docs(i18n): Rules 15×8 coverage audit = 120/120 OK + Fallacies EN residuals NO-OP

docs/investigations/2026-07-03-rules-i18n-coverage-audit.md

@@ -0,0 +1,68 @@
+# Rules i18n Coverage Audit (15 records × 8 languages) — data pendant of verdict #140
+
+**Date:** 2026-07-03 (po-2024, base master `9c19e51a`)
+**Scope:** `Cards/Rules/Argumentum Rules - Cards.csv` — cell-by-cell coverage audit, **read-only** on the CSV.
+**Dispatch:** ai-01 deep-queue (msg `kp26j6`) — data pendant of the visual verdict #140 (5 game sets confirmed visually on the FR TarotCards bundle).
+
+## TL;DR
+
+- **Primaire (résidus trad Fallacies EN) = NO-OP.** 0 empty `desc_en` cell across all 1408 Fallacies rows; PK 371 and 607 (cited as empty) are **already populated** (120 / 79 chars). Scanner #647 TOTAL = 0; 0 EN cell identical-to-FR; 0 EN cell reading as French. The #351/#308 drift (40 PK) was resolved in an earlier cycle. Nothing to fill, no gpt-5.5 call warranted.
+- **Secondaire (audit Rules coverage) = 120/120 OK.** All 15 records × 8 languages are populated, in the correct script, with 0 residual contamination. Data-side confirmation of the visual verdict #140 (5 game sets intact).
+
+## Rules audit — coverage matrix (code=truth)
+
+15 records (`Rules_01`…`Rules_15`), 8 language columns (`Text`/`Text_en`/`Text_ru`/`Text_pt`/`Text_ar`/`Text_es`/`Text_zh`/`Text_fa`). For each cell: populated? + script matches expected (latin for fr/en/es/pt, cyrillic for ru, arabic for ar/fa, cjk for zh)?
+
+| pk | fr | en | ru | pt | ar | es | zh | fa |
+|---|---|---|---|---|---|---|---|---|
+| Rules_01 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_02 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_03 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_04 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_05 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_06 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_07 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_08 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_09 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_10 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_11 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_12 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_13 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_14 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| Rules_15 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+
+**Empty cells: 0. Script mismatches: 0.** All 120 cells populated with the correct script for their language.
+
+### False-positive probes (verified legitimate, not contamination)
+
+Two flags from the initial heuristic scan were investigated and **cleared**:
+
+- **Rules_01 `Text_zh`** flagged "latin?" by the script heuristic. Real content: `# Argumentum\n## 说谎者学校` — "说谎者学校" (5 hanzi, "School of Liars"). The latin-script word "Argumentum" (the brand name, kept in latin across all languages) inflated the latin count past the threshold; the CJK content is present and correct. ✅ Not a defect.
+- **`embromador` (PT, Rules_04/05/06/12)** flagged by a garbage probe. Real meaning: the **canonical PT term for "baratineur / smooth talker"** (the player role), used 77× across `Cards/Scenarii/...Cards.csv` PT. It is the correct game-term translation, not MT garbage. ✅ Not a defect. (The probe was over-broad — `embromador` was incorrectly borrowed from the #642 worklist garbage list.)
+
+### Post-#640 contamination re-check
+
+Scanner #647 on `Cards/Rules/Argumentum Rules - Cards.csv` = **0 finding**. The #633 garbage ("English Channel", "picareta", etc.) purged by #640 is confirmed gone — no residual MT contamination in any language column.
+
+## Primaire (Fallacies EN residuals) — NO-OP evidence
+
+The dispatch cited "2 empty `desc_en` (PK 371 and 607)" + "re-scan drift EN/RU/PT (#351/#308 = 40 PK)". Verified empirically (code=truth, per [[matcher-no-match-is-not-content-absent]] / [[test-counter-empirical-dotnet-test]]):
+
+| Check | Method | Result |
+|---|---|---|
+| Empty `desc_en` (FR canon non-empty) | `csv` scan, 1408 rows | **0** |
+| PK 371 `desc_en` | direct read | **120 chars** (populated) |
+| PK 607 `desc_en` | direct read | **79 chars** (populated) |
+| Scanner #647 drift | committed scanner | **TOTAL: 0** (all datasets) |
+| EN cell identical-to-FR | accent-normalised compare | **0** |
+| EN cell reading as French | stopword-density heuristic | **0** |
+
+The PKs cited exist (PK range 0–1407), but their `desc_en` are **already filled**. The #351/#308 drift (40 PK) was resolved in an earlier cycle. **No fill, no gpt-5.5 call warranted** — reporting the no-op rather than fabricating work.
+
+## Method / reproducibility
+
+- Rules audit: `csv` read of `Cards/Rules/Argumentum Rules - Cards.csv`, script classification per cell (latin/cyrillic/arabic/cjk by Unicode block counts), garbage probe + manual verification of the 2 flags.
+- Fallacies no-op: `csv` read of `Cards/Fallacies/Argumentum Fallacies - Taxonomy.csv`, direct PK reads, scanner #647 run.
+- Read-only on both CSVs. 0 write.
+
+Relates #640 (Rules i18n refonte), #647 (scanner), #140 (visual verdict), #351/#308 (drift, resolved). Base master `9c19e51a`.