUIRED]}
+    if any(m in text for m in _PERMANENT_MARKERS):
+        return {"class": PERMANENT, "evidence": "permanent marker present",
+                "policy": POLICY[PERMANENT]}
+    if any(m in text for m in _KNOWN_DEFECT_MARKERS):
+        return {"class": KNOWN_PROCESS_DEFECT,
+                "evidence": "known process-defect marker present",
+                "policy": POLICY[KNOWN_PROCESS_DEFECT]}
+    if any(m in text for m in _INTERMITTENT_MARKERS):
+        return {"class": INTERMITTENT,
+                "evidence": "intermittent marker present",
+                "policy": POLICY[INTERMITTENT]}
+    if any(m in text for m in _TRANSIENT_MARKERS):
+        return {"class": TRANSIENT, "evidence": "transient marker present",
+                "policy": POLICY[TRANSIENT]}
+    return {"class": UNKNOWN, "evidence": "no marker matched",
+            "policy": POLICY[UNKNOWN]}
@@ -0,0 +1,146 @@
+"""forge.py — T18 §18L: minimum stable forge abstraction boundary.
+
+Rune today is Gitea-centric; Scaffolding must not make Gitea itself part
+of canonical construction semantics. This module defines the boundary:
+Git remains the source-history authority; a Forge adapter handles the
+remote-services surface (push/CI status/PR/protected checks/release
+metadata). GiteaForge is the real working adapter, delegating to the
+existing gitea_auth + scripts (rule A: no duplicate implementation).
+
+A future Forgejo/GitHub/GitLab adapter implements the same interface
+without changing Scaffolding's core model.
+"""
+from __future__ import annotations
+
+from typing import Protocol
+
+
+class ForgeAdapter(Protocol):
+    """The minimum stable forge boundary. All methods may raise
+    ForgeUnavailable — callers treat forge failures as blocking-for-that-
+    operation, never as construction-truth changes."""
+
+    def push(self, repo: str, branch: str) -> dict: ...
+
+    def commit_status(self, repo: str, sha: str) -> str: ...
+
+    def ci_runs_for_sha(self, repo: str, sha: str) -> list[dict]: ...
+
+    def create_or_patch_release(self, repo: str, tag: str,
+                                title: str, body: str) -> dict: ...
+
+
+class ForgeUnavailable(RuntimeError):
+    """The forge could not be reached or refused the operation."""
+
+
+class GiteaForge:
+    """Real adapter. Delegates to the existing canonical implementations
+    (gitea_auth credential ladder; deploy_qualify verdict logic) — no
+    duplicated API clients (rule A)."""
+
+    def __init__(self, server: str | None = None, repo: str = "conductor/rune"):
+        self._server = server  # None → scripts' env/credential resolution
+        self._repo = repo
+
+    def _env(self) -> dict:
+        import os
+        env = dict(os.environ)
+        if self._server:
+            env["GITEA_SERVER"] = self._server
+        env.setdefault("GITEA_REPO", self._repo)
+        return env
+
+    def push(self, repo: str, branch: str) -> dict:
+        import subprocess
+        proc = subprocess.run(["git", "push", "gitea", f"master:{branch}"],
+                              capture_output=True, text=True)
+        if proc.returncode != 0:
+            raise ForgeUnavailable(f"push failed: {proc.stderr[:200]}")
+        return {"pushed": True, "branch": branch}
+
+    @staticmethod
+    def _gitea_auth():
+        """Canonical scripts/gitea_auth credential ladder (rule A: no
+        duplicate auth). scripts/ is not a package — path shim, the
+        established spec-loaded-scripts pattern."""
+        import sys
+        from pathlib import Path as _P
+        scripts = _P(__file__).resolve().parents[2] / "scripts"
+        if str(scripts) not in sys.path:
+            sys.path.insert(0, str(scripts))
+        import gitea_auth
+        return gitea_auth
+
+    def commit_status(self, repo: str, sha: str) -> str:
+        import json
+        import os
+        import urllib.request
+        gitea_auth = self._gitea_auth()
+        server = self._server or os.environ.get(
+            "GITEA_SERVER", "https://gitea.thingscouldgetdicey.online")
+        if not gitea_auth.has_token():
+            raise ForgeUnavailable("no forge credentials (gitea token)")
+        req = urllib.request.Request(
+            f"{server}/api/v1/repos/{repo}/commits/{sha}/status",
+            headers={"Authorization": f"token {gitea_auth.get_token()}"})
+        try:
+            with urllib.request.urlopen(req, timeout=15) as r:
+                data = json.loads(r.read())
+        except OSError as exc:
+            raise ForgeUnavailable(f"status unreachable: {exc}") from exc
+        return data.get("state", "unknown")
+
+    def ci_runs_for_sha(self, repo: str, sha: str) -> list[dict]:
+        import json
+        import os
+        import urllib.request
+        gitea_auth = self._gitea_auth()
+        server = self._server or os.environ.get(
+            "GITEA_SERVER", "https://gitea.thingscouldgetdicey.online")
+        if not gitea_auth.has_token():
+            raise ForgeUnavailable("no forge credentials (gitea token)")
+        req = urllib.request.Request(
+            f"{server}/api/v1/repos/{repo}/actions/runs?limit=50",
+            headers={"Authorization": f"token {gitea_auth.get_token()}"})
+        try:
+            with urllib.request.urlopen(req, timeout=15) as r:
+                data = json.loads(r.read())
+        except OSError as exc:
+            raise ForgeUnavailable(f"runs unreachable: {exc}") from exc
+        return [run for run in data.get("workflow_runs", [])
+                if str(run.get("head_sha", "")) == sha]
+
+    def create_or_patch_release(self, repo: str, tag: str,
+                                title: str, body: str) -> dict:
+        import json
+        import os
+        import urllib.request
+        gitea_auth = self._gitea_auth()
+        server = self._server or os.environ.get(
+            "GITEA_SERVER", "https://gitea.thingscouldgetdicey.online")
+        if not gitea_auth.has_token():
+            raise ForgeUnavailable("no forge credentials (gitea token)")
+        # delegate to the canonical idempotent publisher
+        import subprocess
+        import sys
+        script = None
+        for cand in ("scripts/create_release.py",):
+            if (sys.path and __import__("pathlib").Path(cand).is_file()):
+                script = cand
+                break
+        if script:
+            proc = subprocess.run(
+                [sys.executable, script, "--tag", tag, "--title", title,
+                 "--body", body], capture_output=True, text=True)
+            if proc.returncode != 0:
+                raise ForgeUnavailable(
+                    f"release publication failed rc={proc.returncode}: "
+                    f"{(proc.stderr or proc.stdout)[-200:]}")
+            return {"published": True, "tag": tag}
+        raise ForgeUnavailable("canonical release publisher not found")
+
+
+def default_forge() -> ForgeAdapter:
+    """The working adapter (rule 18L: prove Gitea still works)."""
+    return GiteaForge()
@@ -0,0 +1,108 @@
+"""graph_projection.py — T18 §18Q: stable versioned READ-ONLY graph
+projection from a structural Scaffold, suitable for later local-web
+visualisation, GUI, TUI tree, or programmatic queries.
+
+This is a PROJECTION — never another authority (rule 18Q). Structural
+truth remains the Scaffold (structural.json) / Spine. Built THROUGH the
+self-scaffolded upgrade (contract 18q-graph-projection): the upgrade
+started from the maintained Scaffold, per the 18P proof.
+
+Formats: versioned JSON (programmatic), DOT and Mermaid (visual —
+dependency-cruiser-proven human-usable formats per the 18D research).
+"""
+from __future__ import annotations
+
+PROJECTION_SCHEMA_VERSION = 1
+
+
+def project_json(model) -> dict:
+    """Stable versioned JSON projection: nodes + edges, no internal
+    storage layout leaked (Substrate-safe seam, rule 18E)."""
+    nodes = []
+    for e in model.doc["entities"]:
+        nodes.append({
+            "id": e["id"],
+            "kind": e.get("kind"),
+            "maturity": e.get("maturity"),
+            "name": (e.get("identities") or {}).get("name"),
+            "ext": {k: v for k, v in (e.get("ext") or {}).items()
+                    if k in ("source", "bounded_adapter")},
+        })
+    edges = []
+    for r in model.doc["relationships"]:
+        edges.append({
+            "type": r.get("type"),
+            "from": r.get("from"),
+            "to": r.get("to"),
+            "authority": r.get("authority"),
+            "status": r.get("status"),
+        })
+    return {
+        "schema_version": PROJECTION_SCHEMA_VERSION,
+        "project": getattr(model, "project_id", ""),
+        "generated_at": _now(),
+        "node_count": len(nodes),
+        "edge_count": len(edges),
+        "nodes": nodes,
+        "edges": edges,
+    }
+
+
+def project_dot(model) -> str:
+    """Graphviz DOT — clusters per domain, edges labeled by type."""
+    lines = ["digraph structural {", "  rankdir=LR;",
+             '  node [shape=box, style="rounded,filled", fillcolor="#eef"];']
+    for e in model.doc["entities"]:
+        eid = e["id"].replace('"', "'")
+        label = (e.get("identities") or {}).get("name") or \
+            e["id"].split(":", 1)[-1]
+        lines.append(f'  "{eid}" [label="{label} ({e["kind"]})"];')
+    for r in model.doc["relationships"]:
+        if r["type"] == "belongs_to" and r.get("status") == "root":
+            continue
+        a = str(r["from"]).replace('"', "'")
+        b = str(r["to"]).replace('"', "'")
+        lines.append(f'  "{a}" -> "{b}" [label="{r["type"]}"];')
+    lines.append("}")
+    return "\n".join(lines)
+
+
+def project_mermaid(model) -> str:
+    """Mermaid flowchart — renders on GitHub/GitLab/many viewers."""
+    lines = ["graph LR"]
+    ids = {e["id"] for e in model.doc["entities"]}
+    safe = {}
+    for i, eid in enumerate(sorted(ids)):
+        safe[eid] = f"n{i}"
+    for e in model.doc["entities"]:
+        label = ((e.get("identities") or {}).get("name")
+                 or e["id"].split(":", 1)[-1]).replace('"', "'")
+        lines.append(f'  {safe[e["id"]]}["{label} ({e["kind"]})"]')
+    for r in model.doc["relationships"]:
+        if r["type"] == "belongs_to" and r.get("status") == "root":
+            continue
+        if r["from"] in safe and r["to"] in safe:
+            lines.append(f'  {safe[r["from"]]} -->|{r["type"]}| '
+                         f'{safe[r["to"]]}')
+    return "\n".join(lines)
+
+
+def write_all(model, out_dir) -> dict:
+    """Write all three formats; returns paths. Read-only over truth —
+    writing projections never mutates the model."""
