docs(taxonomy): #498 coverage-status refresh — two-layer finding + reconciliation trajectory

[jsboige]

Summary

Refresh the stale docs/taxonomy/498-coverage-status.md (dated 2026-07-05, master 70bd1605) with a code=truth scan at master 7796c127. Docs-only proposition (gated) — 0 prod CSV write. Flagged stale by both ai-01 and po-2024 (the §5 next-targets were wrong post-#753/#760).

Central finding — the AIF mapping has two layers, never reconciled

The "coverage" number depends on which AIF layer you measure:

Métrique	Rows	% of 1408	Serialized by
Attack layer (AIF_attackType+attackedNode)	93	6.6%	#753 (back-fill §7) + #760 (scaleup)
Skos layer (skosDirectRef/ExceptionRef/MappingType)	70	5.0%	pre-chantier baseline
Union (any AIF signal)	145	10.3%	—
Fully-modeled (both layers)	18	1.3%	intersection

The two layers overlap only at 18 rows. The previous "70/1408 (5.0%)" was not wrong — it measured the skos layer, which predates the attack columns. "93" measures the attack layer. Two distinct counters, resolved here.

Cross-tab (code=truth, 1408 rows) — the reconciliation gap

	skos ✓	skos ✗
attack ✓	18 (fully-modeled)	75 (attack-only)
attack ✗	52 (skos-only)	1263 (unmapped)

• 52 skos-only — vetted native tokens, no attack columns. Half-done. Highest-ROI back-fill, 0 fabrication risk (tokens already exist + verified). Family: Cheating 13, Insufficiency 10, ML 9, Faulty logics 8, Math 6, Influence 3, Obstruction 3.
• 75 attack-only — 44 ML (skos already proposed + ratified in merged cluster docs PR-5..12, awaiting deep serialization) + 31 scaleup (need skos modeling, fail-loud docs(taxonomy): #141 crossLink enrichment sample run (real output curated from #626) #677).

Reprioritized trajectory (§5)

Reconcile the two layers before opening new clusters ex-nihilo:

• P1 — back-fill attack columns for the 52 skos-only rows (bounded, vetted, 0 fabrication).
• P2 — deep-serialize the already-proposed skos for the 44 ML attack-only rows.
• P3 — new cluster docs on small clean residuals (Complication exagérée 6u, Opération inappropriée 5u) + Obstruction rebut nuclei (Évasion 1313, Procès en incohérence 1361).

Completing the half-done rows lifts 2-layer completeness 1.3% → ~7-8% for bounded effort.

Distribution (code=truth)

61 undercut / 29 undermine / 3 rebut — RA/I/CA deterministic map (#707§4 (a)) exact. The 3 rebut (1282/1313/1361) = first Fallacies rebut, all Obstruction.

Discipline

• ✅ Read body before action — refreshed against the actual CSV (code=truth scan), not the memo. Resolved the 70-vs-93 discrepancy empirically before publishing any number.
• ✅ Docs-only, gated — 0 prod CSV write; only 498-coverage-status.md changed (198 ins / 128 del). No .mcp.json/tmp//__pycache__ staged.
• ✅ 0 fabrication docs(taxonomy): #141 crossLink enrichment sample run (real output curated from #626) #677 — references only native tokens in use (42 DirectRef + 50 ExceptionRef distinct).
• ✅ VERIFIED (scan) vs RAPPORTÉ (git log) vs PENDING (docs(taxonomy): #498 Fallacies AIF — Relativisme abusif cluster (first rebut cluster, gated proposal) #766) labeled throughout.
• ❌ [runtime pending] feat(dnn): #649 RulesExplorer i18n — mirror loc() culture cascade #674/[HOLD post-tag] fix(cardpen): #629 convert relative /Cards paths to raw-master URLs (Option 1) #666/chore(dnn): #131 migrate 12 RazorComponent templates to Razor14 (source-level, runtime pending) #596 NOT touched. No self-merge — gated review ai-01.

🤖 Worker po-2024 — autonomous chantier-doc refresh (dispatch ai-01 hgicfj, primaire).