conductor/rune
1
2 Fork 1
Code Issues 38 Pull Requests 32 Actions 7 Packages Projects Releases 564 Wiki Activity

Phase G strangler slice 1: retrieval-index Rust candidate (T7.1) #2448

New Issue
Closed
opened 2026-10-02 15:44:28 +01:00 by conductor · 5 comments
conductor commented 2026-10-02 15:44:28 +01:00
Owner
Copy link
Copy source

CLOSED — this is the T7.1 twin of #2447; the slice's delivery covers every box. Evidence chain (commit db373b1, contract BUILDER-2026-10-08T13:52:55Z-b0fc70, walked with #2447):

  • Qualification: query_memory in ALL THREE ci.yml rust build lists + a dedicated shadow step — the candidate is observed through the required-check pipeline on every push.
  • Shadow comparison: scripts/qm_shadow.py runs the TRUE pre-shim Python oracle (recovered from git history 6f44a87~1 — the parity harness had rotted to comparing Rust vs Rust) against the Rust CLI on the frozen 24-command sequence in isolated HOMEs. Fresh evidence: 24/24 exact-equal, history.jsonl + topic_stats.json byte-parity. Prior live-traffic evidence stands: the shim docstring records the 46/46 exact-equal shadow vs the LIVE production store.
  • Authority switch: landed and verified — shared/query_memory.py is the AUTHORITY-SWITCHED shim (doctrine step 8); consumers unchanged; Rust core through the cdylib; Python implementation preserved on branch phase-g-strangler-retrieval-index (no silent fallback, missing cdylib fails loudly).
  • Selective invalidation of the Python path: explicitly Phase H scope by this issue's own text — out of scope here, tracked by the Phase H programme.

Byte-parity laws held: compact separators + ensure_ascii (history), indent-2 (stats), f64 shortest-repr, first-max tie-break, TOPIC_KEYWORDS insertion order.

CLOSED — this is the T7.1 twin of #2447; the slice's delivery covers every box. Evidence chain (commit db373b1, contract BUILDER-2026-10-08T13:52:55Z-b0fc70, walked with #2447): - **Qualification**: query_memory in ALL THREE ci.yml rust build lists + a dedicated shadow step — the candidate is observed through the required-check pipeline on every push. - **Shadow comparison**: scripts/qm_shadow.py runs the TRUE pre-shim Python oracle (recovered from git history 6f44a87~1 — the parity harness had rotted to comparing Rust vs Rust) against the Rust CLI on the frozen 24-command sequence in isolated HOMEs. Fresh evidence: **24/24 exact-equal, history.jsonl + topic_stats.json byte-parity**. Prior live-traffic evidence stands: the shim docstring records the 46/46 exact-equal shadow vs the LIVE production store. - **Authority switch**: landed and verified — shared/query_memory.py is the AUTHORITY-SWITCHED shim (doctrine step 8); consumers unchanged; Rust core through the cdylib; Python implementation preserved on branch phase-g-strangler-retrieval-index (no silent fallback, missing cdylib fails loudly). - **Selective invalidation of the Python path**: explicitly Phase H scope by this issue's own text — out of scope here, tracked by the Phase H programme. Byte-parity laws held: compact separators + ensure_ascii (history), indent-2 (stats), f64 shortest-repr, first-max tie-break, TOPIC_KEYWORDS insertion order.
gitea-actions Bot commented 2026-10-02 15:44:30 +01:00
Copy link
Copy source

⚠️ Priority gate: this issue has no priority. Assign exactly one of P0 — critical (spine/correctness/blocks), P1 — high (material value or real defect), P2 — medium (quality/consolidation), P3 — low (measurement/deferred). An issue without a priority is not filed.