+    import json
+    from pathlib import Path
+    out = Path(out_dir)
+    out.mkdir(parents=True, exist_ok=True)
+    j = out / "structural_graph.json"
+    d = out / "structural_graph.dot"
+    mm = out / "structural_graph.mmd"
+    j.write_text(json.dumps(project_json(model), indent=1, sort_keys=True))
+    d.write_text(project_dot(model))
+    mm.write_text(project_mermaid(model))
+    return {"json": str(j), "dot": str(d), "mermaid": str(mm)}
+
+
+def _now() -> str:
+    import time
+    return time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
@@ -0,0 +1,69 @@
+"""impact_preview.py — T18 §18G: visible change-impact preview.
+
+For consequential changes, a user-visible STRUCTURED projection of what
+a construction will touch — BEFORE build. Derived from the structural
+Scaffold's impact closure + the Build Basis summary. Contains
+structured conclusions/evidence only — NEVER private chain-of-thought
+(rule 18G). This is a projection, not an authority.
+"""
+from __future__ import annotations
+
+PREVIEW_SCHEMA_VERSION = 1
+
+
+def build_preview(model, closure: dict, *, basis: dict | None = None,
+                  contract_scope: list[str] | None = None) -> dict:
+    """Assemble the preview from an impact_closure (structural_model)
+    and optionally the Build Basis summary. Pure derivation — no hidden
+    reasoning, no mutation."""
+    if closure.get("error"):
+        return {"schema_version": PREVIEW_SCHEMA_VERSION,
+                "error": closure["error"]}
+    scope = set(closure.get("mutation_scope", []))
+    observation = closure.get("observation_scope") or {}
+    deps = observation.get("direct_dependencies", [])
+    # candidate files: identities' source paths if modeled
+    candidate_files = []
+    for eid in sorted(scope):
+        e = model.get(eid) or {}
+        src = (e.get("identities") or {}).get("source")
+        if src:
+            candidate_files.append(src)
+    return {
+        "schema_version": PREVIEW_SCHEMA_VERSION,
+        "target": closure.get("target"),
+        "affected_domains": ([closure.get("owning_domain")]
+                             if closure.get("owning_domain") else []),
+        "affected_subsystems": sorted(
+            e for e in scope if str(e).startswith("subsystem:")),
+        "consumers_connectors": sorted(
+            e for e in scope if e != closure.get("target")),
+        "candidate_files": sorted(set(candidate_files))[:50],
+        "contract_scope_additions": sorted(
+            set(contract_scope or []) - set(candidate_files)),
+        "state_implications": closure.get("state_surfaces", []),
+        "environment_runtime_implications": sorted(
+            model.deployment_surfaces(closure["target"])
+            if model.get(closure["target"]) else []),
+        "migration_implications": (basis or {}).get("implications",
+                                                   {}).get("migration", []),
+        "risks_uncertainties": [
+            *(basis or {}).get("risks", []),
+            *closure.get("uncertainty", []),
+        ],
+        "verification_plan": closure.get("verification_scope", []),
+        "rollback_implications": [
+            "revert contract scope commits; deployment rollbacks via "
+            "last-good symlink (canonical deploy machinery)"],
+        "basis_summary": (basis or {}).get("depth"),
+    }
+
+
+def validate_preview(preview: dict) -> list[str]:
+    errs = []
+    if preview.get("schema_version") != PREVIEW_SCHEMA_VERSION:
+        errs.append("schema_version mismatch")
+    for f in ("target", "verification_plan", "risks_uncertainties"):
+        if f not in preview:
+            errs.append(f"missing {f}")
+    return errs
@@ -0,0 +1,190 @@
+"""iteration.py — T19 §19E/19H/19K: the canonical dependency-aware
+iteration engine — the "quicksand" solution as STANDARD Scaffolding
+behaviour.
+
+Canonical loop (rule 19H):
+
+    intent → target → research/Build Basis → scaffold reconcile →
+    PREDICTIVE impact → bounded mutation scope + wider observation
+    scope → preview → isolated construction → verification →
+    DESCRIPTIVE impact → predicted-vs-observed delta → structural
+    truth update → selective validation invalidation → obligations →
+    iterate → harden.
+
+Built ON the T18 foundation (rule C): structural_model.impact_closure,
+build_basis, impact_preview, drift_resume — no second anything.
+"""
+from __future__ import annotations
+
+import time
+from pathlib import Path
+
+from rune.capability_ledger import CapabilityLedger
+from rune.behaviour_ledger import BehaviourLedger
+
+
+def _now() -> str:
+    return time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+
+
+# ── 19E: surgical context packets ────────────────────────────────────────
+
+def context_packet(model, ledger: CapabilityLedger, target: str, *,
+                   intent: str = "", basis_summary: dict | None = None,
+                   max_local_files: int = 12) -> dict:
+    """LOCAL + CONNECTED context for one structural target (rule 19E).
+    Bounded and proportional: the whole graph is NEVER dumped. Local =
+    the target's own ledger entry; connected = 1-hop structural
+    neighbours with their ledgers on demand."""
+    entity = model.get(target)
+    if entity is None:
+        return {"error": f"unknown structural target {target!r}"}
+    caps = ledger.by_component(target)
+    local = ledger.context_packet(caps[0]["capability_id"]) if caps else None
+    beneath = [b for b in model.beneath(target)
+               if not model.get(b).get("identities", {}).get("source")]
+    files: list[str] = []
+    for eid in [target] + model.children(target):
+        e = model.get(eid) or {}
+        src = (e.get("ext") or {}).get("source") or \
+            (e.get("identities") or {}).get("source")
+        if src:
+            files.append(src)
+    for cap in caps:
+        files.extend(cap.get("source_locations", []))
+    closure = model.impact_closure(target)
+    # interface-mediated consumers: consumers of the interfaces this
+    # target provides (1-hop through the connector, still bounded)
+    iface_consumers = sorted({c for iface in model.interfaces_for(target)
+                              for c in model.consumers(iface)
+                              if c != target})
+    connected = {
+        "dependencies": model.dependencies(target),
+        "consumers": sorted(set(model.consumers(target)) |
+                            set(iface_consumers)),
+        "interface_mediated_consumers": iface_consumers,
+        "interfaces_provided": model.interfaces_for(target),
+        "state_owned": model.state_owned(target),
+        "verification": model.verification_surfaces(target),
+        "deployment": model.deployment_surfaces(target),
+        "uncertainty": closure.get("uncertainty", []),
+    }
+    return {
+        "schema_version": 1,
+        "intent": intent,
+        "target": target,
+        "target_kind": entity.get("kind"),
+        "owning_domain": model.owning_domain(target),
+        "local": local,
+        "local_files": sorted(set(files))[:max_local_files],
+        "beneath_count": len(beneath),
+        "connected": connected,
+        "basis_summary": basis_summary,
+        "generated_at": _now(),
+    }
+
+
+# ── 19H: predictive vs descriptive impact ────────────────────────────────
+
+def predictive_impact(model, target: str) -> dict:
+    """What MAY be affected BEFORE the change (structural closure)."""
+    return model.impact_closure(target)
+
+
+def descriptive_impact(model, target: str, *,
+                       observed_changed: list[str],
+                       generation: str = "") -> dict:
+    """What was MATERIALLY affected AFTER the change — from actual
+    evidence (diff/observations), never assumed from the graph alone."""
+    predicted = model.impact_closure(target)
+    predicted_scope = set(predicted.get("mutation_scope", []))
+    observed = set(observed_changed)
+    return {
+        "target": target,
+        "generation": generation,
+        "predicted": sorted(predicted_scope),
+        "observed": sorted(observed),
+        "unexpected": sorted(observed - predicted_scope),   # prediction missed
+        "unaffected_as_predicted": sorted(predicted_scope - observed),
+        "recorded_at": _now(),
+    }
+
+
+def record_impact_delta(project_store: Path, delta: dict) -> Path:
+    """Persist predicted-vs-observed deltas — they compound into better
+    future impact prediction (rule 19H)."""
+    p = Path(project_store) / "impact_deltas.jsonl"
+    import json
+    with p.open("a") as f:
+        f.write(json.dumps(delta, sort_keys=True) + "\n")
+    return p
+
+
+# ── 19I: selective invalidation bridge ───────────────────────────────────
+
+def invalidate_behaviours(model, behaviours: BehaviourLedger, *,
+                          changed_entities: set[str], generation: str = ""
+                          ) -> list[dict]:
+    """Selective invalidation driven by STRUCTURAL truth: behaviours
+    whose dependencies/owner intersect the changed closure reopen;
+    everything else keeps its confidence (rule 19I)."""
+    return behaviours.invalidate_for_change(
+        changed_entities=changed_entities, generation=generation)
+
+
+# ── 19K: post-change obligations ─────────────────────────────────────────
+
+def post_change_obligations(model, target: str, *,
+                            behaviours: BehaviourLedger | None = None,
+                            changed_entities: set[str] | None = None,
+                            ) -> list[dict]:
+    """Machine-derived follow-up obligations from STRUCTURAL truth
+    (rule 19K) — never an improvised generic checklist."""
+    changed = changed_entities or {target}
+    obligations: list[dict] = []
+    seen: set[tuple] = set()
+
+    def add(kind: str, target_e: str, why: str) -> None:
+        key = (kind, target_e)
+        if key not in seen:
+            seen.add(key)
+            obligations.append({"kind": kind, "target": target_e,
+                                "why": why})
+
+    for eid in sorted(changed):
+        # consumers of a changed component must be inspected
+        for consumer in model.consumers(eid):
+            add("inspect_consumer", consumer,
+                f"consumes/depends on changed {eid}")
+        # provided interfaces: every consumer of the interface is affected
+        for iface in model.interfaces_for(eid):
+            add("verify_interface", iface,
+                f"provided by changed {eid}")
+            for consumer in model.consumers(iface):
+                add("inspect_consumer", consumer,
+                    f"consumes {iface} provided by changed {eid}")
+        # state compatibility
+        for store in model.state_owned(eid):
+            add("check_state_compatibility", store,
+                f"state owned by changed {eid}")
+        # interfaces the target consumes
+        for iface in model.consumers(eid) if False else []:
+            pass
+        consumed = sorted(r["to"] for r in model.doc["relationships"]
+                          if r["type"] == "consumes" and r["from"] == eid)
+        for iface in consumed:
+            add("check_interface_contract", iface,
+                f"changed {eid} consumes {iface} — contract must still hold")
+        # deployment
+        for dep in model.deployment_surfaces(eid):
+            add("verify_deployment", dep,
+                f"deployed surface of changed {eid}")
+    # behaviour retests
+    if behaviours is not None:
+        for rec in behaviours.doc["behaviours"].values():
+            deps = set(rec.get("dependencies") or []) | {
+                rec.get("owner_component")}
+            if deps & changed:
+                add("retest", rec["behaviour_id"],
+                    f"depends on changed {sorted(deps & changed)}")
+    return obligations
@@ -0,0 +1,316 @@
+"""layout_contract.py — persistent UI/TUI layout authority.
