release(v0.65.0): "How much of this is correct by accident?" — 8 artifacts, 7 implemented

[avrabe]

v0.65.0 — "How much of this is correct by accident?"

Cuts the release: version bump, CHANGELOG fold, and the ARCHMODEL deferral recorded with its
blocker re-verified at cut time.

8 artifacts, 7 implemented. RQ-65-PARITY, RQ-65-MUTANTS, RQ-65-ALIASCLASS, RQ-65-DECLINE,
RQ-65-MVPCORE, RQ-65-FUNCN, RQ-65-FLOORSHAPE. RQ-65-ARCHMODEL defers on spar#445 — external,
in another repository, re-verified OPEN on 2026-09-09 (last updated 2026-09-03).

Two things in this PR that would not have reddened anything

ANCHOR_TAG deliberately NOT bumped. claims.yaml carries ANCHOR_TAG = "v0.64.0" as a
load-bearing anchor, not a stale reference. A global 0.64.0 -> 0.65.0 sweep across the 56 files
containing that string would have moved it, and measured in a scratch clone that yields
A0 anchor v0.65.0: AHEAD of the window's previous tag — a red main on the first post-tag run.
It moves in the v0.66 planning PR, when that window opens. Verified after the bump:
status-evidence-anchor: v0.64.0 ... lag 0.

RQ-65-PARITY and RQ-65-FUNCN had NO changelog entry. Both merged before the ## [Unreleased]
header existed (RQ-65-FLOORSHAPE created it), so the release's foundational must artifact was
invisible in the changelog. Nothing in CI reads that file, so nothing would ever have caught it.
Both entries are restored here.

What the release section says about itself

• 19 % mutation survival (4 of 21 byte-changing), frame and seed stated, three red-first
  controls KILLED, framed as an upper bound.
• 21 defects, none from a user report — 20 from running something new, 1 from reading something
  old (The spec-suite gap was declared closed by a COMPILE census — the behavioural half stayed open and nothing sampled it until #197 #1217; no oracle would flag prose).
• Four defects in the instruments: a census compiling .wast on a path that types every
  parameter i32 (i64.add summing one argument's two halves, exit 0, counted ok); a ledger pin
  bound to the decoder while the failure was in lowering; an audit walk whose vfp_defs claimed
  calls do not clobber s0-s15; a delivery floor passing at EXACTLY its floor while 39 of 67 commits
  were missing.
• The subtraction metric went the WRONG WAY: selector +156 lines across three waivers,
  sel_dsl_rules unmoved at 80. Published rather than explained away.
• Decline-honesty's bill, where it lands: feat(reach): RQ-65-MVPCORE (#1017) — MVP core re-derived on the executed path: single-module .wast routing (oracle first), merge refusals, start-function invocation; 14/114 -> 12/80 #1232's refusals shrank the home-register audit's own
  corpus from 8317 to 7191 functions.

Falsification

Three CI-pinned invariants, none asserted in prose alone: a recorded-KILLED mutation surviving its
recorded killer fails mutation-survey-discrimination and voids the 19 %; a change to the audit's
192 planted probes means the instrument moved; a moved parity pin must be explained.

Refs #1189, #197, #1180, #1183, #1017, #1136, #1208

🤖 Generated with Claude Code

https://claude.ai/code/session_01YJK5LZZEkV5smCY1jKn18L

[temper-pulseengine]

Automated review for PR #1234

pulseengine/synth:release/v0.65.0 → pulseengine/synth:main

Verdict: 💬 Comment

Summary: The commit message does not clearly explain the changes made to the codebase. It only mentions that the version number was updated, but it doesn't provide any details about what specific changes were made or why they were necessary.

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

Findings (1):

1. Cargo.toml:1

   version = "0.64.0"
   

   The version number in Cargo.toml was updated from 0.64.0 to 0.65.0.

---

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 7d33268

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!