[surlymochan]

v0.5.33 — Milestone M13 "声称抽取 A/B 口径修复" accepted

Patch release. One merged PR since 0.5.32 (PR #179, squash ceece47), and the milestone it completes is now accepted.

Milestone M13 accepted (issue #69; acceptance detail issue #177, comment 6815860645; #177 closed 2026-11-12T05:50:45Z). parseClaimCounts now distinguishes the two X/Y claim writings: a second number smaller than the first is read as passed/failed (472/12 → failed 12, 240/0 → failed 0), a second number greater than or equal to the first is read as passed/total (61/64 → failed 3). Slice 1 (PR #179) adds the disambiguation rule plus regression tests for both writings and boundaries.

Acceptance evidence (measured, #180 comment 6815694863): the main round directly measured 8 groups on merged main (ceece47) — all correct; slice 1's 6 new regression assertions independently re-run and passing. #160 real re-review: verdict still mismatch, 5 claims, first claim corrected to 472 passed / 12 failed (total unset), other 4 identical to the M12 JSON; the 3 mismatches are all replay environment gaps / pre-existing failures (go not on PATH, pre-existing npm test i6d failure, skill:diff environment gap), 0 pointing at a PR quality issue. #173's two pytest claims still parse as 61 passed / total 64 / failed 3 — the new reading does not harm passed/total claims. Regression: audit suites 120/123, the 3 failures same names and same cause as the clean-main baseline (I6d pytest, pytest not on PATH), zero regression.

Honest note: the three mismatches in the #160 re-review are replay environment gaps / pre-existing failures, not PR quality issues; correcting the claim to 472/12 does not flip #160's mismatch verdict. This milestone closes the extraction bug filed as issue #175 (closed 2026-11-12T05:51:04Z, together with evidence issue #180, closed 2026-11-12T05:50:55Z).

Acceptance ledger: the issue #5 ledger is settled as a conclusion on the MiniMax full-matrix recheck (135/135 complete, 2026-11-10) — P2-1 passed, P1-2 / P1-3 / P2-3 did not pass — so no bench numbers are claimed in this release.

Local test (this branch): full npm test 275 tests, 274 pass / 1 fail — the single failure is the pre-existing test/i6d-acceptance-demo.test.js:52 ([GREEN] FAIL: expected verdict=verified exit=0, 1 !== 0), which fails identically on a clean-main (ceece47) baseline re-run of the same file; unrelated to this release (version/docs only). The three I6d pytest failures did not appear in this full run. drift:scan exit 0; pack:check exit 0 (dry-run tarball @surlymochan/evalid@0.5.33, 208 files, 510.8 kB). skill:diff fails on this VM only — missing ~/.codex/skills/xflow (environment gap, identical message on clean main). release:pack as a chain stops at verify because of the single pre-existing failure; the remaining gates were run once each above.

Files: package.json, package-lock.json, CHANGELOG.md, RELEASE_NOTES.md, docs/launch-dossier.md. No code changes. Do not merge in this round — merge after main-round review.