+
+SCAFFOLD-TRANCHE-003. Preserves explicit user interface-placement decisions
+as durable structural constraints so future builders must adapt around them.
+
+THE PROBLEM (tranche brief §11): a user says "version information belongs
+HERE"; a model implements it; later unrelated work moves it because another
+model prefers elsewhere. Unacceptable. User-approved layout is STRUCTURE:
+code adapts to the layout, never the reverse.
+
+Authority model:
+    LayoutContract     AUTHORITY — the intended invariant
+    layout observation EVIDENCE  — what an adapter saw at some time
+    region             vocabulary in which contracts are expressed
+
+DRIFT SEMANTICS — deliberate exception to general scaffold drift (§22):
+    General project drift: reality wins, scaffold reconciles.
+    USER-LOCKED layout drift: USER INTENT wins; the observation that
+    violates a locked contract is marked DRIFTING, the contract is NOT
+    rewritten, a maintenance_item is created, and the structural gate
+    fails until repair or explicit user-approved amendment.
+
+Validation is STRUCTURAL (adapter seam), never screenshot-based (§17-18):
+an adapter observes {component: region/order/parent}; the checker compares
+observation to contract. No adapter → UNKNOWN; UNKNOWN is never PASS (§23).
+"""
+
+from __future__ import annotations
+
+from rune.structural_store import add_entity, entity, now, relationships, resolve_status
+
+__all__ = [
+    "define_region",
+    "bind_component",
+    "amend_layout_contract",
+    "record_layout_observation",
+    "check_layout",
+    "layout_drift",
+    "STATUS",
+]
+
+STATUS = {"ACTIVE", "AMENDED", "RETIRED", "DRIFTING"}
+
+
+# ── regions (open-world vocabulary) ───────────────────────────────────────
+
+
+def define_region(
+    doc: dict,
+    region_id: str,
+    *,
+    parent: str = "",
+    order: int | None = None,
+    orientation: str = "",
+    sizing: str = "",
+    visibility: str = "",
+    tech: dict | None = None,
+    provenance: str = "scaffold",
+    evidence_refs: list[str] | None = None,
+) -> dict:
+    """Define/reinforce a layout region. Regions are open-world and
+    project-defined — header/sidebar/main/inspector are examples, not enum.
+    `tech` carries framework metadata (web DOM / Qt widgets / TUI panes)."""
+    identities = {
+        "parent": parent,
+        "orientation": orientation,
+        "sizing": sizing,
+        "visibility": visibility,
+        "provenance": provenance,
+        "tech": dict(tech or {}),
+    }
+    if order is not None:
+        identities["order"] = order
+    return add_entity(
+        doc,
+        id=f"region:{region_id}",
+        kind="layout_region",
+        identities=identities,
+        evidence_refs=evidence_refs or [f"region-defined:{provenance}"],
+    )
+
+
+# ── component identity + contracts ─────────────────────────────────────────
+
+
+def bind_component(
+    doc: dict,
+    component: str,
+    region_id: str,
+    *,
+    lock: bool = False,
+    provenance: str = "scaffold",
+    order: int | None = None,
+    parent_of: str = "",
+    anchor: str = "",
+    sizing: str = "",
+    visibility: str = "",
+    evidence_ref: str = "",
+    implementation: str = "",
+    **framework_attrs,
+) -> dict:
+    """Bind a conceptual UI component (ui:<name>) to an approved layout
+    requirement. The component id is STABLE STRUCTURAL IDENTITY: it survives
+    filename/implementation/framework changes — `implementation` is a
+    refreshable reference, never the key.
+
+    lock=True + provenance=user  → USER-LOCKED: changing requires explicit
+    user-authorized amendment (amend_layout_contract); models may NOT
+    silently override (§14-15).
+
+    Framework-specific detail (CSS-ish or widget-ish) rides in
+    **framework_attrs — open-world, never first-class core fields (§19)."""
+    if not component.startswith("ui:"):
+        component = f"ui:{component}"
+    if entity(doc, f"region:{region_id}") is None:
+        raise ValueError(f"unknown region: {region_id} (define_region first)")
+    if lock and provenance != "user":
+        raise ValueError("lock=True requires user provenance — a scaffold/builder/model can never self-issue a lock")
+    identities = {
+        "component": component,
+        "region": region_id,
+        "lock": lock,
+        "provenance": provenance,
+        "anchor": anchor,
+        "sizing": sizing,
+        "visibility": visibility,
+        "implementation": implementation,
+        "framework_attrs": dict(framework_attrs),
+    }
+    if order is not None:
+        identities["order"] = order
+    if parent_of:
+        identities["parent_of"] = parent_of
+    e = add_entity(
+        doc,
+        id=f"layout:{component}",
+        kind="layout_contract",
+        identities=identities,
+        maturity="DECLARED",
+        evidence_refs=[evidence_ref or f"layout-bind:{provenance}"],
+    )
+    e.setdefault("ext", {})["amendments"] = []
+    e["status"] = "ACTIVE"
+    return e
+
+
+def amend_layout_contract(
+    doc: dict,
+    component: str,
+    *,
+    new_region: str = "",
+    new_order: int | None = None,
+    unlock: bool = False,
+    authorized_by: str = "",
+    reason: str = "",
+    evidence_ref: str = "",
+) -> dict:
+    """Amend a layout contract. A USER-LOCKED contract requires
+    authorized_by='user' — enforced, not documented (§15). Every amendment
+    is recorded on the contract; the original binding remains in history."""
+    cid = component if component.startswith("ui:") else f"ui:{component}"
+    e = entity(doc, f"layout:{cid}")
+    if e is None:
+        raise ValueError(f"no layout contract for {cid}")
+    if e["identities"].get("lock"):
+        if authorized_by != "user":
+            raise ValueError(
+                f"{cid} is USER-LOCKED: amendment requires explicit user "
+                f"authorization (got authorized_by={authorized_by!r})"
+            )
+        if not reason:
+            raise ValueError("user-locked amendment requires a reason")
+    changes: dict = {}
+    if new_region:
+        if entity(doc, f"region:{new_region}") is None:
+            raise ValueError(f"unknown region: {new_region}")
+        e["identities"]["region"] = new_region
+        changes["region"] = new_region
+    if new_order is not None:
+        e["identities"]["order"] = new_order
+        changes["order"] = new_order
+    if unlock:
+        if authorized_by != "user":
+            raise ValueError("unlock requires user authorization")
+        e["identities"]["lock"] = False
+        changes["lock"] = False
+    e["ext"]["amendments"].append(
+        {
+            "at": now(),
+            "authorized_by": authorized_by,
+            "reason": reason[:200],
+            "evidence": evidence_ref,
+            "changes": changes,
+        }
+    )
+    e["observed_at"] = now()
+    return e
+
+
+# ── observations (evidence, NOT authority) ────────────────────────────────
+
+
+def record_layout_observation(doc: dict, adapter: str, observed: dict) -> list[dict]:
+    """Record what a structural adapter saw: {ui_component: {"region": ...,
+    "order": ..., "parent": ...}}. Observations are evidence appended to the
+    store; they never rewrite contracts (§17-18, §22)."""
+    if not adapter:
+        raise ValueError("observation requires adapter identity")
+    out = []
+    for comp, obs in observed.items():
+        c = comp if comp.startswith("ui:") else f"ui:{comp}"
+        out.append(
+            {"kind": "layout_observed", "component": c, "adapter": adapter, "observed": dict(obs), "observed_at": now()}
+        )
+    doc["observations"].extend(out)
+    return out
+
+
+# ── deterministic validation (the gate) ────────────────────────────────────
+
+
+def check_layout(doc: dict, observed: dict) -> dict:
+    """layout_contract_check gate. Verdicts (§23):
+        PASS     — every checked contract complies
+        FAIL     — a contract is verifiably violated (locked violations are
+                   additionally classed USER_LOCKED_VIOLATION — user intent,
+                   not code preference, defines correctness)
+        UNKNOWN  — no observation available for a locked contract (no
+                   adapter evidence) — NEVER converted to PASS
+    Flexible components may be reorganized by builders via amendment, but a
+    live observation contradicting any contract still fails the gate."""
+    results = []
+    verdict = "PASS"
+    for e in doc["entities"]:
+        if e["kind"] != "layout_contract" or e.get("status") == "RETIRED":
+            continue
+        cid = e["identities"]["component"]
+        required_region = e["identities"]["region"]
+        locked = bool(e["identities"].get("lock"))
+        required_order = e["identities"].get("order")
+        obs = observed.get(cid) or observed.get(cid.split("ui:")[-1])
+        if obs is None:
+            results.append(
+                {"component": cid, "locked": locked, "verdict": "UNKNOWN", "reason": "no adapter observation available"}
+            )
+            if locked:
+                verdict = "UNKNOWN" if verdict == "PASS" else verdict
+            continue
+        problems = []
+        got_region = obs.get("region", "")
+        if got_region != required_region:
+            problems.append(f"region {got_region!r} != required {required_region!r}")
+        if required_order is not None and obs.get("order") is not None and obs["order"] != required_order:
+            problems.append(f"order {obs['order']} != required {required_order}")
+        if problems:
+            r = {
+                "component": cid,
+                "locked": locked,
+                "verdict": "FAIL",
+                "reason": "; ".join(problems),
+                "observed": obs,
+                "required_region": required_region,
+            }
+            if locked:
+                r["drift_class"] = "USER_LOCKED_VIOLATION"
+            verdict = "FAIL"  # any contract violation fails the gate (§14)
+            results.append(r)
+        else:
+            results.append({"component": cid, "locked": locked, "verdict": "PASS"})
+    return {"verdict": verdict, "checks": results}
+
+
+def layout_drift(doc: dict, observed: dict) -> dict:
+    """Reconcile layout observations with contracts. USER-LOCKED drift:
+    the CONTRACT is preserved (user intent wins — the deliberate exception
+    to general reality-wins drift), the violating observation is marked,
+    and a maintenance_item records the inconsistency. Flexible drift: just
+    recorded — builder may reorganize within constraints."""
