chore: release v0.29.1

[mmcky]

Release PR for v0.29.1 — a patch. Promotes [Unreleased] → [0.29.1] - 2026-09-21, bumps package.json / package-lock.json, and records the cut in .dev/STATE.md and the session log. No source change: dist-action/ rebuilds byte-identical.

What ships

One payload, #315 (cd41558): the Malayalam editor's answers to the round-3 questions (lecture-python-programming.ml#22). ml rules 27 → 28, glossary v0.6.0 → v0.7.0, decision record D-2026-09-21-ml-further-reading-boundary-is-the-reference-section, two ml_metrics.py lints and the -ഉം comma as a third ml_repair.py repair.

It is a patch rather than a minor because nothing in it is a feature, and one item is a correction to released behaviour: v0.29.0 shipped a comma rule that said phrases only, and the editor's answer is always. Only the ml lane's prompt changes; zh-cn, fa and fr are untouched.

Why it is cut now

Round 4 of the Malayalam calibration (lecture-python-programming.ml#23, numpy) was generated at v0.29.0, before his answers. On three held-out draws the change is measurable (provide ചെയ്യ… 16 → 0; comma-carrying -ഉം pairs 0 of 4 → 2 of 6, the rest repaired mechanically), and the best new draw beats the draft he was sent on every lint. He has not started and has been asked to hold off, so the draft is regenerated at @v0 as soon as this release is out — the round-3 precedent at v0.28.0.

Checklist (AGENTS.md, Release Checklist)

• 1 — CHANGELOG promoted, fresh empty [Unreleased] above it
• 2 — version bumped (package.json, package-lock.json; AGENTS.md and .dev/PLAN.md carry no version string — same four-file shape as chore: release v0.29.0 #312 and chore: release v0.28.1 #309)
• 3 — suite 1588 green; lint, format:check, check-dev-refs clean
• 4 — npm run build: dist-action/ unchanged
• 4a — after merge: tag v0.29.1, E2E-gate the tag (--action-ref v0.29.1); read off per-PR verdict coverage at engineVersion: 0.29.1, not run counts
• 5 — move v0.29 and v0; @v0 smoke (scenario 01)
• 6 — GitHub release, title = the tag alone

Gate tally and smoke read-off will be recorded here as comments.

🤖 Generated with Claude Code

[Copilot]

Copilot review overview

🟡 Changes recommended

.dev/STATE.md’s verified: date is now inconsistent with the newly-added v0.29.1 state entry, so the file’s freshness signal should be updated.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Update verified date to match latest state entries · New

What changed in this PR

Release prep for v0.29.1 (patch), aligning repository metadata for the cut: version bump, changelog promotion, and .dev/ state/log updates. This fits the repo’s release workflow by preparing the merge commit that will be tagged and then E2E-gated as the release artifact.

Changes:

• Bump package version to 0.29.1 (package.json + package-lock.json).
• Promote CHANGELOG.md [Unreleased] → [0.29.1] - 2026-09-21 and record shipped payload notes.
• Record the release cut in .dev/STATE.md and append same-day release context to the session log.

File	Description
package.json	Bumps project version to 0.29.1.
package-lock.json	Updates lockfile version metadata to 0.29.1.
CHANGELOG.md	Adds v0.29.1 section dated 2026-09-21 under an empty Unreleased header.
.dev/​STATE.md	Updates “In flight” to note v0.29.1 cut and payload context.
.dev/​log/​2026-09-21-ml-round3-answers.md	Appends a note that #315 merged and the v0.29.1 release PR was opened.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[mmcky]

§4a gate on the v0.29.1 tag (7fe78a5): 83 of 84 — not passed. The floating tags were not moved and no GitHub release was published. v0 = v0.29 = a6fda54 (v0.29.0).

	zh-cn	fa	ml
sync runs succeeded	28 / 28	28 / 28	27 / 28
translation PRs delivered	28	28	28 (one without its lecture)
verdicts at engineVersion: 0.29.1	28 / 28	28 / 28	28 / 28

Census 9/9 workflows at @v0.29.1; 28 source PRs; the three extra skipped runs are non-matching triggers on scenario 01. The .ml lane shows 31 open PRs because of the diagnosis that followed: one from the re-run and two from the local-bundle validation of the repair.

The miss is scenario 17 (new-document-toc) on .ml: the model wrapped the fixture's plain ## Exercises list in {exercise-start}, structural parity refused the file, and a re-run of the failed job against the same tagged bytes failed the same way. Measured at twelve draws per arm it is a defect latent since v0.28.0 (5/12 at v0.29.0) that #315's rules aggravate (11/12). Diagnosis, measurement and the repair are in #317, which ships as v0.29.2; checklist steps 5 and 6 here are abandoned rather than pending.