Phase 0 — wire sync automation into the two unwired zh-cn series (tracking)

[mmcky]

An audit of the zh-cn translation program (2026-07-06) found that of the three launched series, only lecture-python-programming → lecture-python-programming.zh-cn has action-translation sync automation wired in — and it is in excellent shape (27/27 lectures, zero drift, last English change synced same-day, AI review on every PR). The other two launched series have no automation in their source repos and are drifting:

Pair	Coverage	Last content update	Drift since
lecture-python-programming → .zh-cn	27/27 (100%)	2026-06-19 (auto)	0 PRs
lecture-python-intro → lecture-intro.zh-cn	49/52 (94%)	2026-04-09	~28 lecture PRs
lecture-python.myst → lecture-python.zh-cn	79/121 (65%)	2025-11-26	40+ lecture PRs

Phase 0 goal: replicate the proven lecture-python-programming setup on the other two pairs so drift is handled automatically from here on. Catch-up translation of missing lectures in the intermediate series is out of scope (that is Phase 2 of the broader rollout); Phase 0 only reconciles existing drift and turns the automation on.

Tracking

• Track A — intro pair: [action-translation] Wire zh-cn sync automation and reconcile drift (Phase 0, Track A) lecture-python-intro#782
• Track B — intermediate pair: [action-translation] Wire zh-cn sync automation and reconcile drift (Phase 0, Track B) lecture-python.myst#947

Shared prerequisites

• Confirm TRANSLATION_PAT and ANTHROPIC_API_KEY secrets are available to both source repos (same secrets the lecture-python-programming workflows use)
• Decide action version pin for new workflows (lecture-python-programming currently pins QuantEcon/action-translation@v0.15.0)
• Regenerate onboarding reports with current tooling before any reconciliation — the sample report in [action-translation] Sample onboarding report lecture-python-intro -> lecture-intro.zh-cn lecture-python-intro#676 dates from 2025-12 and is stale

Definition of done

Both source repos have a sync-translations-zh-cn.yml workflow that opens a translated PR in the target repo when an English lecture PR is merged, both target repos run the review workflow on translation PRs, and one end-to-end sync has been verified on each pair (via a real merged PR or \translate-resync).

Related: the onboarding-report sample in QuantEcon/lecture-python-intro#676 demonstrates the per-file drift-decision flow this rollout depends on.

[mmcky]

Track A (intro ↔ intro.zh-cn) is complete — closed via QuantEcon/lecture-python-intro#782 (2026-07-18): 41-lecture forward resync merged, .translate/ + heading maps onboarded, sync/review/rebase wired @v0, e2e verified (QuantEcon/lecture-intro.zh-cn#267, review PASS).

The run doubled as a production shake-down of the CLI path and produced QuantEcon/lecture-python.myst#104–#107 plus sightings on QuantEcon/lecture-python.myst#89/#94 — worth triaging before Track B (intermediate ↔ lecture-python.zh-cn) runs the same playbook at 2.5× the file count: the QuantEcon/lecture-python.myst#106 discovery gap and QuantEcon/lecture-python.myst#107 localisation-regression class are the two that cost real review effort this time.

[mmcky]

Progress reconciliation (2026-07-21) — the body's tracking list predates both waves; here is live state against the definition of done.

Requirement	intro pair (Track A)	intermediate pair (Track B)
Source repo sync workflow	✅ merged 2026-07-18 (lecture-python-intro#795), pinned @v0	🔴 the one remaining gap — myst#979 open; ANTHROPIC_API_KEY org-secret visibility to verify
Target review workflow	✅ wired + field-verified (~15 live events)	✅ zh-cn#100 merged; on @v0 with today's PAT + fork guard
One verified end-to-end sync	✅ \translate-resync → intro.zh-cn#267 (reviewed, merged), plus first organic sync observed 2026-07-19 (intro.zh-cn#268)	🔴 blocked on the sync workflow; mix_model remains the natural candidate (fresh drift from myst#978)

Both drift-recovery waves completed and merged (41 PRs intro, 69 intermediate). The shared-prerequisite checkbox "decide action version pin" is settled program-wide: floating @v0 (project-translation#9, closed today). The stale-report checkbox is superseded — both pairs got full status --check-sync inventories during their waves.

Track A is done to the definition. Track B needs exactly two things: land myst#979 and verify one end-to-end sync. Per the sequencing decision in project-translation's PLAN, the intermediate audit precedes turning that automation on, so drift-recovery is verified converged before sync starts maintaining it.

[mmcky]

Track B's remaining two requirements now have a dedicated execution plan with live drift state and resume pointers: QuantEcon/project-translation#14 (audit → drift mini-wave → land myst#979 → e2e verify → close this tracker).

[mmcky]

Track B is done — the last gate, one verified organic end-to-end sync, was met today.