+    record_layout_observation(doc, "drift-check", observed)  # evidence, not authority
+    result = check_layout(doc, observed)
+    drifts = [c for c in result["checks"] if c["verdict"] == "FAIL"]
+    staled = []
+    for d in drifts:
+        le = entity(doc, f"layout:{d['component']}")
+        if le is None:
+            continue
+        if d.get("locked"):
+            le["status"] = "DRIFTING"  # contract preserved, violation marked
+            staled.append(d["component"])
+            add_entity(
+                doc,
+                id=f"maintenance:layout-drift-{d['component']}",
+                kind="maintenance_item",
+                identities={
+                    "issue": f"locked layout contract violated: {d['reason']}",
+                    "evidence": f"adapter observation {d.get('observed')}",
+                    "approval": "user",
+                    "priority": "high",
+                    "status": "open",
+                },
+                evidence_refs=["layout-drift"],
+            )
+        le.setdefault("ext", {})["last_violation"] = {"at": now(), "observed": d.get("observed"), "reason": d["reason"]}
+    # any previously-DRIFTING locked contract now compliant returns to ACTIVE
+    for e in doc["entities"]:
+        if e["kind"] == "layout_contract" and e.get("status") == "DRIFTING":
+            c = check_layout(doc, observed)
+            comp = e["identities"]["component"]
+            mine = next((x for x in c["checks"] if x["component"] == comp), None)
+            if mine and mine["verdict"] == "PASS":
+                e["status"] = "ACTIVE"
+    return {"gate": result["verdict"], "drifted_locked": staled, "checks": result["checks"]}
+
+
+# keep relationships import used (regions may later carry containment rels)
+_ = relationships
@@ -241,6 +241,13 @@ class TaskStateRepository:
        with self._lock:
            return self._tasks.get(task_id)

+    def list_all(self) -> list[TaskStateRecord]:
+        """Every tracked task record (T17 17F: user-facing counters and
+        listings derive from THIS authority, not the legacy operator
+        cache)."""
+        with self._lock:
+            return list(self._tasks.values())
+
    def list_open(self, channel_id: str) -> list[TaskStateRecord]:
        with self._lock:
            return sorted(self._open_in_channel(channel_id), key=lambda t: t.updated_at)
@@ -1,366 +0,0 @@
-#!/usr/bin/env python3
-"""Build Mode Plugin — State machine orchestrator.
-
-Each phase is an independent module. The state machine coordinates:
-1. Read current state from checkpoint file
-2. Execute the current phase's module
-3. Write output to checkpoint file
-4. Advance to next phase
-5. Repeat
-
-Key insight from Gauntlet: "Checkpoint files on disk. Reports generated
-from checkpoints, not memory." If the conversation dies, the next session
-resumes from the last checkpoint. No drift because state is on disk.
-
-Architecture:
- Each phase is a separate module in phases/
- State machine reads checkpoint.json to know current phase
- Each phase writes its output to checkpoint.json
- Cron tasks fire each phase independently
- Ledger tracks progress for resumability
-"""
-
-import json
-import os
-import sys
-from datetime import datetime, timezone
-from typing import Optional, Dict, Any, List
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# CONFIGURATION
-# ═══════════════════════════════════════════════════════════════════════════════
-
-PLUGIN_DIR = os.path.dirname(os.path.abspath(__file__))
-PHASES_DIR = os.path.join(PLUGIN_DIR, "phases")
-BUILDS_DIR = os.path.expanduser("~/.rune/work/builds")
-WORKTREE_DIR = os.path.expanduser("~/.rune/worktrees")
-REPO_DIR = "/mnt/workdrive/rune"
-
-# Phase execution order
-PHASE_ORDER = [
-    "assess",
-    "research",
-    "feasible",
-    "consult",
-    "consent",
-    "plan",
-    "build",
-    "verify",
-    "correct",
-    "integrate",
-    "record",
-]
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# STATE MACHINE
-# ═══════════════════════════════════════════════════════════════════════════════
-
-
-class BuildStateMachine:
-    """State machine for build mode. Reads/writes checkpoint files."""
-
-    def __init__(self, build_id: str):
-        self.build_id = build_id
-        self.build_dir = os.path.join(BUILDS_DIR, build_id)
-        self.checkpoint_path = os.path.join(self.build_dir, "checkpoint.json")
-        self.ledger_path = os.path.join(self.build_dir, "ledger.json")
-        self.trajectory_path = os.path.join(self.build_dir, "trajectory.json")
-
-        os.makedirs(self.build_dir, exist_ok=True)
-
-    def load_checkpoint(self) -> Dict[str, Any]:
-        """Load current state from checkpoint file."""
-        if os.path.exists(self.checkpoint_path):
-            with open(self.checkpoint_path) as f:
-                return json.load(f)
-        return self._initial_checkpoint()
-
-    def save_checkpoint(self, checkpoint: Dict[str, Any]):
-        """Save state to checkpoint file."""
-        checkpoint["updated_at"] = datetime.now(timezone.utc).isoformat()
-        with open(self.checkpoint_path, "w") as f:
-            json.dump(checkpoint, f, indent=2)
-
-    def _initial_checkpoint(self) -> Dict[str, Any]:
-        """Create initial checkpoint."""
-        return {
-            "build_id": self.build_id,
-            "created_at": datetime.now(timezone.utc).isoformat(),
-            "updated_at": datetime.now(timezone.utc).isoformat(),
-            "current_phase": "assess",
-            "phase_status": "pending",
-            "phases": {},
-            "description": "",
-            "worktree_path": None,
-            "complexity_score": 0,
-            "estimated_steps": 0,
-            "batches": [],
-            "corrections": 0,
-            "max_corrections": 3,
-            "files_changed": [],
-            "contracts": [],
-            "failure_patterns": [],
-        }
-
-    def get_current_phase(self) -> str:
-        """Get current phase from checkpoint."""
-        checkpoint = self.load_checkpoint()
-        return checkpoint.get("current_phase", "assess")
-
-    def get_phase_status(self) -> str:
-        """Get current phase status from checkpoint."""
-        checkpoint = self.load_checkpoint()
-        return checkpoint.get("phase_status", "pending")
-
-    def advance_phase(self):
-        """Advance to next phase."""
-        checkpoint = self.load_checkpoint()
-        current = checkpoint["current_phase"]
-
-        # Mark current phase as complete
-        checkpoint["phases"][current] = "complete"
-
-        # Find next phase
-        try:
-            idx = PHASE_ORDER.index(current)
-            if idx + 1 < len(PHASE_ORDER):
-                next_phase = PHASE_ORDER[idx + 1]
-                checkpoint["current_phase"] = next_phase
-                checkpoint["phase_status"] = "pending"
-            else:
-                checkpoint["phase_status"] = "complete"
-        except ValueError:
-            pass
-
-        self.save_checkpoint(checkpoint)
-
-    def set_phase_status(self, status: str, details: Dict[str, Any] = None):
-        """Set current phase status."""
-        checkpoint = self.load_checkpoint()
-        checkpoint["phase_status"] = status
-        if details:
-            checkpoint.update(details)
-        self.save_checkpoint(checkpoint)
-
-    def log_to_ledger(self, entry: Dict[str, Any]):
-        """Append entry to ledger."""
-        ledger = []
-        if os.path.exists(self.ledger_path):
-            with open(self.ledger_path) as f:
-                ledger = json.load(f)
-
-        entry["timestamp"] = datetime.now(timezone.utc).isoformat()
-        ledger.append(entry)
-
-        with open(self.ledger_path, "w") as f:
-            json.dump(ledger, f, indent=2)
-
-    def log_trajectory(self, action: str, details: Dict[str, Any] = None):
-        """Append to trajectory log (like SWE-agent .traj files)."""
-        trajectory = []
-        if os.path.exists(self.trajectory_path):
-            with open(self.trajectory_path) as f:
-                trajectory = json.load(f)
-
-        entry = {
-            "timestamp": datetime.now(timezone.utc).isoformat(),
-            "action": action,
-            "details": details or {},
-        }
-        trajectory.append(entry)
-
-        with open(self.trajectory_path, "w") as f:
-            json.dump(trajectory, f, indent=2)
-
-    def get_resume_point(self) -> Optional[str]:
-        """Get phase to resume from."""
-        checkpoint = self.load_checkpoint()
-        current = checkpoint.get("current_phase")
-        status = checkpoint.get("phase_status")
-
-        if status == "complete":
-            # All phases complete
-            return None
-
-        return current
-
-    def can_resume(self) -> bool:
-        """Check if build can be resumed."""
-        return self.get_resume_point() is not None
-
-    def get_status(self) -> Dict[str, Any]:
-        """Get build status."""
-        checkpoint = self.load_checkpoint()
-        return {
-            "build_id": self.build_id,
-            "description": checkpoint.get("description", ""),
-            "current_phase": checkpoint.get("current_phase", "assess"),
-            "phase_status": checkpoint.get("phase_status", "pending"),
-            "phases": checkpoint.get("phases", {}),
-            "batches": len(checkpoint.get("batches", [])),
-            "corrections": checkpoint.get("corrections", 0),
-            "worktree": checkpoint.get("worktree_path"),
-            "complexity_score": checkpoint.get("complexity_score", 0),
-            "estimated_steps": checkpoint.get("estimated_steps", 0),
-        }
-
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# PHASE EXECUTION
-# ═══════════════════════════════════════════════════════════════════════════════
-
-
-def execute_phase(build_id: str, phase: str = None) -> Dict[str, Any]:
-    """Execute a phase. If phase is None, execute current phase."""
-    sm = BuildStateMachine(build_id)
-
-    if phase is None:
-        phase = sm.get_current_phase()
-
-    # Check if phase module exists
-    phase_module_path = os.path.join(PHASES_DIR, f"{phase}.py")
-    if not os.path.exists(phase_module_path):
-        return {"success": False, "error": f"Phase module not found: {phase}"}
-
-    # Import and execute phase module
-    try:
-        import importlib.util
-
-        spec = importlib.util.spec_from_file_location(f"phase_{phase}", phase_module_path)
-        module = importlib.util.module_from_spec(spec)
-        spec.loader.exec_module(module)
-
-        # Execute phase
-        result = module.execute(sm)
-
-        # Log to trajectory
-        sm.log_trajectory(f"phase_{phase}", {"result": result})
-
-        # Advance to next phase if successful
-        if result.get("success"):
-            sm.advance_phase()
-
-        return result
-
-    except Exception as e:
-        sm.log_trajectory(f"phase_{phase}_error", {"error": str(e)})
-        return {"success": False, "error": str(e)}
-
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# CLI
-# ═══════════════════════════════════════════════════════════════════════════════
-
-
-def main():
-    if len(sys.argv) < 2:
-        print("Usage: build_mode_plugin.py [start|status|resume|execute|abort]")
-        sys.exit(1)
-
-    cmd = sys.argv[1]
-
-    if cmd == "start":
-        if len(sys.argv) < 3:
-            print("Usage: build_mode_plugin.py start <description>")
-            sys.exit(1)
-        description = " ".join(sys.argv[2:])
-        build_id = datetime.now(timezone.utc).strftime("%Y%m%d-%H%M%S")
-        sm = BuildStateMachine(build_id)
-        checkpoint = sm.load_checkpoint()
-        checkpoint["description"] = description
-        sm.save_checkpoint(checkpoint)
-        sm.log_to_ledger({"action": "build_started", "description": description})
-        print(f"✅ Build {build_id} started")
-        print(f"   Description: {description}")
-        print(f"   Checkpoint: {sm.checkpoint_path}")
-
-    elif cmd == "status":
-        if len(sys.argv) < 3:
-            # List all builds
-            if os.path.exists(BUILDS_DIR):
-                builds = os.listdir(BUILDS_DIR)
-                if builds:
-                    print(f"Builds: {len(builds)}")
-                    for build_id in sorted(builds):
-                        try:
-                            sm = BuildStateMachine(build_id)
-                            status = sm.get_status()
-                            print(
-                                f"  {build_id}: {status['description']} ({status['current_phase']}: {status['phase_status']})"
-                            )
-                        except Exception:
-                            pass
-                else:
-                    print("No builds found.")
