W0 closeout: harness citation scenario (#256.3 verified), deck refresh, D3 decision record

[mmcky]

The W0 (#258) tail, each item decided by the owner in turn. Fixes #7. References #256 (defect 3 verified — evidence commented there) and #221 (closes when this lands, via the decision record).

#256.3 — citation backfill verified on the harness, against the deployed release

The harness fixtures carried zero {cite} keys — the reason #117 was unexercisable at v0.25.0 release gating. This PR adds scenario 27 plus the fixtures that arm the guard: target repos now reset with a minimal _config.yml (whose bibtex_bibfiles is what switches the backfill on) and a references.bib deliberately missing the test key, while the source repo carries it (ArrowDebreu1954).

Already run, first try, scoped to zh-cn at --action-ref v0 — confirmed beforehand that v0 peels to c74e3aa = v0.25.0, the deployed bytes: sync run 31448329289 green, and QuantEcon/test-translation-sync.zh-cn#710 delivers all three files — the translated lecture with the {cite} role intact in Chinese prose, references.bib with the entry appended under the append-only #117 marker, and the state file. #256's defect-3 box and #258's verification box are ticked with this evidence.

One finding recorded for #259: the PR's translation-sync-metadata files[] declares only the markdown — the delivered bib and state file are undeclared. Delivery exceeding declaration is the benign inverse of #256 defect 1; the declared-vs-delivered assertion should reconcile both directions.

Estate note: the scoped run leaves the source test repo carrying only the zh-cn sync workflow (documented scoped-run behavior); the next unscoped harness run restores fa/ml.

#7 — deck refreshed (owner call: refresh, not wontfix)

presentations/action-translation-sync.* → action-translation.md, content rewritten against v0.25.0: Claude Sonnet 5, three operational modes, verdict-block routing with shadow mode, deterministic guardrails, the translate CLI surface, trust gates, the real production estate, and current counts (1,516 tests, 27 E2E scenarios). PDF/HTML regenerated with marp 4.5.0 and visually checked page-by-page; stale v0.7.0-era claims (Sonnet 4.5, "357 terms", 183 tests) are gone. build.sh and the presentations README follow the rename.

D3 — bot identity direction recorded

.dev/decisions/D-2026-08-11-bot-identity-machine-user.md: when the sync bot's identity moves off the personal PAT it moves to the quantecon-services machine user, not a GitHub App — #221's measurement (Apps always report author_association=NONE, so an App silently breaks the resync trust gate in ~17 gate copies) is the deciding evidence, plus the two riding constraints (name the token, and nothing the bot posts may contain the command string). Migration stays deferred; #61 remains the migration task; #221 closes once this record is on main.

Verification

Suite 1516/1516, lint, format:check, check-dev-refs (168/168) all green. bash -n on the harness script plus a --dry-run --scenarios 27 smoke before the real run. dist-action/ changes only in the sourcemap (a comment edit in fetchBibliographies whose "harness repos have no _config.yml" claim this PR makes stale).

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR closes out the remaining W0 items by (1) extending the GitHub E2E harness with a new citation/backfill scenario to verify the deployed release behavior end-to-end, (2) refreshing and renaming the Marp slide deck to reflect the current action-translation system, and (3) recording the D3 decision on bot identity in .dev/decisions/ (with supporting log + changelog entries).

Changes:

• Add E2E scenario 27 plus fixtures (_config.yml + references.bib) to make bibliography backfill exercisable in harness runs.
• Refresh the presentations deck (rename to action-translation.*, update content, regenerate artifacts) and update the build script/README accordingly.
• Add D3 decision record and W0 closeout log entry; document the harness verification in CHANGELOG.md.

Reviewed changes

Copilot reviewed 14 out of 19 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tool-test-action-on-github/test-action-on-github.sh	Reset source/targets with bib fixtures; adds scenario 27 and updates scenario counts/messages.
tool-test-action-on-github/test-action-on-github-data/base-references.bib	New source-side bibliography fixture containing the backfill key.
tool-test-action-on-github/test-action-on-github-data/base-references-target.bib	New target-side bibliography fixture deliberately missing the test key.
tool-test-action-on-github/test-action-on-github-data/base-config.yml	New minimal target _config.yml fixture arming bibtex_bibfiles.
tool-test-action-on-github/test-action-on-github-data/27-add-citation-minimal.md	New scenario input introducing {cite} usage.
tool-test-action-on-github/README.md	Updates harness documentation to 27 scenarios and documents the new Phase 5 scenario.
src/index.ts	Updates fetchBibliographies doc comment to reflect new harness fixture reality.
presentations/README.md	Updates instructions and filenames to action-translation.* and notes refresh context.
presentations/build.sh	Updates build script to build action-translation.{pdf,html} from action-translation.md.
presentations/action-translation.md	New refreshed Marp deck source content.
presentations/action-translation.html	Regenerated HTML output for the refreshed deck.
presentations/action-translation.pdf	Regenerated PDF output for the refreshed deck.
presentations/action-translation-sync.md	Removes the old deck source.
presentations/action-translation-sync.html	Removes the old generated HTML output.
CHANGELOG.md	Adds an Unreleased entry documenting scenario 27 and the deployed-release verification evidence.
.dev/log/2026-08-11-w0-closeout.md	Adds W0 closeout session log capturing decisions/verification outcomes.
.dev/decisions/D-2026-08-11-bot-identity-machine-user.md	Records D3 bot identity direction and constraints for the eventual migration.

---

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