QuantEcon/lecture-python.myst#979 merged (c10a3a8b), so sync-translations-zh-cn.yml is live on @v0 with both org secrets confirmed visible. Two \translate-resync triggers then exercised the full path: source comment → sync run → translated PR with the correct labels → verdict-v2 AI review on @v0 → verdict block posted. Results: QuantEcon/lecture-python.zh-cn#198 (inventory_dynamics, from myst#623) and #199 (perm_income, from myst#975). Both source PRs had merged before the workflow existed, so their drift had never reached the target — the resyncs both verified the path and did real reconciliation work.

That means both tracks are complete and this tracker can close once #198 and #199 merge (the edition is not literally in-sync until they do). Track A was verified 2026-07-19 on QuantEcon/lecture-intro.zh-cn#268.

Three findings from the runs worth carrying forward, since they are the kind of thing that only shows up in production:

Reading the run list is actively misleading. A sync applies its labels in a single call, GitHub emits one labeled event per label, and here the labels were applied twice — so ten Review Translations runs appear, nine correctly skipped by the label filter and one success. At a glance that reads as total failure, which is the same surface as the #131 symptom. Check for the single success before concluding anything.

The two PRs gated for different reasons, and only one was valid. #198 was routed to editor by a genuine catch: the glossary specifies 边缘分布 for "marginal distribution" and the translation used 边际分布 throughout. #199 was routed by diff check failed: positionCorrect, with three confident findings asserting the source's code change was never applied — verified false, the target carries it identically (def time_path(T, rng), rng.standard_normal, the added rng = np.random.default_rng(), both call sites). Filed as #148: the four diffChecks are model output but gate absolutely, so a hallucinated boolean routes a correct PR to a human and, more importantly, will bias the Stage 4 floor calibration that #147 exists to gather.

Coverage reached 121/121 first, which is what made #979 safe to merge — its own body had held it back on the grounds that 42 lectures were untranslated, a condition resolved by the #196 seed on 2026-07-22 and verified live before merging.

Refs: QuantEcon/project-translation#14 (Track B in-sync tracker), #147, #148, QuantEcon/project-translation status/intermediate-zh-cn.md.

[mmcky]

Phase 0 complete — both tracks

QuantEcon/lecture-python.zh-cn#198 and #199 merged, so the intermediate pair is now 121 files, 121 aligned — zero drift, zero gaps. Both tracks of this tracker are finished:

• Track A — intro pair — wired 2026-07-18, organic sync observed 2026-07-19 (🌐 [translation-sync] FIX: remove unused get_cmap and to_rgba imports in laffer_adaptive lecture-intro.zh-cn#268)
• Track B — intermediate pair — 69-PR resync wave merged 2026-07-19, coverage completed 2026-07-22, automation live and e2e verified 2026-07-23

The engine is now deployed across 21 workflows in 11 repos, every one on @v0 — all three English source series wired, and all five editions receiving automated sync, AI review and rebase. Catalogue: QuantEcon/project-translation#7.

Suggest closing this tracker. QuantEcon/lecture-python.myst#947 (the Track B tracker in the source repo) can close with it.

What Phase 0 deliberately did not do

Worth restating at close, because the scope decision still holds: Phase 0 was "wire automation only; catch-up translation deferred". Catch-up in fact happened along the way for the intermediate pair (the 37-lecture seed in #196), so all five editions are now at full coverage — but the seeded lectures are unreviewed machine output, and that debt is real and growing. It sits with the human-review program (QuantEcon/project-translation#15), not here.

Findings from Track B worth carrying into the next phase

Silent-drift defects do not announce themselves. The test harness this program validates releases on had been drifting for months — source workflows six releases behind, and hand-written .translate/ state using key names the engine does not recognise, so readFileState() returned undefined and both harness targets read as untranslated. Fixed under #109. Nothing had failed; it had just stopped testing what it appeared to test.

Merged, released and working are three separate states, each requiring its own check. #131 shipped but silently skipped review; the v0.22.0 release PR merged without anyone cutting the tag; and a green workflow run can mean the job was skipped by a label filter that no longer matches.

A run list of skipped is not failure. A sync applies labels in one call, GitHub emits one labeled event per label, and the labels can land twice — ten review runs appear where nine are correctly skipped and one succeeds.

The reviewer can be confidently wrong. On #199 it asserted, in three separate findings, that the source's code change had never been applied; hand-diffing showed the target carried it identically. Filed as #148, because the four diffChecks gate absolutely but are model output.

[mmcky]

Both tracks look complete as of 2026-07-23 — recommending closure.

The tracking checkboxes in the body were never ticked, so recording the state here rather than leaving it to inference.