-            else:
-                print("No builds found.")
-        else:
-            build_id = sys.argv[2]
-            try:
-                sm = BuildStateMachine(build_id)
-                status = sm.get_status()
-                print(f"Build: {status['build_id']}")
-                print(f"  Description: {status['description']}")
-                print(f"  Phase: {status['current_phase']} ({status['phase_status']})")
-                print(f"  Batches: {status['batches']}")
-                print(f"  Corrections: {status['corrections']}")
-                print(f"  Worktree: {status['worktree']}")
-                print(f"  Complexity: {status['complexity_score']}")
-                print(f"  Estimated steps: {status['estimated_steps']}")
-            except FileNotFoundError:
-                print(f"Build {build_id} not found")
-
-    elif cmd == "resume":
-        if len(sys.argv) < 3:
-            print("Usage: build_mode_plugin.py resume <build_id>")
-            sys.exit(1)
-        build_id = sys.argv[2]
-        try:
-            sm = BuildStateMachine(build_id)
-            if sm.can_resume():
-                resume_point = sm.get_resume_point()
-                print(f"Resuming build {build_id} from phase: {resume_point}")
-                # Execute the current phase
-                result = execute_phase(build_id)
-                print(f"Result: {result}")
-            else:
-                print(f"Build {build_id} cannot be resumed (all phases complete)")
-        except FileNotFoundError:
-            print(f"Build {build_id} not found")
-
-    elif cmd == "execute":
-        if len(sys.argv) < 3:
-            print("Usage: build_mode_plugin.py execute <build_id> [phase]")
-            sys.exit(1)
-        build_id = sys.argv[2]
-        phase = sys.argv[3] if len(sys.argv) > 3 else None
-        result = execute_phase(build_id, phase)
-        print(json.dumps(result, indent=2))
-
-    elif cmd == "abort":
-        if len(sys.argv) < 3:
-            print("Usage: build_mode_plugin.py abort <build_id>")
-            sys.exit(1)
-        build_id = sys.argv[2]
-        try:
-            sm = BuildStateMachine(build_id)
-            checkpoint = sm.load_checkpoint()
-            worktree = checkpoint.get("worktree_path")
-            if worktree:
-                subprocess.run(
-                    ["git", "worktree", "remove", worktree, "--force"],
-                    cwd=REPO_DIR,
-                    capture_output=True,
-                )
-            sm.log_to_ledger({"action": "build_aborted"})
-            print(f"✅ Build {build_id} aborted")
-        except FileNotFoundError:
-            print(f"Build {build_id} not found")
-
-    else:
-        print(f"Unknown command: {cmd}")
-        sys.exit(1)
-
-
-if __name__ == "__main__":
-    main()
@@ -1,115 +0,0 @@
-#!/usr/bin/env python3
-"""Assess phase — Complexity scoring and codebase impact estimation.
-
-This is the first phase. It scores the user's request to determine:
- If build mode is needed (score 8+)
- How many steps are estimated
- What files will be affected
-
-Output: Updates checkpoint with complexity_score, estimated_steps, description.
-"""
-
-import os
-import sys
-from typing import Dict, Any
-
-# Add parent to path for imports
-sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
-sys.path.insert(0, os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", ".."))
-
-
-def execute(sm) -> Dict[str, Any]:
-    """Execute the assess phase."""
-    checkpoint = sm.load_checkpoint()
-    description = checkpoint.get("description", "")
-
-    if not description:
-        return {"success": False, "error": "No description provided"}
-
-    # Import complexity scorer
-    try:
-        from rune.complexity_scorer import score_request
-
-        breakdown = score_request(description)
-    except Exception as e:
-        return {"success": False, "error": f"Complexity scorer failed: {e}"}
-
-    # Update checkpoint
-    checkpoint["complexity_score"] = breakdown.total
-    checkpoint["estimated_steps"] = _estimate_steps(description)
-    checkpoint["assess_signals"] = breakdown.signals
-    checkpoint["assess_mode"] = breakdown.mode
-    checkpoint["assess_confidence"] = breakdown.confidence
-    sm.save_checkpoint(checkpoint)
-
-    # Log to ledger
-    sm.log_to_ledger(
-        {
-            "action": "assess_complete",
-            "complexity_score": breakdown.total,
-            "estimated_steps": checkpoint["estimated_steps"],
-            "mode": breakdown.mode,
-            "confidence": breakdown.confidence,
-            "signals": breakdown.signals,
-        }
-    )
-
-    # Check if build mode is needed
-    if breakdown.total < 8:
-        return {
-            "success": True,
-            "build_mode_needed": False,
-            "complexity_score": breakdown.total,
-            "estimated_steps": checkpoint["estimated_steps"],
-            "mode": breakdown.mode,
-            "reason": f"Score {breakdown.total} < 8, task mode recommended",
-        }
-
-    return {
-        "success": True,
-        "build_mode_needed": True,
-        "complexity_score": breakdown.total,
-        "estimated_steps": checkpoint["estimated_steps"],
-        "mode": breakdown.mode,
-        "confidence": breakdown.confidence,
-        "signals": breakdown.signals,
-    }
-
-
-def _estimate_steps(description: str) -> int:
-    """Estimate number of steps from description."""
-    steps = 0
-    desc_lower = description.lower()
-
-    # Each distinct verb ≈ 1-2 steps
-    verbs = [
-        "add",
-        "create",
-        "build",
-        "implement",
-        "update",
-        "fix",
-        "refactor",
-        "migrate",
-        "test",
-        "deploy",
-        "configure",
-        "integrate",
-        "design",
-    ]
-    verb_count = sum(1 for v in verbs if v in desc_lower)
-    steps += verb_count * 2
-
-    # Each "and" or "then" ≈ 1 step
-    conjunctions = ["and", "then", "also", "plus"]
-    conjunction_count = sum(1 for c in conjunctions if c in desc_lower)
-    steps += conjunction_count
-
-    # Each file reference ≈ 1 step
-    import re
-
-    file_refs = re.findall(r"\b\w+\.\w+\b", description)
-    steps += len(file_refs)
-
-    # Minimum 3 steps for build mode
-    return max(steps, 3)
@@ -1,42 +0,0 @@
-#!/usr/bin/env python3
-"""Build phase — Execute batches in 3 phases.
-
-Three-phase execution:
-1. Foundation: Types, interfaces, shared utilities
-2. Implementation: Feature code
-3. Integration: Tests, wiring, documentation
-
-Output: Updates checkpoint with build results.
-"""
-
-import os
-import sys
-from typing import Dict, Any, List
-
-
-def execute(sm) -> Dict[str, Any]:
-    """Execute the build phase."""
-    checkpoint = sm.load_checkpoint()
-    batches = checkpoint.get("batches", [])
-
-    if not batches:
-        return {"success": False, "error": "No batches to build"}
-
-    sm.log_to_ledger(
-        {
-            "action": "build_started",
-            "batches": len(batches),
-        }
-    )
-
-    # Build is done by the TUI's model
-    # This module just tracks the state
-
-    sm.set_phase_status("in_progress", {"build_requested": True})
-
-    return {
-        "success": True,
-        "build_needed": True,
-        "batches": batches,
-        "instruction": "Execute batches in 3 phases: Foundation → Implementation → Integration",
-    }
@@ -1,39 +0,0 @@
-#!/usr/bin/env python3
-"""Consent phase — User approval.
-
-Presents the plan to the user and waits for approval.
-Output: Updates checkpoint with approved.
-"""
-
-import os
-import sys
-from typing import Dict, Any
-
-
-def execute(sm) -> Dict[str, Any]:
-    """Execute the consent phase."""
-    checkpoint = sm.load_checkpoint()
-    description = checkpoint.get("description", "")
-    complexity_score = checkpoint.get("complexity_score", 0)
-    estimated_steps = checkpoint.get("estimated_steps", 0)
-    patterns = checkpoint.get("consult_patterns", [])
-
-    sm.log_to_ledger(
-        {
-            "action": "consent_started",
-            "description": description,
-        }
-    )
-
-    # Consent is handled by the TUI
-    # This module just prepares the approval request
-
-    return {
-        "success": True,
-        "consent_needed": True,
-        "description": description,
-        "complexity_score": complexity_score,
-        "estimated_steps": estimated_steps,
-        "patterns": patterns,
-        "instruction": "Present plan to user and wait for approval",
-    }
@@ -1,68 +0,0 @@
-#!/usr/bin/env python3
-"""Consult phase — Knowledge base check + failure pattern lookup.
-
-Checks llm-craft and textual-craft knowledge bases for best practices.
-Queries failure pattern database for similar past failures.
-Output: Updates checkpoint with patterns, best_practices.
-"""
-
-import os
-import sys
-from typing import Dict, Any
-
-
-def execute(sm) -> Dict[str, Any]:
-    """Execute the consult phase."""
-    checkpoint = sm.load_checkpoint()
-    description = checkpoint.get("description", "")
-
-    sm.log_to_ledger(
-        {
-            "action": "consult_started",
-            "description": description,
-        }
-    )
-
-    # Query knowledge bases
-    patterns = []
-    best_practices = []
-
-    try:
-        # Check llm-craft knowledge base
-        kb_path = os.path.join(os.path.dirname(__file__), "..", "..", "skills", "llm-craft", "knowledge.md")
-        if os.path.exists(kb_path):
-            with open(kb_path) as f:
-                content = f.read()
-                # Extract relevant patterns
-                if "test" in description.lower():
-                    patterns.append("Use contract testing for interface verification")
-                if "refactor" in description.lower():
-                    patterns.append("Refactor in small steps, verify after each change")
-                if "new feature" in description.lower():
-                    patterns.append("Start with interface definition, then implementation")
-    except Exception:
-        pass
-
-    # Query failure patterns
-    failure_patterns = checkpoint.get("failure_patterns", [])
-    if failure_patterns:
-        best_practices.append(f"Learn from {len(failure_patterns)} past failures")
-
-    # Update checkpoint
-    checkpoint["consult_patterns"] = patterns
-    checkpoint["consult_best_practices"] = best_practices
-    sm.save_checkpoint(checkpoint)
-
-    sm.log_to_ledger(
-        {
-            "action": "consult_complete",
-            "patterns": patterns,
-            "best_practices": best_practices,
-        }
-    )
-
-    return {
-        "success": True,
-        "patterns": patterns,
-        "best_practices": best_practices,
-    }
@@ -1,46 +0,0 @@
-#!/usr/bin/env python3
-"""Correct phase — Fix failed batches.
