60 Commits

Include renames

Author	SHA1	Message	Date
Nick	eadb65a6dc	chore: census milestone sync	2026-10-10 22:01:46 +01:00
Nick	b7501b9804	feat(B3-v3): clean immutable baseline — #165 SimpleMem/A-MEM/Letta through full fixed-depth pipeline ... ADAPT-1 CLOSED. B3-v3 clean baseline on frozen SHA 5a1ad11/v0.64.0. Zero pipeline modifications during run. Zero manual artifact edits. Chain: ISSUE(0.16s) → INWARD(0.34s) → OUTWARD(1106s, 12 sources/8 read) → APPLICABILITY(970s, 10 candidates, INNOVATION_WARRANTED) → INNOVATION(386s, BUILD, 5 candidates) → ADVERSARIAL(809s after TIMEOUT retry, 21 attacks, 5 survivors, SURVIVES_WITH_MITIGATIONS) → SYNTHESIS(444s) → QUALIFY(REFINE_SYNTHESIS ×2 → PROCEED 13/13). Total wall ~4897s (~82 min) across 12 stage executions. First-attempt pass: ISSUE, INWARD, APPLICABILITY, INNOVATION, SYNTHESIS v1. Native retries: OUTWARD (schema-variant normalization), ADVERSARIAL (stuck process → retry succeeded), SYNTHESIS (2 REFINE_SYNTHESIS loops). Also fixes OUTWARD artifact resolution: the file-channel path now uses _resolve_worker_artifact (shared helper) instead of a bespoke check, so prose-stdout + valid file works correctly (the #165 GATE_RETRY pattern). 1179+ tests pass (baseline suite unchanged).	2026-09-10 06:36:07 +01:00
conductor	11df1983f4	research-165-r1: QualifyArtifact — PROCEED (13/13 checks)	2026-09-10 06:33:57 +01:00
conductor	ade5d862eb	research-165-r1: SynthesisArtifact (10 steps, 18 invariants, 24 tests)	2026-09-10 06:33:53 +01:00
conductor	f98858b5fe	research-165-r1: AdversarialArtifact (21 attacks, 5 survivors, SURVIVES_WITH_MITIGATIONS)	2026-09-10 06:33:48 +01:00
conductor	849a75f079	research-165-r1: InnovationArtifact (BUILD, 5 candidates)	2026-09-10 06:33:44 +01:00
conductor	1d4662a1c7	research-165-r1: ApplicabilityArtifact (10 candidates, INNOVATION_WARRANTED)	2026-09-10 06:33:40 +01:00
conductor	fe25aa3ccc	research-165-r1: OutwardArtifact (12 sources, 8 read)	2026-09-10 06:33:36 +01:00
conductor	2b323e7a44	research-165-r1: QUALIFY PROCEED — pipeline COMPLETE (clean B3-v3 baseline, frozen SHA 5a1ad11/v0.64.0)	2026-09-10 06:33:33 +01:00
Nick	deba21fe54	feat(TASK-I): PL0 passive Experience Ledger — 10 seed Process Experiences from real runs ... scripts/process_experience.py: ExperienceLedger with 12 root-cause layers, 3 result types, confidence tracking, evidence references, correction tracking, recurrence classification, cost measurement, reusable flag. 10 seed Process Experiences from real runs (#161/#163/#164/#310/#311): - 5 failures (schema drift, field naming, missing disposition, validator mismatch, duplicate contract authority) - 3 successes (gate blocked missing M1, INNOVATION correctly skipped, real external evidence produced decision-changing results) - 2 inefficiencies (routing waste, detection gap) All 10 marked reusable_lesson=True. No autonomous self-editing — observation and evidence capture only.	2026-09-10 02:22:49 +01:00
Nick	203e6ed41d	fix(RESEARCH): schema-variant normalization + disposition prompt + multi-dimension matching + URL check narrowing ... Findings from B3-v2 fresh Route-C run on #164 (real research task): 1. Schema-variant normalization: the worker emitted search_receipts/ read_receipts/sources/evidence_claims at top level instead of discovery/reading subsections. _normalize_outward_artifact now maps documented variants to the canonical structure (a documented mapping, not a rescue — the worker DID produce the data). Also maps worker 'urls' (list) → 'url' and 'classification' → 'source_class'. 2. ADVERSARIAL prompt now demands the disposition field explicitly (the template predated the disposition contract — the worker omitted it). 3. Adversarial gate: multi-dimension matching — an attack with 'duplicate_authority / Rune constraint violations' addresses BOTH dimensions. Removed the break-after-first-match. 4. OUTWARD URL check narrowed: sources with no URL are legitimately discovered-but-not-fetched (the receipt check handles READ status). 5. SYNTHESIS gate caught a real design gap on #164: mitigation M1 (decide the integration surface FIRST) was not represented — the gate refusing is the system working. B3-v2 paused for synthesis v2. Real acceptance on #164: OUTWARD 13 sources/6 read (415s), APPLICABILITY 7 candidates/STANDARD_SOLUTION_SUFFICIENT (369s), ADVERSARIAL 10 attacks/SURVIVES_WITH_MITIGATIONS (reconciled), SYNTHESIS gate correctly blocked. All artifacts on Gitea master. 1168 tests pass; ruff clean.	2026-09-10 01:20:57 +01:00
