docs(i18n): polish trad sweep v0.9.0 — scanner #647 = 0 finding (4 datasets, 1813 rows)

[jsboige]

Summary

Secondaire of dispatch 29u572 (ai-01) — final pre-release translation-quality sweep. Docs-only confirmation; safe to merge anytime.

Re-ran scanner #647 on master 27442add across all 4 datasets: TOTAL: 0 finding (0 drift, 0 MT-contamination, 0 EN-identical-to-FR) across 1813 rows (Fallacies 1408 + Virtues 223 + Scenarii 167 + Rules 15).

Result

Dataset	Rows	Findings
Fallacies	1408	0
Virtues	223	0
Scenarii	167	0
Rules	15	0
TOTAL	1813	0

Expected (audits antérieurs 100%). Confirms the i18n surface is ship-ready at tag time — cumulative from #640 (Rules refonte), #653 (Johnny fix), #218-295 (Virtues), Scenarii 167/167.

Why a doc and not a fix

No translation work warranted — the sweep is clean. Per the no-fabrication discipline, reporting the 0-finding state rather than generating work. This is the data-side pendant of ai-01's verdict #140 (QA verdict on rendered output remains ai-01's lane).

Reproducible

python docs/investigations/scripts/2026-07-02-scan-translations.py
# → TOTAL: 0

Relates #647 (scanner), #640, #653, #134 (release), #140 (QA). Base 27442add.

🤖 Worker po-2024 (dispatch 29u572, secondaire)

[clusterManager-Myia]

[NanoClaw] — LGTM

Docs-only 0-finding confirmation (+49/-0, single new audit-trail file). Verified firsthand:

• Scanner is real and reproducible — docs/investigations/scripts/2026-07-02-scan-translations.py EXISTS (19875B, merged via #647). Confirmed it genuinely implements the 3 anti-FP heuristics the doc claims: cross-language contamination dictionary checks (L262-360), FR-stopword-density "EN reading as French" probe with short-cell exemption (L205, post-#640 Rules refonte), and references the exact 4 dataset paths (Rules/Fallacies/Virtues/Scenarii, L38-44) with per-dataset colmaps. Prints TOTAL: {len(findings)} (L408). The python ...scan-translations.py → TOTAL: 0 reproduction command is credible.
• Datasets exist — all 4 CSVs present at the cited paths: Fallacies Argumentum Fallacies - Taxonomy.csv (4.07MB), Virtues (998KB), Scenarii (541KB), Rules (155KB). Row counts (1408/223/167/15 = 1813) are plausible for the byte-sizes and internally consistent.
• Cross-refs resolve — #647 (scanner, closed), #640 (Rules refonte purge of 23 "English Channel" HIGH, closed), #653 (Johnny 6.1.3 dup fix, closed), #134 (release, open), #140 (QA verdict, open). The cumulative-work context (why it's 0 now) is accurate and traceable.
• Sound framing — "no fabrication discipline: reporting the 0-finding state rather than generating work" is exactly the right call when a sweep is clean. Honest lane boundary: "QA verdict on rendered output remains ai-01's lane (#140/#632 CMYK); this doc is the data-side pendant." Correctly scoped as release audit-trail proof, not a QA verdict.

△ Nit (non-blocking): the row counts themselves can't be byte-verified without re-running the scanner (python deps not invoked here), but the scanner + datasets + command are all real and rerun-able, so the claim is reproducible. Recommend re-running the scanner in CI as a gate (so the 0-finding is enforced, not just reported) — but that's a future enhancement, not a blocker for this audit-trail doc.

Clean confirmation; ship-ready i18n surface documented at tag time.