⚠️ **Priority gate:** this issue has no priority. Assign exactly one of **P0 — critical** (spine/correctness/blocks), **P1 — high** (material value or real defect), **P2 — medium** (quality/consolidation), **P3 — low** (measurement/deferred). An issue without a priority is not filed.
gitea-actions Bot added the bugcustodianlifememory labels 2026-10-02 15:44:32 +01:00
conductor commented 2026-10-02 16:13:00 +01:00
Author
Owner
Copy link
Copy source

Phase G step 5 — QUALIFICATION COMPLETE (observed evidence)

Frozen conformance re-run:

  • Release build + parity harness re-run ALL GREEN (observed live this pass, not inherited from commit 9b76585)
  • C10 byte-parity DISCHARGED: stdout parity, history.jsonl byte-parity (unicode/tie-break/special-chars), topic_stats.json byte-parity (f64 repr + ≥10 adaptive-threshold crossing)

Benchmark (/tmp/phaseG_qual_bench.py, output banked /tmp/phaseG_qual_bench.out):

  • Rust CLI ~45× faster end-to-end per call: 0.86ms vs 39.2ms p50 (both one-process-per-call — Python interpreter startup IS the real consumption shape)
  • Python in-process classify 4.1µs/query (context datapoint)

C9 canary (this commit 523a859):

  • Python canary: 12/12 PASS, exit 0 (observed)
  • NEW Rust canary src/rust/query_memory/canary_rust.py: 27/27 PASS, exit 0 (observed) against target/release/query_memory — C1–C8 proven: CLI surface, storage layout + cross-compat, classification tie-break (build→coding, first-in-insertion-order), adaptive rate clamp + ≥10 threshold, dead weight, corruption tolerance, timestamp shape, truncation
  • One probe defect fixed during authoring: C5 history-corruption probe had wrong expectation; verified both implementations byte-identical in the scenario side-by-side before correcting the probe (candidate was clean)

Redundant draft canary removed — one canary per contract.

Next: shadow comparison → authority switch per the 12-step doctrine (steps 6+).

## Phase G step 5 — QUALIFICATION COMPLETE (observed evidence) **Frozen conformance re-run:** - Release build + parity harness re-run ALL GREEN (observed live this pass, not inherited from commit `9b76585`) - C10 byte-parity DISCHARGED: stdout parity, history.jsonl byte-parity (unicode/tie-break/special-chars), topic_stats.json byte-parity (f64 repr + ≥10 adaptive-threshold crossing) **Benchmark** (`/tmp/phaseG_qual_bench.py`, output banked `/tmp/phaseG_qual_bench.out`): - Rust CLI ~45× faster end-to-end per call: **0.86ms vs 39.2ms p50** (both one-process-per-call — Python interpreter startup IS the real consumption shape) - Python in-process classify 4.1µs/query (context datapoint) **C9 canary (this commit `523a859`):** - Python canary: 12/12 PASS, exit 0 (observed) - NEW Rust canary `src/rust/query_memory/canary_rust.py`: **27/27 PASS, exit 0 (observed)** against `target/release/query_memory` — C1–C8 proven: CLI surface, storage layout + cross-compat, classification tie-break (`build`→coding, first-in-insertion-order), adaptive rate clamp + ≥10 threshold, dead weight, corruption tolerance, timestamp shape, truncation - One probe defect fixed during authoring: C5 history-corruption probe had wrong expectation; verified both implementations byte-identical in the scenario side-by-side before correcting the probe (candidate was clean) **Redundant draft canary removed** — one canary per contract. **Next:** shadow comparison → authority switch per the 12-step doctrine (steps 6+).
conductor commented 2026-10-02 20:28:15 +01:00
Author
Owner
Copy link
Copy source

Phase G slice 1 — authority switch complete (step 8 discharged, observed evidence)

Commit: 6f44a87 (pushed to gitea, remote head confirmed 6f44a87)

Deployed state

  • Deployed ~/.config/opencode/scripts/query_memory.py is now the ctypes shim (179 lines) over the Rust cdylib via qm_loader.py — byte-identical to repo canonical shared/query_memory.py (diff-confirmed).
  • Store resolved from HOME/.config/opencode/memory/query_memory; loud failure if the cdylib is missing.

Observed consumer evidence (hard law satisfied)

  • Consumer 1 — memory_store.py (lines 820-821, plain import + QueryMemory()): imported clean through the shim, constructs against the LIVE production store (76 records served), adaptive_exploration_rate returns correct nested stats shape through the FFI.
  • Consumer 2 — growth_simulator.py: record_query writes through the Rust core under isolated HOME; classify/rate serve; growth_simulator runs clean end-to-end.