Nick	a997a0ae4c	feat(D4+D5): executor-owned artifact targets + cognitive-stage mutation boundary (closes #298 scope) ... D4 — ARTIFACT TARGET CONTRACT IS EXECUTOR-OWNED: - _declared_artifact_path(stage): absolute path computed from artifact_dir + STAGE_ARTIFACT_FILES[stage]; injected verbatim into every worker prompt (APPLICABILITY/ADVERSARIAL/SYNTHESIS/INNOVATION/OUTWARD/QUALIFY) - Worker never derives its own path — producer and consumer share one authoritative output location (the B3-v2 root cause: worker's prompt said research/issue-162/ but executor's artifact_dir was research/issue-162-b3v2/) - Acceptance: nonstandard artifact_dir (issue-162-b3v2) now gets the correct path injected D5 — COGNITIVE_STAGE_MUTATION_BOUNDARY: - mutation_boundary.py: capture_worktree_state (status porcelain + HEAD) before cognitive stages; verify_no_source_mutation after; new changes outside declared artifact paths → UNAUTHORIZED_STAGE_MUTATION - HEAD movement detected (cognitive stages must never commit) - allowed paths: declared artifact + telemetry + consumed markers - Wired into execute(): cognitive stages only (ISSUE/IMPLEMENT skip); failure → STAGE_FAILED, no advancement - Tests: clean stage passes; source mutation detected → FAILED + no advancement + mutation visible for human review BLE001 narrowed in mutation_boundary (git errors are OSErrors). 1164 tests pass; ruff clean.	2026-09-09 23:57:59 +01:00
Nick	f5ee557ee3	fix(CORRECTION-1/2): D2 producer contract + sentinel-tolerant reader; D3 telemetry wired into production path ... ADAPT-1 baseline corrections (per handoff: fix D2+D3, record D1 as FIXED_DEPTH_ROUTE_INAPPLICABLE — no OUTWARD semantics change). CORRECTION-1 (D2): producer contract enforced - Worker prompts updated: artifact file must contain exactly one JSON document; StageCompleted markers go to stdout ONLY - _load_worker_artifact_file(): canonical reader requires pure JSON; LEGACY_COMPATIBILITY_ONLY permits exactly the known 'StageCompleted: <STAGE>' trailing sentinel (regex-anchored, no arbitrary salvage); corrupt JSON is never masked - Acceptance: pure JSON OK; exact sentinel OK; arbitrary trailing text rejected; corrupt JSON rejected; all 6 stage prompts carry the contract CORRECTION-2 (D3): StageAttempt telemetry wired into real production path - research_event_relay: marks tool windows on tool running/completed/error/ failed events (observational, never changes classification) - StageExecutor._run_relay_with_telemetry: marks model windows around relay.run(); attempt threaded through _execute_stage into all 8 stage methods (telemetry_attempt kwarg, default None) - PROVEN on a real production-path execution: ADVERSARIAL record shows wall=0.3006, model_time=0.3001 (1 window), tool_time=0.3001 (1 window) — overlapping by design, non-additive model respected - telemetry failures never raise into the stage path (observational) D1 NOT a code correction: recorded as FIXED_DEPTH_ROUTE_INAPPLICABLE_TO_LOCAL_EVIDENCE_TASKS — EXPECTED_ROUTING_LIMITATION / ADAPTIVE_REQUIREMENT. OUTWARD external- evidence semantics preserved unchanged. 1163 tests pass (was 1158); ruff clean.	2026-09-09 22:18:50 +01:00
