[hoklims]

Summary

Freezes the candidates of evidence-v3 rewritten from assertSame(a, b) to Bun's native expect(a).toBe(b) (ADR 0004), as a new replay-corpus entry candidates-v3-tobe:

• inputs in tests/replay-corpus/inputs/candidates-v3-tobe/ (same worlds and policy as v3, one designated test per candidate);
• a 5.0.0 manifest produced by assertledger@2.1.0 with bun-test-designated, decision VERIFIED;
• corpus.json entry with rewriteOf: "evidence-v3", replayed by the current build like every other entry.

tests/replay-corpus.test.ts gains a rewritten v3 candidates suite:

• every v3 evidence entry has a rewrite;
• the rewrite is re-derived from the frozen v3 request: same worlds, same policy, same candidate ids and files, each file equal to the mechanical assertSame → expect().toBe() rewrite;
• the rewritten manifest reaches the v3 decision, the same per-candidate status, kills, reasons and gates, and the same observation outcomes.

E11 (#111) established that toBe accepts exactly the pairs Object.is accepts on Bun 1.4.2, which is what makes this rewrite verdict-preserving.

Proof

Negative witnesses in a disposable worktree at 783df8e, each red on ERR_ASSERTION in the designated test, then green after restore:

Witness	Mutant	Red test
rewrite-matcher	strong.test.ts: .toBe( → .toEqual(	designates expect(a).toBe(b)…
rewrite-worlds	rewritten request.json: world 1 subject altered	designates expect(a).toBe(b)…
verdict-divergence	rewritten manifest: WEAK_ORACLE → ELIGIBLE	reaches the verdicts…
dropped-rewrite	corpus.json: rewriteOf removed	rewrite every v3 candidate corpus

• pnpm check: 995 tests pass, 0 fail.
• The fixture inputs/candidates-v3-tobe/repository/package.json is {"type":"module"}, no scripts: no route change.

Limits

• The equivalence covers the two v3 candidates and Bun 1.4.2 only.
• The manifest was produced by 2.1.0 on Windows x64, trusted-local mode; replayed by 2.1.0 and the current build only.
• The future native-matcher profile (v6) must reach the same verdicts; not checked here.
• No independent proof auditor ran.

Test plan

• pnpm check
• Four negative witnesses red then green
• CI green

Refs #84

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-27T14:47:23.994809Z	783df8e	PR opened
🔒 Security Review	✅ Completed 2026-10-27T14:48:21.686157Z	783df8e	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.