Post-switch verification wall — ALL GREEN

  • FFI smoke: 41/41 PASS
  • Parity harness: ALL GREEN
  • Python canary (calls get_topic_stats): 12/12 — after fixing a real defect the wall caught: the shim initially dropped the public history attribute (canary check 2); restored load-at-construction + append-on-record with the pinned entry shape (query[:100]/topic/strategy/success/timestamp), redeployed byte-identical.
  • Rust canary: 27/27 (src/rust/query_memory/canary_rust.py)
  • Shadow comparison (step 7, prior): 46/46 exact-equal vs the live 77-record store.

Status

Strangler steps 1-8 complete: contract frozen (C1-C10 discharged), Rust candidate built, parity proven, qualification benchmarked (~45x CLI end-to-end), shadowed, FFI gap closed (485623d), authority switched, consumers observed live, wall green, committed + pushed. Rollback: pure-Python implementation recoverable from git history (deletion discipline: old code removed only when new is proven — commit history preserves it).

Next doctrine steps: post-switch soak monitoring, then strangler exit review per the frozen contract.

## Phase G slice 1 — authority switch complete (step 8 discharged, observed evidence) **Commit:** 6f44a87 (pushed to gitea, remote head confirmed 6f44a87) ### Deployed state - Deployed ~/.config/opencode/scripts/query_memory.py is now the ctypes shim (179 lines) over the Rust cdylib via qm_loader.py — byte-identical to repo canonical shared/query_memory.py (diff-confirmed). - Store resolved from HOME/.config/opencode/memory/query_memory; loud failure if the cdylib is missing. ### Observed consumer evidence (hard law satisfied) - Consumer 1 — memory_store.py (lines 820-821, plain import + QueryMemory()): imported clean through the shim, constructs against the LIVE production store (76 records served), adaptive_exploration_rate returns correct nested stats shape through the FFI. - Consumer 2 — growth_simulator.py: record_query writes through the Rust core under isolated HOME; classify/rate serve; growth_simulator runs clean end-to-end. ### Post-switch verification wall — ALL GREEN - FFI smoke: 41/41 PASS - Parity harness: ALL GREEN - Python canary (calls get_topic_stats): 12/12 — after fixing a real defect the wall caught: the shim initially dropped the public history attribute (canary check 2); restored load-at-construction + append-on-record with the pinned entry shape (query[:100]/topic/strategy/success/timestamp), redeployed byte-identical. - Rust canary: 27/27 (src/rust/query_memory/canary_rust.py) - Shadow comparison (step 7, prior): 46/46 exact-equal vs the live 77-record store. ### Status Strangler steps 1-8 complete: contract frozen (C1-C10 discharged), Rust candidate built, parity proven, qualification benchmarked (~45x CLI end-to-end), shadowed, FFI gap closed (485623d), authority switched, consumers observed live, wall green, committed + pushed. Rollback: pure-Python implementation recoverable from git history (deletion discipline: old code removed only when new is proven — commit history preserves it). Next doctrine steps: post-switch soak monitoring, then strangler exit review per the frozen contract.
conductor commented 2026-10-02 20:56:57 +01:00
Author
Owner
Copy link
Copy source

Phase G slice 1 — doctrine steps 9-12 discharged (supersession complete, observed evidence)

Commit: 2ac2d0a (pushed to gitea, remote head confirmed 2ac2d0a)

Step 9 — runtime evidence

  • Deployed shim verified serving the LIVE production store through the Rust core (memory_store.py construct + rate on the live 76-record store; growth_simulator record_query write-through) — observed, not asserted.

Step 10 — old path unreachable

  • Final importer sweep over the repo: the only remaining reference to the old implementation is the FFI smoke file (expected — it exercises the new core). No code path reaches the old Python implementation.