Nick	6c85b8066c	feat(RP12): post-cleanup real-progression re-run on #163 + pipeline fixes proven live ... RP12/#275 — real issue #163 (temporal graph backend spike) through the FULL canonical path on post-RP11 code: ISSUE ✓ → INWARD ✓ → OUTWARD ✓ (45 sources, 10 read w/ receipts, 34 tool calls) → APPLICABILITY ✓ (10 candidates, route INNOVATION_WARRANTED — first live exercise of the conditional route) → INNOVATION ✓ (conclusion ADAPT, 1 candidate with kill_test — NOT manufactured novelty) → ADVERSARIAL ✓ (13 attacks; honest finding: graph tier is NOT bi-temporal as claimed — survivors carry per-candidate dispositions; disposition SURVIVES_WITH_MITIGATIONS) → SYNTHESIS ✓ → QUALIFY REFINE_SYNTHESIS (3 concrete interface gaps — narrowed semantics working) → SYNTHESIS v2 ✓ (refine findings addressed) → QUALIFY v2 PROCEED ✓ (13/13 checks). Pipeline COMPLETE, artifacts persisted to Gitea master. Fixes proven live during the run: - OUTWARD prose-stdout file-channel fallback (RP4 fix backported; first exercise on this path) - INNOVATION executor + gate built (RP5/#268 — first run needing it; conclusion enum enforced, derived_from/kill_test required, NO_INNOVATION_ WARRANTED valid) - adversarial prompt now demands the disposition enum (pre-correction template gap) - adversarial normalizer carries disposition through - synthesis prompt carries QUALIFY refine findings as round-2 constraints - readiness invariant correctly blocked misplaced blocking-unknowns (caught my own error — semantics verified live) Also: retired-workflow references removed from test fixture; no retired code on tree (verified: 0 run_full_research/--watch references). 1130 tests pass; ruff clean.	2026-09-09 13:30:23 +01:00
Nick	25afdc7024	merge: canonical Research-to-Build pipeline (rp2-canonical-stage-model) into master ... RP11/#274 step 1 — consolidate to ONE canonical authority. The RSC0-9 research scripts on master (DEFINE/OUTWARD_DISCOVERY/OUTWARD_READING stage model) are superseded by the canonical pipeline proven through RP2-RP10 (epic #257, children #264-#273): ISSUE→INWARD→OUTWARD→APPLICABILITY→ conditional INNOVATION→ADVERSARIAL(disposition)→SYNTHESIS→QUALIFY. Zero conflicts vs current master; the #161 harness (PR #279) is preserved.	2026-09-09 11:14:49 +01:00
conductor	c7a9a9163a	research-161-r1: QualifyArtifact — PROCEED verdict with source-verified findings	2026-09-09 01:31:36 +01:00
conductor	d60f02e10a	research-161-r1: QUALIFY PROCEED — 13/13 readiness checks, pipeline COMPLETE (canonical RP8). Synthesis e584bbf895b3 ready for literal MiMo Code handoff.	2026-09-09 01:31:35 +01:00
conductor	39b48c9099	research-161-r1: SynthesisArtifact — retrieval-quality benchmark harness build instructions (hash e584bbf895b3)	2026-09-09 01:22:11 +01:00
conductor	ecc7039d84	research-161-r1: SYNTHESIS gate PASSED (harness design, 11 steps, 13 invariants, 10 mitigations carried) — tree at QUALIFY (canonical RP7)	2026-09-09 01:22:10 +01:00
conductor	2baafd6681	research-161-r1: AdversarialArtifact (16 attacks incl. embed-availability + cache-pollution findings, 2 falsification kill-tests, targeted arXiv verification)	2026-09-09 00:46:52 +01:00
conductor	c46680be9d	research-161-r1: ADVERSARIAL gate PASSED (16 attacks, 6 survivors, 1 rejected on duplicate-authority) — tree at SYNTHESIS (canonical RP6)	2026-09-09 00:46:51 +01:00
conductor	2c2df6d3a5	research-161-r1: ApplicabilityArtifact (9 classified candidates, repo-verified integration points)	2026-09-09 00:30:43 +01:00
conductor	5577505ccb	research-161-r1: APPLICABILITY gate PASSED (9 candidates, route STANDARD_SOLUTION_SUFFICIENT) — tree at ADVERSARIAL (canonical RP4)	2026-09-09 00:30:41 +01:00
conductor	9989d6e7fd	research-161-r1: task tree after OUTWARD	2026-09-09 00:16:27 +01:00
conductor	9dc21e1a0a	research-161-r1: OutwardArtifact	2026-09-09 00:16:15 +01:00
conductor	2af31ce492	research: persist tree after OUTWARD (issue #263, round 1)	2026-09-08 23:18:50 +01:00
conductor	b2092ae819	research: persist tree after INWARD (issue #263, round 1)	2026-09-08 23:18:49 +01:00
Nick	9d1e0da8e5	feat(research): OR1 — unify OUTWARD_DISCOVERY+OUTWARD_READING into OUTWARD ... One coherent MiMo external-research pass. Discovery and reading remain visible as subsections inside the artifact, not separate top-level stages. Artifact has discovery and reading subsections. Event relay streams both.	2026-09-08 23:17:51 +01:00
conductor	651715fd80	research: persist tree after ADVERSARIAL (issue #262, round 1)	2026-09-08 23:09:48 +01:00
conductor	d623559c77	research: persist tree after OUTWARD_READING (issue #262, round 1)	2026-09-08 22:40:22 +01:00
conductor	aa2bc29e67	research: persist tree after OUTWARD_DISCOVERY (issue #262, round 1)	2026-09-08 22:35:45 +01:00