Track A — intro pair (lecture-python-intro → lecture-intro.zh-cn): done 2026-07-18. Drift recovered via a 41-PR translate forward wave (#225–#265, all merged), .translate/ metadata and heading maps onboarded (zh-cn#266), sync/review/rebase wired (intro#795 + zh-cn#212), and end-to-end verified — a \\translate-resync on intro#793 produced zh-cn#267, AI-reviewed PASS, merged. First organic sync observed 2026-07-19 (zh-cn#268 from intro#798).

Track B — intermediate pair (lecture-python.myst → lecture-python.zh-cn): done 2026-07-23. The 69-PR resync wave merged 2026-07-19; coverage reached 121/121 with the #196 seed on 2026-07-22; sync automation landed with QuantEcon/lecture-python.myst#979 on 2026-07-23 with both org secrets verified visible. The last gate — one verified organic end-to-end sync — was met twice, by \\translate-resync on myst#623 and #975, producing zh-cn#198 and #199 with full verdict-v2 blocks.

So the Phase 0 goal as stated — replicate the proven programming setup on the other two pairs so drift is handled automatically from here on — is met: all three English source repos are wired and all five editions receive automated sync PRs with AI review and rebase.

Catch-up translation of missing lectures was explicitly out of scope here (Phase 2), and that has since been done anyway for the intermediate series.

Follow-up work that outlived this tracker, and should not hold it open:

• Native review of the 37 machine-translated seeds in lecture-python.zh-cn, and of the new Tier 1 exercise sections
• The terminology audit at Audit: 边际分布 → 边缘分布 for "marginal distribution" across 8 remaining lectures (54 occurrences) lecture-python.zh-cn#201
• -n -W + set -eo pipefail on lecture-intro.zh-cn's ci.yml

Closing is yours to call — I have not closed it, since a tracking issue's completion is a judgement about scope rather than a fact I can verify from the checklist.

[mmcky]

Closing — Phase 0 is complete. Checkboxes ticked; two of them need a note rather than a silent tick.

Definition of done, item by item

The stated bar was: both source repos have a sync workflow that opens a translated PR when an English lecture PR is merged, both target repos run the review workflow, and one end-to-end sync has been verified on each pair.

	Status
Source sync workflows	✅ lecture-python-intro (2026-07-18) and lecture-python.myst (QuantEcon/lecture-python.myst#979, 2026-07-23)
Target review workflows	✅ both, plus rebase
End-to-end verified — Track A	✅ \\translate-resync on intro#793 → zh-cn#267, AI-reviewed PASS, merged. First organic sync observed 2026-07-19 (zh-cn#268 from intro#798)
End-to-end verified — Track B	✅ twice — \\translate-resync on myst#623 and #975 → QuantEcon/lecture-python.zh-cn#198 and #199, both with full verdict-v2 blocks

Drift reconciliation, which was the other half of the goal: 41 PRs on Track A (#225–#265) and 69 on Track B (zh-cn#101–#171), all merged.

Two checkboxes that deserve an explanation

"Confirm TRANSLATION_PAT and ANTHROPIC_API_KEY secrets are available to both source repos" — done, but the premise carried a wrong name. The cross-repo push secret is QUANTECON_SERVICES_PAT (org-level), not TRANSLATION_PAT; that was found during the fr rollout's prerequisite check on 2026-07-14. Both org secrets were verified visible to lecture-python.myst before #979 merged, and ANTHROPIC_API_KEY was separately granted to the fa harness target on 2026-07-23.

"Regenerate onboarding reports with current tooling" — ticked as superseded rather than done as written. The intent was to replace the stale 2025-12 sample in QuantEcon/lecture-python-intro#676 before reconciling. What actually happened is better: both pairs got full translate status --check-sync triage inventories during their own Phase 0 waves (2026-07-18/19), archived in project-translation/reports/. The #676 sample is history and the reconciliation it was meant to inform is finished.

The remaining item — "decide action version pin" — resolved 2026-07-18 as floating @v0 fleet-wide, which has since carried five releases to the estate without hand-bumping.

What outlives this tracker, and should not hold it open

• Native review of the 37 machine-translated seeds in lecture-python.zh-cn, and of the new Tier 1 exercise sections
• The terminology audit at Audit: 边际分布 → 边缘分布 for "marginal distribution" across 8 remaining lectures (54 occurrences) lecture-python.zh-cn#201, and the fleet-wide adherence question now at Measure glossary adherence across the estate — delivery is fixed, compliance has never been measured #154
• -n -W + set -eo pipefail on lecture-intro.zh-cn's ci.yml, plus a cold Build Cache dispatch on its main

Catch-up translation was explicitly out of scope here (Phase 2) and has since been done for the intermediate series anyway — coverage is 121/121.

Estate state at closure: all three English source repos wired, all five editions receiving automated sync PRs with AI review and rebase, engine at v0.23.0 via @v0.