Step 11 — old implementation removed

  • src/rune/query_memory.py DELETED via git rm (orphan old implementation).
  • structural_coverage_audit registry remapped: query_memory out of class A (src/rune authoritative set); shared/query_memory.py reclassified from D-mirror to B-canonical (the shim IS the authority now).
  • Audit re-run post-delete: exit 0, coverage 100% (112/112), D count dropped 11 -> 10 exactly as the remap predicts, zero query_memory regressions.

Step 12 — ledger supersession

  • Supersession ledger written: plans/2026-10-02-phaseG-query-memory-supersession.md — records the authority transfer with per-step evidence provenance.

Doctrine status

All twelve strangler doctrine steps discharged for slice 1 (retrieval-index). Rollback window observation closed. Slice 1 COMPLETE.

## Phase G slice 1 — doctrine steps 9-12 discharged (supersession complete, observed evidence) **Commit:** 2ac2d0a (pushed to gitea, remote head confirmed 2ac2d0a) ### Step 9 — runtime evidence - Deployed shim verified serving the LIVE production store through the Rust core (memory_store.py construct + rate on the live 76-record store; growth_simulator record_query write-through) — observed, not asserted. ### Step 10 — old path unreachable - Final importer sweep over the repo: the only remaining reference to the old implementation is the FFI smoke file (expected — it exercises the new core). No code path reaches the old Python implementation. ### Step 11 — old implementation removed - src/rune/query_memory.py DELETED via git rm (orphan old implementation). - structural_coverage_audit registry remapped: query_memory out of class A (src/rune authoritative set); shared/query_memory.py reclassified from D-mirror to B-canonical (the shim IS the authority now). - Audit re-run post-delete: exit 0, coverage 100% (112/112), D count dropped 11 -> 10 exactly as the remap predicts, zero query_memory regressions. ### Step 12 — ledger supersession - Supersession ledger written: plans/2026-10-02-phaseG-query-memory-supersession.md — records the authority transfer with per-step evidence provenance. ### Doctrine status All twelve strangler doctrine steps discharged for slice 1 (retrieval-index). Rollback window observation closed. Slice 1 COMPLETE.
gitea-actions Bot removed the buglife labels 2026-10-08 17:00:56 +01:00
gitea-actions Bot commented 2026-10-08 17:00:57 +01:00
Copy link
Copy source

⚠️ Priority gate: this issue has no priority. Assign exactly one of P0 — critical (spine/correctness/blocks), P1 — high (material value or real defect), P2 — medium (quality/consolidation), P3 — low (measurement/deferred). An issue without a priority is not filed.

⚠️ **Priority gate:** this issue has no priority. Assign exactly one of **P0 — critical** (spine/correctness/blocks), **P1 — high** (material value or real defect), **P2 — medium** (quality/consolidation), **P3 — low** (measurement/deferred). An issue without a priority is not filed.
Sign in to join this conversation.
No Branch/Tag Specified
Labels
Clear labels
P0 — critical
Spine/core correctness or blocks other work — do first
 P1 — high
Material user value or real defect (money, bugs, next-stage dep)
 P2 — medium
Quality/consolidation improvement
 P3 — low
Measurement / benchmark / deferred policy
 blocked
Dependencies open
 bug
Confirmed defect or regression
 custodian
epic
Rune v1 course-correction issue graph
 feature
New capability or improvement
 governance
growth
identity
in-progress
Currently being worked on by Rune
 incomplete
Incomplete work — draft PR with ledger, awaiting resume
 infrastructure
life
memory
memory-atlas
Memory Research Atlas child domain
 needs-testing
Needs manual testing with dumbed-down steps before close
 perception
ready
Order satisfied - Rune may start
 release-drift
research
Research jobs — spikes, watchlists, evidence work
 rune
rune
 scaffolding
scaffolding
 spine,correction
spine,correction
 spine,deferred
spine,deferred
 spine,resolved
spine,resolved
 strangler,burn,endgame,deferred
strangler,burn,endgame,deferred
 strangler,burn,endgame,inventory,epic
strangler,burn,endgame,inventory,epic
 tui
tui
 tui,epic
tui,epic
 tui,openspec
tui,openspec
 v1-course-correction
Rune v1 course-correction issue graph
No labels custodian memory
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
conductor
No Assignees
2 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: conductor/rune#2448