conductor	bbb4af2feb	research: persist tree after OUTWARD_DISCOVERY (issue #256, round 1)	2026-09-08 22:32:46 +01:00
conductor	a77d3cad08	research: persist tree after OUTWARD_DISCOVERY (issue #261, round 1)	2026-09-08 22:32:45 +01:00
conductor	aba9802247	research: persist tree after OUTWARD_DISCOVERY (issue #256, round 1)	2026-09-08 22:32:37 +01:00
conductor	a046472db6	research: persist tree after OUTWARD_DISCOVERY (issue #260, round 1)	2026-09-08 22:32:34 +01:00
conductor	c5db9a3923	research: persist tree after OUTWARD_DISCOVERY (issue #262, round 1)	2026-09-08 22:32:29 +01:00
conductor	04b733b645	research: persist tree after OUTWARD_DISCOVERY (issue #260, round 1)	2026-09-08 22:32:27 +01:00
Nick	d5687057ab	chore(research): record local backend execution evidence (DEFINE+INWARD completed, tree on master)	2026-09-08 21:32:57 +01:00
Nick	5a9825fa7f	fix(research): DEFINE and INWARD use deterministic paths — MiMo only for bounded semantic inspection	2026-09-08 21:23:40 +01:00
Nick	efc895bd30	feat(research): R1-R2 deterministic DEFINE + INWARD context builder ... DEFINE: deterministic contract compilation from Gitea issue. No MiMo unless ambiguous. Classifies question type, extracts sub-questions, produces required-stage plan with skip reasons. INWARD context builder: deterministic. Matches issue keywords against umbrella manifest and source/test files. Produces bounded candidate context. MiMo only invoked for semantic inspection of the bounded candidates, not broad agent wandering. METHODOLOGY STAGE ≠ MANDATORY LLM CALL.	2026-09-08 21:18:42 +01:00
conductor	56c0074b96	docs(research): file 4-phase research round for #245 (headless stage)	2026-09-08 09:11:42 +01:00
conductor	f6beca546a	docs(research): file 4-phase research round for #245 (headless stage)	2026-09-08 08:31:50 +01:00
Nick	2a9ddbbf6b	feat(v1): establish runtime authority map	2026-08-29 02:14:47 +01:00
Nick	f12d60c217	feat(tui): add governed local tool objects	2026-08-28 11:26:04 +01:00
Nick	9997dac413	Build mode research: auto-detection, pipeline improvements, contextual awareness ... Three parallel research agents completed: 1. Auto-detection (NOVEL — no existing tool does this): - Simple keyword scoring algorithm (0-3 task, 4-7 task+plan, 8+ build) - Signal categories: request analysis, codebase impact, historical patterns, risk - Zero cost for 90% of requests (regex), LLM fallback for 10% ambiguous - UX: auto-switch high confidence, suggest medium, stay low 2. Pipeline improvements (from Devin, SWE-agent, Aider, Cursor): - File-level isolation (worktrees per batch) - Inline verification after every edit (not just batch boundaries) - Contract testing between batches (highest-value verification) - Selective rollback + failure-aware re-planning after 2 failures - Three-phase execution: Foundation → Implementation → Integration - Trajectory recording for learning - Dynamic batch sizing (1-3 files default, 80%+ success rate) 3. Contextual awareness: - Hybrid keyword+LLM classification - Mode transitions mid-task (upgrade/downgrade with confirmation) - Context signals: codebase, request, conversation, user preferences - Progressive disclosure: start light, escalate if complex Key research insights: - Agent reliability: 96% → 34% from 8K to 128K tokens (LOCA-bench, ICML 2026) - 56.2% of agent turns are exploration, not execution (FastContext, 2026) - Single-file changes: ~80% success vs <50% for 4-6 files (SWE-bench) - Keep detection simple — gauntlet's complexity was its downfall Pinned findings updated at research/build-mode/PINNED-FINDINGS.md	2026-08-23 15:28:00 +01:00
Nick	a26f82b174	Build mode: worktree isolation, ledger, batch sentinels (v0.51.0) ... Build mode orchestrator (build_mode.py): - Full pipeline: ASSESS → RESEARCH → FEASIBLE → CONSULT → CONSENT → PLAN → BUILD → VERIFY → CORRECT → INTEGRATE → RECORD - Worktree isolation: every build creates a worktree for reversibility - Ledger checklist: tracks progress at each step for resumability - Batch sentinels: per-batch verification (not full canary) - Scope guardian: full verification after all batches pass - Recursive retry: 3 attempts max per batch - Resume from last completed phase if credits run out TUI commands: - /build start <description> — start new build - /build status — list all builds - /build resume <id> — resume build from last phase - /build abort <id> — abort build, delete worktree - /build ledger <id> — show build ledger Pinned findings at research/build-mode/PINNED-FINDINGS.md 92/92 canary passing	2026-08-23 15:16:45 +01:00
