docs(taxonomy): #847 AIF audit Layer (c) label-distortion flag (staged, read-only) — MEASURE not verdict

[jsboige]

What

#847 AIF structural audit — Layer (c) label-distortion flag (dispatched by ai-01 msg-20260723T024907-2l19wg, tick 89, PRIMARY). The §2.1 half of Layer (c), companion to #853 (the §2.2 reference-reachability half, merged 9f3caaef). Read-only — 0 run live, 0 prod-CSV. MEASURE, not verdict. INPUT for ai-01 synthesis, GATED jsboige ratification.

§2.1 has two bullets — only one is mechanically computable

§2.1 bullet	What	Mechanical?	This instrument
(2) lexical availability	term borrowed from another tradition	Yes (lexical facts)	✅ borrowed_from_lang
(1) distorted scope	label stretched/narrowed	No (semantic)	⛔ proxies tested, non-discriminatory (§3)

Headline (39 grouping nodes)

• 1 borrowed: Ad hominem (raw Latin — confirmed against the taxonomy's own Latin column).
• 38 unflagged: no mechanical distortion signal (unflagged ≠ "label fits" — it's the absence of a flag).
• Axis-C lexical pressure on the grouping layer is LOW (1/39): the taxonomy is FR descriptive phrases / FR calques, importing raw Latin only where no FR single-term exists.

The dead proxies (saves future work — §3)

Two obvious mechanical proxies for stretched/narrowed were tested and found non-discriminatory (reported in CSV as context, NOT used to derive label_fit):

1. Concept-dispersion (dispersion_fr) — saturated: median 0.94 (d1) / 1.00 (d2/d3). Each fallacy IS its own Wikipedia article → distinct-slugs ≈ leaves everywhere → can't distinguish tight from sprawled.
2. EN-grouping divergence (n_en_groupings) — 1:1 everywhere: the EN grouping columns are a FR translation/calque, NOT an independent EN-tradition tree → no axis-C structural signal lives in the grouping layer.

⇒ A defensible stretched/narrowed flag needs human semantic read (or a gated LLM pass). This delivers the mechanically-grounded part and documents the dead-end so no future worker re-tries it.

Cross-layer value — resolves the c-DEFER on Ad hominem

Node	label_fit	acompte tag	Combined reading
Ad hominem (d2)	borrowed (latin)	A-candidate	tradition-divergence → bridge-node, do NOT bill as defect

Corroborates #853 (Ad hominem's Obstruction base = 69% Wikipedia-anchored). The mechanism-heterogeneity is the tradition's (Latin scholastic term spanning cultures), not a jsboige grouping error. The bridge holds.

Of the 7 A-candidate (tree-tension) nodes, only Ad hominem carries a mechanical axis-C label signal; the other 6 (Erreur de raisonnement, Argument bâclé, Causalité douteuse, Mauvaise interprétation, Pensée biaisée, Biais naturels) have no labelling-distortion flag → left as axis-A candidates for jsboige (layer c bills none as tradition artifacts).

Governance

• MEASURE not verdict. 0 reorganisation wording. Synthesis + verdict = ai-01/jsboige. Same governance as docs(taxonomy): #847 AIF structural-homogeneity acompte (layers a+b, read-only) — MEASURE not verdict #850/docs(taxonomy): #847 AIF audit Layer (c) reference-reachability (staged, read-only) — MEASURE not verdict #853.
• 0 prod-CSV (T&A freeze). Docs-only artefact. DatasetUpdater untouched (Enabled=false).
• borrowed = lexical fact (taxonomy Latin column + morphology, conservative). unflagged = absence of flag, not a positive "exact" verdict.
• Calques (e.g. Pétition de principe ← petitio principii) = gray zone, not flagged (FR words, not a raw borrow) — flagged for human read.

Artefact (docs-only)

docs/taxonomy/847-layerc-label-distortion.{md,csv} — 39 internal nodes.

Refs

Method aif-structural-audit-method.md §2.1, §4, §5. Companion #853 (9f3caaef). Acompte #850 (ccf9d911). Tracking #847, chantier #498. Base master 82a1e027.

🤖 Generated with Claude Code