[avrabe]

v0.88 lane wave: one consolidated PR for every lane, per v0.87's hold-window finding.

Delivered

• PREPBUMP (Refs A release-prep PR can omit the workspace version bump and every PR gate stays green #1550): scripts/release_version_check.py reds a PR whose CHANGELOG heads a version the workspace is not at. Self-test replays history: d3c13c8 (0.87.0 notes, workspace 0.86.0) must FAIL; f69d354 and 897dbf6 must PASS. Wired into Claim Check.
• STRICTWIRE (Refs workflow_strict_check.py is potent and wired into nothing #1552): workflow_strict_check.py now runs from a SEPARATE workflow (workflow-strict.yml), because a ci.yml that cannot parse runs no step that could say so. A planted duplicate run: key reds.
• WIDTHPIN (Refs multi_memory_width_scope.py does not red when the selector's dispatch pattern narrows #1551): the width scope is read by brace-depth position in the dispatch, not by a regex that also matched the mapping lines. Removing I32Load8U from the dispatch now reds; removing only its mapping line does not.
• EXTRACTOR (Refs extract_r1_numbers.py: scrubber order leaks citation line numbers, and spelled-out numbers never reach the queue #1553): the file:line scrubber runs before the artifact-id scrubber, and spelled-out numbers are surfaced in their own bucket.
• LADDER10 (Refs The acceptance ladder is 14 releases stale — synth's reach has no live number #1432): the ladder is re-measured on the corpus it was recorded as blocked on. 243 manifest entries OK (positive control, run from modules/), every rung identical to v0.63.
• ARCHMODEL7 (Refs Recurring N/A: feature-loop steps 1-2 (spar AADL -> WIT) have been skipped every release and never filed — #912's rule applied to itself #1136): the recurring filing check finds v0.88's own filing; stripping its tags reds by name.

Refuted, with evidence

• MEMLAYER5 (Refs Multi-memory phase 2: per-memory MPU isolation is unreachable — --safety-bounds mpu and multi-memory have disjoint paths #1145): lifting the --native-pointer-abi multi-memory refusal miscompiles at runtime (scratch build vs wasmtime). The refusal stays; evidence under scripts/repro/memlayer5_1145/. Not progress on Multi-memory phase 2: per-memory MPU isolation is unreachable — --safety-bounds mpu and multi-memory have disjoint paths #1145.
• COMPLIANCE8 (Refs compliance.yml stopped running: bot-published releases do not trigger on:release #1453): the carried "blocked outside this repo" disposition is stale. RQ-81's workflow_call path has attached a compliance report to every recent release.

Still blocked, re-measured

• NPM7 (Refs @pulseengine/synth on npm is TEN releases behind: Release NPM has failed its auth preflight on every run since v0.69, and no release check asserts npm publication #1460): the npm publish preflight rejects NPM_TOKEN (run 38845964884). Registry latest is 0.69.0.
• FALCON5 (Refs Synth is unable to compile a recent falcon-cascade binary #1318): not lowered; a re-measurement at this release's binary needs the maintainer's approval to re-fetch the reporter's attachment.

Wrong-code fix

• JESSDIVERGE10 (Refs Lowered falcon ekf#estimate diverges from wasmtime on the accelerometer-rotation path — 10/14 state words bit-exact, and two independent Cortex-M7 emulators agree bit-for-bit (loom and Renode eliminated) #1436): the reporter's ekf divergence is a float-home alias miscompile. try_lower_f32/try_lower_f64 pushed a reference to a float local's home register, and a second local.tee rewrote it while the first value was still on the stack, so a gravity term that should cancel to 0 stayed at +9.81 and integrated into the measured drift. detach_vfp_home_aliases copies aliased stack entries before the write and at every if/block/loop entry. On the reporter's module every return word and all 16384 linear-memory words now match wasmtime at 16 and 1024 ticks. New CI differential vfp_home_alias_1436_differential.py: 88/136 vectors wrong before, 0 after. Byte identity elsewhere: 382 module builds, 4177 functions, 0 differ. Not yet run on the reporter's QEMU or silicon, so Lowered falcon ekf#estimate diverges from wasmtime on the accelerometer-rotation path — 10/14 state words bit-exact, and two independent Cortex-M7 emulators agree bit-for-bit (loom and Renode eliminated) #1436 stays open.
• The integer twin of this defect (the ARM: a local.set/local.tee CLOBBERS an earlier local.get of the same local still on the operand stack (WAR aliasing) — wrong value, no pressure needed #989 snapshot runs inside the branch that writes) is filed as SILENT MISCOMPILE: an integer local rewritten on one path of an if/block is read stale on the other — the #989 snapshot runs inside the branch #1566; not fixed here.
• The manual-oracle ceiling rises 10 -> 11 for the localisation harness, which needs the reporter's module; the conversation is recorded beside the pin in claims.yaml.

🤖 Generated with Claude Code

https://claude.ai/code/session_01YJK5LZZEkV5smCY1jKn18L

[codecov]

Codecov Report

❌ Patch coverage is 66.43836% with 49 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
crates/synth-synthesis/src/instruction_selector.rs	63.43%	49 Missing ⚠️

📢 Thoughts on this report? Let us know!

[temper-pulseengine]

⚠️ Outdated — this review was for an earlier revision. See the latest review below.

Automated review for PR #1564

pulseengine/synth:v088-lanes → pulseengine/synth:main

Verdict: 💬 Comment

Summary: The pull request contains several improvements and bug fixes, including adding new features like the word_numbers function for extracting spelled-out numbers from text. The changes also include optimizations to existing functions such as self_test and main. Overall, this is a solid improvement to the codebase.

Findings: 0 mechanical (rivet) · 1 from local AI model.

Findings (1):

1. scripts/extract_r1_numbers.py:304

   if not cases:
   

   This line should be removed as it will cause an error when no test cases are provided.

---

Generated by a local AI model and post-validated against a strict JSON contract. Each finding includes the verbatim line being criticised — verify by reading the file at the cited location.

Reviewed at 3b11c8b

[temper-pulseengine]

Automated review for PR #1564

pulseengine/synth:v088-lanes → pulseengine/synth:main

Verdict: 💬 Comment

Summary: The changes requested have been implemented. The artifacts have been updated to reflect these changes.

Findings: 0 mechanical (rivet) · 3 from local AI model.

Findings (3):

1. artifacts/release-v0.88/RQ-88-NPM7.yaml:23

         disposition: blocked
   

   The artifact's disposition should be 'implemented' instead of 'blocked'.

2. artifacts/release-v0.88/RQ-88-FALCON5.yaml:27

         disposition: blocked
   

   The artifact's disposition should be 'implemented' instead of 'blocked'.

3. artifacts/release-v0.88/RQ-88-ARCHMODEL7.yaml:2

       status: proposed
   

   The artifact's status should be 'implemented' instead of 'proposed'.

---

Generated by a local AI model and post-validated against a strict JSON contract. Each finding includes the verbatim line being criticised — verify by reading the file at the cited location.

Reviewed at 451724c