-
-3 attempts max per batch. After 2 failures, re-plan instead of blindly retrying.
-Output: Updates checkpoint with corrections.
-"""
-
-import os
-import sys
-from typing import Dict, Any
-
-
-def execute(sm) -> Dict[str, Any]:
-    """Execute the correct phase."""
-    checkpoint = sm.load_checkpoint()
-    corrections = checkpoint.get("corrections", 0)
-    max_corrections = checkpoint.get("max_corrections", 3)
-
-    if corrections >= max_corrections:
-        return {
-            "success": False,
-            "error": f"Max corrections ({max_corrections}) reached",
-            "instruction": "Escalate to user with diagnostic summary",
-        }
-
-    sm.log_to_ledger(
-        {
-            "action": "correct_started",
-            "corrections": corrections,
-        }
-    )
-
-    # Correct is done by the TUI
-    # This module just tracks the state
-
-    checkpoint["corrections"] = corrections + 1
-    sm.save_checkpoint(checkpoint)
-
-    sm.set_phase_status("in_progress", {"correct_requested": True})
-
-    return {
-        "success": True,
-        "correct_needed": True,
-        "corrections": corrections + 1,
-        "instruction": "Fix failed batches, re-verify. After 2 failures, re-plan.",
-    }
@@ -1,37 +0,0 @@
-#!/usr/bin/env python3
-"""Feasible phase — Feasibility gate + complexity report.
-
-Checks if the task is feasible based on research findings.
-Output: Updates checkpoint with feasible, risk, complexity_report.
-"""
-
-import os
-import sys
-from typing import Dict, Any
-
-
-def execute(sm) -> Dict[str, Any]:
-    """Execute the feasible phase."""
-    checkpoint = sm.load_checkpoint()
-    description = checkpoint.get("description", "")
-    research = checkpoint.get("research_findings", {})
-
-    # Feasibility is determined by the TUI's model
-    # This module just tracks the state
-
-    sm.log_to_ledger(
-        {
-            "action": "feasible_started",
-            "description": description,
-        }
-    )
-
-    sm.set_phase_status("in_progress", {"feasibility_requested": True})
-
-    return {
-        "success": True,
-        "feasibility_needed": True,
-        "description": description,
-        "research": research,
-        "instruction": "Assess feasibility based on research findings",
-    }
@@ -1,32 +0,0 @@
-#!/usr/bin/env python3
-"""Integrate phase — Scope guardian (full test suite).
-
-Runs full verification after all batches pass.
-Output: Updates checkpoint with integration results.
-"""
-
-import os
-import sys
-from typing import Dict, Any
-
-
-def execute(sm) -> Dict[str, Any]:
-    """Execute the integrate phase."""
-    checkpoint = sm.load_checkpoint()
-
-    sm.log_to_ledger(
-        {
-            "action": "integrate_started",
-        }
-    )
-
-    # Integrate is done by the TUI
-    # This module just tracks the state
-
-    sm.set_phase_status("in_progress", {"integrate_requested": True})
-
-    return {
-        "success": True,
-        "integrate_needed": True,
-        "instruction": "Run full test suite: canary + ruff + pyright + contracts + TUI validator",
-    }
@@ -1,41 +0,0 @@
-#!/usr/bin/env python3
-"""Plan phase — Work manager creates batches with dependency DAG.
-
-Creates batches of 1-3 files each, routes to local or cloud models.
-Output: Updates checkpoint with batches.
-"""
-
-import os
-import sys
-from typing import Dict, Any, List
-
-
-def execute(sm) -> Dict[str, Any]:
-    """Execute the plan phase."""
-    checkpoint = sm.load_checkpoint()
-    description = checkpoint.get("description", "")
-    complexity_score = checkpoint.get("complexity_score", 0)
-    estimated_steps = checkpoint.get("estimated_steps", 0)
-
-    sm.log_to_ledger(
-        {
-            "action": "plan_started",
-            "description": description,
-            "complexity_score": complexity_score,
-            "estimated_steps": estimated_steps,
-        }
-    )
-
-    # Plan is done by the TUI's model
-    # This module just tracks the state
-
-    sm.set_phase_status("in_progress", {"plan_requested": True})
-
-    return {
-        "success": True,
-        "plan_needed": True,
-        "description": description,
-        "complexity_score": complexity_score,
-        "estimated_steps": estimated_steps,
-        "instruction": "Create batches with dependency DAG, route to models",
-    }
@@ -1,38 +0,0 @@
-#!/usr/bin/env python3
-"""Record phase — Update PRD, GitHub, knowledge base, trajectory log.
-
-Final phase. Updates all documentation and records what was learned.
-Output: Updates checkpoint with record results.
-"""
-
-import os
-import sys
-from typing import Dict, Any
-
-
-def execute(sm) -> Dict[str, Any]:
-    """Execute the record phase."""
-    checkpoint = sm.load_checkpoint()
-    build_id = sm.build_id
-    description = checkpoint.get("description", "")
-
-    sm.log_to_ledger(
-        {
-            "action": "record_started",
-            "build_id": build_id,
-            "description": description,
-        }
-    )
-
-    # Record is done by the TUI
-    # This module just tracks the state
-
-    sm.set_phase_status("in_progress", {"record_requested": True})
-
-    return {
-        "success": True,
-        "record_needed": True,
-        "build_id": build_id,
-        "description": description,
-        "instruction": "Update PRD, commit to GitHub, update knowledge base with learnings",
-    }
@@ -1,41 +0,0 @@
-#!/usr/bin/env python3
-"""Research phase — Deep research on the task.
-
-Always runs for build mode. Uses deep-research skill or parallel sub-agents.
-Output: Updates checkpoint with research_findings.
-"""
-
-import os
-import sys
-from typing import Dict, Any
-
-
-def execute(sm) -> Dict[str, Any]:
-    """Execute the research phase."""
-    checkpoint = sm.load_checkpoint()
-    description = checkpoint.get("description", "")
-
-    if not description:
-        return {"success": False, "error": "No description provided"}
-
-    # Log to ledger
-    sm.log_to_ledger(
-        {
-            "action": "research_started",
-            "description": description,
-        }
-    )
-
-    # Research is done by the TUI's model (not this module)
-    # This module just tracks the state
-    # The actual research happens in the TUI's _send_to_model flow
-
-    # For now, mark as needing external research
-    sm.set_phase_status("in_progress", {"research_requested": True})
-
-    return {
-        "success": True,
-        "research_needed": True,
-        "description": description,
-        "instruction": "Use deep-research skill or parallel sub-agents to research this task",
-    }
@@ -1,40 +0,0 @@
-#!/usr/bin/env python3
-"""Verify phase — Tiered verification.
-
-Tiered verification strategy:
-1. Syntactic: Per-edit lint, type-check, format (<1s)
-2. Semantic: Per-batch unit tests, import validation (<30s)
-3. Integration: Per-wave contract tests, build/compile (<5min)
-4. System: Full test suite, property-based (<15min)
-
-Output: Updates checkpoint with verification results.
-"""
-
-import os
-import sys
-from typing import Dict, Any
-
-
-def execute(sm) -> Dict[str, Any]:
-    """Execute the verify phase."""
-    checkpoint = sm.load_checkpoint()
-    batches = checkpoint.get("batches", [])
-
-    sm.log_to_ledger(
-        {
-            "action": "verify_started",
-            "batches": len(batches),
-        }
-    )
-
-    # Verify is done by the TUI
-    # This module just tracks the state
-
-    sm.set_phase_status("in_progress", {"verify_requested": True})
-
-    return {
-        "success": True,
-        "verify_needed": True,
-        "batches": batches,
-        "instruction": "Run tiered verification: syntactic → semantic → integration → system",
-    }
@@ -0,0 +1,349 @@
+"""programme_continuation.py — T15 §15M: the PROGRAMME CONTINUATION
+CANARY.
+
+The user observation this answers: subtask completion repeatedly ended
+whole programmes. "task done" is not "programme done".
+
+After ANY stage/subtask reports complete, the runner asks one bounded
+question — is the overall construction/programme objective satisfied?
+If not, and eligible authorized stages remain, the NEXT stage starts
+immediately (no user re-prompt, no voluntary stop). Blocked branches
+fall back to pre-authorized independent work.
+
+Boundary (tranche brief): this is a bounded CONSTRUCTION-programme rule
+over Scaffolding stage contracts — NOT the future general-purpose
+Substrate scheduler. It knows nothing outside one programme's stage
+list.
+"""
+
+from __future__ import annotations
+
+import json
+import time
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, cast
+
+
+# stage states
+PENDING = "PENDING"
+ELIGIBLE = "ELIGIBLE"
+WAITING_EVENT = "WAITING_EVENT"
+COMPLETE = "COMPLETE"
+BLOCKED = "BLOCKED"
+APPROVAL_REQUIRED = "APPROVAL_REQUIRED"
+FAILED = "FAILED"
+
+TERMINAL = {COMPLETE, BLOCKED, APPROVAL_REQUIRED, FAILED}
+
+
+@dataclass
+class ProgrammeStage:
+    stage_id: str
+    objective: str
+    state: str = PENDING
+    depends_on: list = field(default_factory=list)   # stage_ids
+    priority: int = 0                                # lower = sooner
+    result: dict = field(default_factory=dict)
+
+    def to_dict(self) -> dict:
+        return {"stage_id": self.stage_id, "objective": self.objective,
+                "state": self.state, "depends_on": list(self.depends_on),
+                "priority": self.priority}
+
+
+@dataclass
+class Programme:
+    """One bounded construction programme: its objective, its authorized
+    stages, and the continuation rule. Persisted to the contract store's
+    project area (durable, reloadable, no transcript dependency)."""
+    programme_id: str
+    objective: str
+    stages: list = field(default_factory=list)  # [ProgrammeStage]
+    fallbacks: list = field(default_factory=list)  # pre-authorized ids
+
+    # ── the continuation canary (15M) ────────────────────────────────
+    def continuation_canary(self) -> dict:
+        """'task done' ≠ 'programme done'. Returns the verdict AND the
+        next action — the runner never has to guess whether to stop."""
