Repository navigation
Run python.zh-cn and programming.zh-cn; publish the first three-edition report #1
Description
Run configs/quantecon/python-zh-cn.yml and programming-zh-cn.yml at current main plus intro.zh-cn; hand-adjudicate a sample of ≥30 tier assignments and ≥50 pair categories per edition (the review-table habit from the study); write the first dated report in project-translation reports/ with S1–S3, F1–F5 per edition and stated per-class precision. python.zh-cn carries the 2026-07 resync wave over 146 files (F2 at scale) and 12 probable whole-file AI passes that QuantEcon/project-translation#22 segregated — these need either overrides or an ai-assisted reading. programming.zh-cn is engine-era from the start with multi-revision state files — the first place the state-file baseline strategy can be tried.
Part of QuantEcon/project-translation#49 (lifecycle metrics, Stage 1). Method and limits: docs/method.md.
Activity
Validate baseline.strategy; align docs with implemented metrics (#9)
Implement the state-file baseline strategy (#1) (#11)
Add reviewed tier overrides for python.zh-cn's probable whole-file AI…
Add reviewed tier overrides for python.zh-cn's probable whole-file AI…
Add the engine's per-file commit subject to programming.zh-cn sync si…
mmcky commentedon Aug 28, 2026
Delivered 2026-08-28. The report is merged as reports/2026-08-28-three-edition-lifecycle-baseline.md (QuantEcon/project-translation#53), with S1–S3 and F1–F5 per edition and stated per-class precision.
What the run required, as this issue anticipated: python.zh-cn's probable whole-file AI passes took the reviewed overrides file (#12 — 7 commits to
ai-assisted, 1 English re-seed toseed, adjudicated from before/after pairs rather than assumed from the plan's instance count). programming.zh-cn was the first outing for thestate-filebaseline strategy (#11), which agreed withscript-jumpon all 28 documents including the adjacent-commit edge case.Adjudication: 106 tier assignments and 179 pair categories sampled (seed 20260828). Tier axis 106/106 after one systematic catch the adjudication itself produced — the engine's 2026-03/04
Update translation:per-file wave readai-assisteduntil #13 added the signature. Pair axis measured indicative: 58% / 58% / 36% precision with three named causes; the dominant one (unpaired additions/deletions not prose-gated) is fixable in textstrata and recommended in the report.Headline: human-authored prose at HEAD 20.5% / 6.2% / 0.0% across intro / python / programming; median baseline survival 48% / 65% / 96%; intro's resyncs consumed human-touched lines at 59% against a 20.5% corpus share — the 2026-08-04 finding reproduced and generalised.
#2 (engine-version strata) and #6 (overwrite evidence) are now unblocked; the report names what each needs.