Nick	b2be0a4d66	Safety noise filter for thinking blocks (v0.45.0) ... Added _filter_safety_noise() to strip safety-related noise from thinking blocks: - 22 patterns covering: AI self-identification, refusal language, safety policy references, ethical hedging, responsibility deflection, apology patterns - Case-insensitive matching - Keeps actual reasoning, removes safety noise - Consecutive blank lines collapsed Use case: abliterated models (0% refusal) still have safety concepts in their training data. The model answers, but its thinking may reference safety. This filter strips the safety noise from the thinking blocks, showing only the actual reasoning in the collapsible ThinkingBubble. 92/92 canary passing	2026-08-23 02:51:56 +01:00
Nick	0bd8f02eec	Tool calling: file read/write/list/search/command for LLM agents (v0.44.0) ... tool_executor.py — new module: - 5 tools: read_file, write_file, list_directory, search_files, run_command - Sandbox: paths resolved + validated against allowed directories - Blocked dirs: ~/.ssh, ~/.gnupg, /etc, /var, /proc, /sys - Atomic writes: temp file → os.replace with .bak backup - Undo stack: last 50 writes reversible - Diff preview: unified diff before any write - Tool call parsing: extracts JSON tool calls from model response - Agent loop: model → tool → result → model continues Operator integration: - Tool definitions added to prompt (JSON schema format) - Tool calls detected in response and executed - Results fed back to model for continuation - Write operations return diff for confirmation TUI integration: - Diff preview shown as confirm message (gold/amber style) - Auto-confirm for now (TODO: user confirmation dialog) Research basis: - Structured outputs (JSON mode) over native tool calling - Grammar-constrained decoding for reliable tool calls - Sandbox + diff + undo for safe file operations - ReAct pattern: model → action → observation → continue 92/92 canary passing	2026-08-23 02:13:13 +01:00
Nick	f1fa2400cc	Fix ruff + pyright issues + batch processing research (v0.42.1) ... Ruff auto-fix (688 issues fixed): - 66 unused imports removed - 43 unsorted imports fixed - 36 deprecated type annotations modernized - 24 whitespace issues cleaned - 156 other auto-fixable issues Critical bugs fixed: - rune_operator.py: model_context used before defined (reordered) - rune_tui.py: sleep_time not imported (added import) - warm_router.py: _load_state undefined (renamed to load_state) Remaining: 258 ruff issues (style nags), 70 pyright errors (type mismatches) Batch processing research: - Optimal batch: 5-15 issues, 1-3 files per LLM session - Group by issue type (not file) for pattern consistency - Context window effective capacity: 25-50% of stated max - Always verify after each batch 54 files changed, 1039 insertions, 763 deletions	2026-08-23 01:34:07 +01:00
Nick	216dea94d3	Add ruff + pyright quality tools (v0.42.0) ... ruff (fast linter): - Configured in pyproject.toml with focused rules: F, E, W, B, SIM, RUF - Ignores style nags (E501, SIM103, SIM115) and Textual patterns (B008, RUF012) - Auto-fixable rules enabled - Current: 953 issues (347 auto-fixable) pyright (static type checker): - Configured in pyrightconfig.json with standard mode - Suppresses noisy optional-member-access warnings - Catches real bugs: undefined variables, type mismatches, missing attributes - Current: 70 errors, 237 warnings Key findings from first run: - 66 unused imports (F401) — auto-fixable - 5 undefined names (F821) — actual bugs - 70 type errors from pyright — None passed where str expected - 2 undefined variables in warm_router.py Research: /research/quality-tools/findings.md 65/65 canary passing	2026-08-23 01:26:50 +01:00