+        complete = [s for s in self.stages if s.state == COMPLETE]
+        if len(complete) == len(self.stages):
+            return {"verdict": "PROGRAMME_COMPLETE",
+                    "next_action": None,
+                    "reason": f"all {len(self.stages)} stages complete"}
+        eligible = self.eligible_stages()
+        if eligible:
+            nxt = eligible[0]
+            return {"verdict": "PROGRAMME_OPEN",
+                    "next_action": ("start", nxt.stage_id),
+                    "reason": (f"{len(complete)}/{len(self.stages)} "
+                               f"complete; {len(eligible)} eligible — "
+                               f"continue immediately")}
+        waiting = [s for s in self.stages
+                   if s.state == WAITING_EVENT]
+        if waiting:
+            return {"verdict": "PROGRAMME_OPEN_WAITING",
+                    "next_action": ("await_event", waiting[0].stage_id),
+                    "reason": "event-driven wait active — no polling"}
+        blocked = [s for s in self.stages if s.state == BLOCKED]
+        if blocked and self.fallbacks:
+            return {"verdict": "PROGRAMME_OPEN",
+                    "next_action": ("fallback", self.fallbacks[0]),
+                    "reason": (f"{len(blocked)} blocked; pre-authorized "
+                               f"fallback available")}
+        if blocked:
+            return {"verdict": "ALL_REMAINING_BLOCKED",
+                    "next_action": None,
+                    "reason": f"{len(blocked)} stage(s) blocked, no fallback"}
+        approval = [s for s in self.stages
+                    if s.state == APPROVAL_REQUIRED]
+        if approval:
+            return {"verdict": "APPROVAL_REQUIRED",
+                    "next_action": None,
+                    "reason": (f"{len(approval)} stage(s) need user "
+                               f"authorization")}
+        # T17 §17E: all stages terminal but some FAILED — the programme
+        # may still close (programme-level acceptance decides), never
+        # hang in UNKNOWN_INTERRUPTION.
+        if all(s.state in TERMINAL for s in self.stages):
+            failed = [s.stage_id for s in self.stages if s.state == FAILED]
+            return {"verdict": "PROGRAMME_COMPLETE_WITH_FAILURES",
+                    "next_action": None,
+                    "reason": (f"all stages terminal; failed: {failed} — "
+                               f"programme-level acceptance decides")}
+        return {"verdict": "UNKNOWN_INTERRUPTION", "next_action": None,
+                "reason": "no stage runnable and none terminal"}
+
+    def eligible_stages(self) -> list:
+        """PENDING stages whose dependencies are COMPLETE, priority-first.
+        ELIGIBLE counts as runnable too — an eligible stage nobody started
+        is exactly the case the continuation canary exists to catch."""
+        done = {s.stage_id for s in self.stages if s.state == COMPLETE}
+        out = [s for s in self.stages
+               if s.state in (PENDING, ELIGIBLE)
+               and all(d in done for d in s.depends_on)]
+        out.sort(key=lambda s: (s.priority, s.stage_id))
+        return out
+
+    # ── durable persistence (canonical store, not transcript) ────────
+    def save(self, base: Path) -> Path:
+        base.mkdir(parents=True, exist_ok=True)
+        p = base / f"{self.programme_id}.json"
+        p.write_text(json.dumps({
+            "programme_id": self.programme_id,
+            "objective": self.objective,
+            "stages": [s.to_dict() for s in self.stages],
+            "fallbacks": list(self.fallbacks),
+            "saved_at": time.strftime("%Y-%m-%dT%H:%M:%SZ",
+                                      time.gmtime()),
+        }, indent=1))
+        return p
+
+    @classmethod
+    def load(cls, base: Path, programme_id: str) -> "Programme | None":
+        p = base / f"{programme_id}.json"
+        if not p.exists():
+            return None
+        d = json.loads(p.read_text())
+        prog = cls(programme_id=d["programme_id"],
+                   objective=d["objective"], fallbacks=d.get("fallbacks", []))
+        for s in d["stages"]:
+            prog.stages.append(ProgrammeStage(**s))
+        return prog
+
+
+def run_programme(prog: Programme, stage_runner, *,
+                  max_transitions: int = 100) -> list[dict]:
+    """Drive one programme to a legitimate stop. `stage_runner(stage)` —
+    provided by the caller — must return a state for the stage. Between
+    EVERY stage: persist → continuation canary → next. Returns the
+    transition log (the 15M proof artifact)."""
+    log: list[dict] = []
+    for _ in range(max_transitions):
+        prog_canary = prog.continuation_canary()
+        log.append({"canary": prog_canary})
+        verdict, action = prog_canary["verdict"], prog_canary["next_action"]
+        if action is None:
+            break  # legitimate stop — reason recorded in the log
+        kind, target = action
+        if kind == "fallback":
+            log.append({"fallback_started": target})
+            break
+        stage = next(s for s in prog.stages if s.stage_id == target)
+        stage.state = stage_runner(stage)   # runner decides the new state
+        log.append({"stage": target, "new_state": stage.state})
+    return log
+
+
+# ── T17 §17D: production wiring — Scaffolding-integrated driver ─────────
+#
+# The 15M canary proved the RULE; this adapter makes it CANONICAL. One
+# programme = a list of stages; each stage maps 1:1 to a Scaffolding
+# contract (stage_id == contract_id). Stages run through the Construction
+# Interface lifecycle (approved→building→verifying→closed); asynchronous
+# stage work runs under the event bridge (WAITING_EVENT is a state, not
+# a stop). Persisted to the scaffold store's programme area — reloadable
+# after restart with no transcript.
+
+DEFAULT_PROGRAMME_AREA = "programmes"
+
+
+def _programme_base(scaffold_store: Path) -> Path:
+    return Path(scaffold_store) / DEFAULT_PROGRAMME_AREA
+
+
+def scaffolding_stage_runner(interface, *, scaffold_store: Path,
+                             daemon_url: str = "",
+                             watchdog_secs: int = 900):
+    """Return a stage_runner for run_programme bound to the Construction
+    Interface. The stage's work is read from stage.result["work"] — a
+    callable (interface, stage) → dict with {"ok": bool, "evidence": str,
+    "async_cmd": [argv] optional}. When async_cmd is present the work
+    runs under the event bridge and the driver resumes on the EVENT;
+    canonical verification still decides completion truth. Stages without
+    work complete as governance-only transitions."""
+    from rune import event_wait as ew  # local import: optional dependency
+
+    def _phase(contract_id: str) -> str:
+        # ALWAYS re-read: transition() saves through its own doc copy, so
+        # a locally-held contract dict goes stale after any transition
+        # (T17 17E live finding — the draft→approved→building chain
+        # silently stopped after the first step on a stale reference).
+        doc = interface.scaffold.load()
+        c = interface.scaffold.contract(doc, contract_id)
+        return (c or {}).get("phase", "")
+
+    def run(stage: ProgrammeStage) -> str:
+        contract_id = stage.stage_id
+        phase = _phase(contract_id)
+        if phase == "":
+            stage.result["error"] = f"no contract {contract_id}"
+            return FAILED
+        if phase == "closed":
+            return COMPLETE  # idempotent re-entry (duplicate signal case)
+        if phase == "draft":
+            interface.transition(contract_id, "approved",
+                                 evidence_ref="programme-stage-start")
+        if _phase(contract_id) == "approved":
+            interface.transition(contract_id, "building",
+                                 evidence_ref="programme-stage-start")
+        work = cast("Any", stage.result.get("work"))
+        if work is None:
+            interface.transition(contract_id, "verifying",
+                                 evidence_ref="governance-only-stage")
+            interface.close(
+                contract_id,
+                observed_evidence=[{"source": "programme-driver",
+                                    "detail": "governance-only stage: "
+                                              "no work, closure by "
+                                              "programme acceptance"}],
+                scope_files=[])
+            return COMPLETE
+        # work is either a callable (interface, stage) → outcome dict, or
+        # a literal outcome spec dict ({"ok", "evidence", "async_cmd"?})
+        outcome: dict = (dict(cast("dict", work(interface, stage)))
+                         if callable(work) else dict(cast("dict", work)))
+        async_cmd = outcome.get("async_cmd")
+        if async_cmd:
+            op = f"stage-{contract_id}-{int(time.time())}"
+            ew.register_process_wait(
+                op, contract_id=contract_id,
+                project=getattr(interface.scaffold, "project_id", ""),
+                objective=stage.objective,
+                watched_for="stage work completion event",
+                watchdog_secs=watchdog_secs)
+            notifier = ew.run_builder_under_events(
+                async_cmd, op, daemon_url=daemon_url or ew.DEFAULT_DAEMON_URL)
+            stage.result["operation_id"] = op
+            stage.result["notifier_pgid"] = notifier.pid
+            result = ew.await_resume_event(op, watchdog_secs=watchdog_secs,
+                                           pgid=notifier.pid)
+            if result["kind"] != "event":
+                stage.result["error"] = f"stage event wait: {result['kind']}"
+                interface.transition(contract_id, "failed",
+                                     evidence_ref="stage-event-lost")
+                return FAILED
+            rec: dict = result.get("wait") or {}
+            outcome["evidence"] = (f"event resume; wait {rec.get('wait_id')} "
+                                   f"verdict={rec.get('verdict')}")
+        if not outcome.get("ok"):
+            stage.result["error"] = outcome.get("evidence", "work failed")
+            # truthful failure: record the evidence ON the contract and
+            # drive it to the failed phase via the legal transition map
+            # (building→failed is legal; no exception swallowing)
+            doc = interface.scaffold.load()
+            c = interface.scaffold.contract(doc, contract_id)
+            if c is not None:
+                c.setdefault("checkpoints", []).append({
+                    "at": time.strftime("%Y-%m-%dT%H:%M:%SZ",
+                                        time.gmtime()),
+                    "phase": c.get("phase", ""),
+                    "evidence": "programme-stage-failed",
+                    "note": (outcome.get("evidence") or "work failed")[:200],
+                })
+                interface.scaffold.save(doc)
+                phase = c.get("phase", "")
+                if phase == "building":
+                    interface.transition(contract_id, "failed",
+                                         evidence_ref="programme-stage-failed")
+            return FAILED
+        # completion truth: work evidence recorded ON the contract, then
+        # closed (evidence lives on the durable object, not the log)
+        if _phase(contract_id) == "building":
+            interface.transition(contract_id, "verifying",
+                                 evidence_ref="programme-stage-verify")
+        doc = interface.scaffold.load()
+        c = interface.scaffold.contract(doc, contract_id)
+        if c is not None:
+            c.setdefault("checkpoints", []).append({
+                "at": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
+                "phase": "verifying",
+                "evidence": "programme-stage-work",
+                "note": (outcome.get("evidence") or "work completed")[:200],
+            })
+            interface.scaffold.save(doc)
+        interface.close(
+            contract_id,
+            observed_evidence=[{"source": "programme-driver",
+                                "detail": outcome.get("evidence", "ok")}],
+            scope_files=[])
+        return COMPLETE
+
+    return run
+
+
+def run_construction_programme(prog: Programme, interface, *,
+                               scaffold_store: Path,
+                               daemon_url: str = "",
+                               watchdog_secs: int = 900,
+                               max_transitions: int = 100) -> list[dict]:
+    """Canonical multi-stage construction continuation (T17 §17D):
+    run_programme with the Scaffolding-bound runner; persists the
+    programme around every transition so a restart resumes from the
+    durable record, never a transcript. Returns the transition log."""
+    runner = scaffolding_stage_runner(
+        interface, scaffold_store=scaffold_store,
+        daemon_url=daemon_url, watchdog_secs=watchdog_secs)
+    log: list[dict] = []
+    base = _programme_base(scaffold_store)
+    for _ in range(max_transitions):
+        prog.save(base)
+        canary = prog.continuation_canary()
+        log.append({"canary": canary})
+        verdict, action = canary["verdict"], canary["next_action"]
+        if action is None:
+            break
+        kind, target = action
+        if kind == "fallback":
+            log.append({"fallback_started": target})
+            break
+        stage = next(s for s in prog.stages if s.stage_id == target)
+        if stage.state == COMPLETE:
+            continue  # duplicate continuation signal: idempotent
+        stage.state = ELIGIBLE
+        new_state = runner(stage)
+        stage.state = new_state
+        log.append({"stage": target, "new_state": new_state})
+        prog.save(base)
+    prog.save(base)
+    return log
@@ -0,0 +1,85 @@
+"""programme_gates.py — T21.2 §21.2G: mandatory acceptance-gate accounting.
+
+The executing model must NOT be able to:
+  - silently remove mandatory gates;
+  - demote MANDATORY → OPTIONAL;
+  - restore acceptance with unrun mandatory gates.
+
+Every mandatory gate ends in exactly one of: PASS | BLOCKED | WAIVED_BY_USER.
+Optional gates may be OPTIONAL_SKIPPED or OPTIONAL_PASS.
+
+This is a governance correction, not new architecture.
+"""
+from __future__ import annotations
+
+import time
+
+MANDATORY_STATUSES = ("PASS", "BLOCKED", "WAIVED_BY_USER")
+OPTIONAL_STATUSES = ("OPTIONAL_SKIPPED", "OPTIONAL_PASS")
+ALL_STATUSES = MANDATORY_STATUSES + OPTIONAL_STATUSES
+
+
+def _now() -> str:
+    return time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+
+
+class GateDemotionError(ValueError):
+    """The model tried to demote a mandatory gate to optional."""
+
+
+class GateAccounting:
+    def __init__(self):
+        self.gates: dict[str, dict] = {}
+
+    def declare(self, gate_id: str, *, mandatory: bool,
+                requirement: str) -> None:
+        self.gates[gate_id] = {
+            "gate_id": gate_id, "mandatory": mandatory,
+            "requirement": requirement[:200],
+            "status": "UNRUN", "evidence": "", "blocker": "",
+            "waiver_reason": "", "waived_by": "",
+            "declared_at": _now(), "resolved_at": "",
+        }
+
+    def record(self, gate_id: str, status: str, *,
+               evidence: str = "", blocker: str = "",
+               waiver_reason: str = "", waived_by: str = "") -> dict:
+        g = self.gates.get(gate_id)
+        if g is None:
+            raise ValueError(f"unknown gate {gate_id!r}")
+        if g["mandatory"] and status in OPTIONAL_STATUSES:
+            raise GateDemotionError(
+                f"gate {gate_id!r} is MANDATORY — cannot be marked "
+                f"{status!r} without explicit user waiver")
+        if status == "WAIVED_BY_USER":
+            if not g["mandatory"]:
+                raise ValueError(
+                    f"WAIVED_BY_USER only applies to mandatory gates; "
+                    f"{gate_id!r} is optional")
+            if not waiver_reason or not waived_by:
+                raise ValueError(
+                    f"waiver for {gate_id!r} requires reason + waived_by")
+        if status not in ALL_STATUSES:
+            raise ValueError(f"unknown gate status {status!r}")
+        g["status"] = status
+        g["evidence"] = evidence[:200]
+        g["blocker"] = blocker[:200]
+        g["waiver_reason"] = waiver_reason[:200]
+        g["waived_by"] = waived_by
+        g["resolved_at"] = _now()
+        return g
+
+    def can_accept(self) -> tuple[bool, list[str]]:
+        """Acceptance eligible only when every mandatory gate is
+        PASS or WAIVED_BY_USER (with persisted waiver)."""
+        blockers = []
+        for gid, g in sorted(self.gates.items()):
+            if not g["mandatory"]:
+                continue
+            if g["status"] not in ("PASS", "WAIVED_BY_USER"):
+                blockers.append(
+                    f"{gid}: {g['status']} — {g.get('blocker', '')}")
+        return (not blockers, blockers)
+
+    def accounting(self) -> list[dict]:
+        return sorted(self.gates.values(), key=lambda g: g["gate_id"])
@@ -0,0 +1,139 @@
+"""project_registry.py — T19 §19C: canonical registry of scaffolded
+projects.
+
+BOUNDARIES (rule 19C, proven by tests):
+  - Registry = DISCOVERY/INDEX (what projects exist, where their truth
+    lives, health/migration state).
+  - Project Scaffold/structural model = project structural truth
+    (never duplicated here).
+  - Spine = Rune's structural self-model. An external project's Scaffold
+    does NOT become Spine structure because Rune knows it exists.
+
+Registry state: one JSON document (versioned, provenance-aware per rule
+G) at <store_dir>/project_registry.json. Operational uses: /testing
+across projects (19L), drift/migration sweeps, re-entry discovery.
+"""
+from __future__ import annotations
+
+import time
+from pathlib import Path
+
+REGISTRY_SCHEMA_VERSION = 1
+
+# project health states (operational: drives /testing and migration
+# consent sweeps)
+HEALTHY = "HEALTHY"
+DRIFTED = "DRIFTED"
+MIGRATION_REQUIRED = "MIGRATION_REQUIRED"
+UNKNOWN = "UNKNOWN"
+
+
+def _now() -> str:
+    return time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime())
+
+
+class ProjectRegistry:
+    def __init__(self, store_dir: Path | str):
+        self.path = Path(store_dir) / "project_registry.json"
+        self.doc = self._load()
+
+    def _load(self) -> dict:
+        if self.path.is_file():
+            import json
+            return json.loads(self.path.read_text())
+        return {"schema_version": REGISTRY_SCHEMA_VERSION,
+                "projects": {}, "updated_at": _now()}
+
+    def save(self) -> None:
+        import json
+        self.doc["updated_at"] = _now()
+        self.path.parent.mkdir(parents=True, exist_ok=True)
+        self.path.write_text(json.dumps(self.doc, indent=1, sort_keys=True))
+
+    def register(self, *, project_id: str, scaffold_location: str,
+                 repo_identity: str = "", forge_repo: str = "",
+                 default_branch: str = "master",
+                 structural_schema_version: int = 1,
+                 runtime_generation: str = "",
+                 last_known_good: str = "") -> dict:
+        """Register/index a scaffolded project. References only — the
+        project's structural truth stays in ITS store (boundary rule)."""
+        entry = {
+            "project_id": project_id,
+            "scaffold_location": scaffold_location,
+            "repo_identity": repo_identity,
+            "forge_repo": forge_repo,
+            "default_branch": default_branch,
+            "structural_schema_version": structural_schema_version,
+            "scaffold_state": {"health": UNKNOWN,
+                               "last_material_update": _now()},
+            "runtime_generation": runtime_generation,
+            "last_known_good": last_known_good,
+            "outstanding_validations": 0,
+            "registered_at": _now(),
+        }
+        self.doc["projects"][project_id] = entry
+        return entry
+
+    def get(self, project_id: str) -> dict | None:
+        return self.doc["projects"].get(project_id)
+
+    def all_projects(self) -> list[dict]:
+        return sorted(self.doc["projects"].values(),
+                      key=lambda p: p["project_id"])
+
+    def set_health(self, project_id: str, health: str, *,
+                   note: str = "") -> None:
+        if health not in (HEALTHY, DRIFTED, MIGRATION_REQUIRED, UNKNOWN):
+            raise ValueError(f"unknown health {health!r}")
+        p = self.doc["projects"][project_id]
+        p["scaffold_state"]["health"] = health
+        p["scaffold_state"]["last_material_update"] = _now()
+        if note:
+            p["scaffold_state"].setdefault("notes", []).append(
+                {"at": _now(), "note": note[:160]})
+
+    def set_outstanding_validations(self, project_id: str,
+                                    count: int) -> None:
+        self.doc["projects"][project_id]["outstanding_validations"] = count
+
+    def mark_generation(self, project_id: str, *, current: str,
+                        last_known_good: str = "") -> None:
+        p = self.doc["projects"][project_id]
+        p["runtime_generation"] = current
+        if last_known_good:
+            p["last_known_good"] = last_known_good
+
+    def resolve_cwd(self, project_id: str) -> str:
+        """T21.2 §21.2B: canonical REPOSITORY/worktree execution cwd for
+        a project — NOT scaffold_location. Precedence (21.2B-3):
+          explicit contract/worktree cwd > programme cwd > registry
+          repo cwd > unresolved/refuse.
+        Registry stores repo_identity as "git:<path>" or a bare path.
+        Scaffold storage is NEVER used as a repo-cwd proxy. Raises for
+        unknown projects (refuse mutation rather than guessing)."""
+        entry = self.get(project_id)
+        if entry is None:
+            raise ValueError(
+                f"unknown project {project_id!r} — resolve identity "
+                f"before mutation; never execute from HOME")
+        repo = entry.get("repo_identity", "")
+        if repo.startswith("git:"):
+            repo = repo[len("git:"):]
+        repo = repo.strip()
+        if not repo:
+            raise ValueError(
+                f"project {project_id!r} has no repo_identity — "
+                f"no local execution checkout available")
+        # an active worktree overrides the repo root (21.2B-3 precedence)
+        wt = entry.get("active_worktree", "")
+        if wt:
+            return wt
+        return repo
+
+    def needs_migration(self, proposed_schema_version: int) -> list[str]:
+        """Projects whose structural schema is older than a proposed
+        version (feeds the 18I consent sweep — never silent upgrades)."""
+        return sorted(pid for pid, p in self.doc["projects"].items()
+                      if p.get("structural_schema_version", 1)
+                      < proposed_schema_version)