docs(taxonomy): #499 Virtues scale-up phase 2 batch 2 — Rigueur mathematique depth-3..4 leaves (gated)

[jsboige]

Summary

#499 Virtues scale-up — Phase 2, batch 2: Rigueur mathématique family (16 depth-3…4 leaves).

Batch 1 (Langage exact, 14 leaves) was merged via #530 (fc2013fc). This batch 2 confirms the method reproduces across families (different mirror family, different scheme distribution).

Family	leaves	status
Langage exact	14	✅ batch 1 (#530 merged)
Rigueur mathématique	16	✅ this batch
Raisonnement valide	51	next
Échange enrichissant	40	next
Argument pertinent	29	next
Honnêteté intellectuelle	23	next
Présentation intègre	21	next
Remaining after this batch	164

Method — identical to Phase 1 (#510) and batch 1 (#530)

• gpt-5.5 /v1/responses reasoning:low, grounding catalog of 7 family PKs + 63 real depth-3 fallacies.
• Family mirror hard constraint: every Rigueur mathématique leaf → Erreur mathématique (PK 594).
• Three independent anti-fab layers, re-checked against the REAL corpus:
  1. Catalog membership — every opposed PK ∈ the 63-set.
  2. Ground-truth — every PK re-verified against the real 1408-row Fallacies CSV, opposed-PK ↔ text_fr character-for-character.
  3. Mirror consistency — every row's prevented_family_pk = 594.
• Result: 16/16 annotated, 0 violations across all 3 layers.
• Anti-fab validator tools(taxonomy): reusable anti-fab validator for scale-up CSVs #518 (kind=virtues): ✓ CLEAN — 16 rows, 0 HARD, 0 WARN.

Granularity note

Depth-4 leaves oppose the most specific applicable math fallacy — e.g. "Exactitude numérique" (pk 75) → Erreur de calcul (681) + Imprécision (667), not a broad sampling fallacy. The "Sign" scheme dominance (10/16) reflects the probabilistic/correlational nature of most math-rigor virtues — a meaningful cross-family contrast with batch 1 (Langage exact, where "Verbal Classification" dominated).

Files

• docs/taxonomy/499-virtues-scaleup-phase2-rigmath.md — proposal (method, scope, anti-fab, the 16 rows).
• docs/taxonomy/499-scaleup-phase2-rigmath-annotations.csv — 16 rows, 10-col presentation schema.

Gate boundaries (HARD, all respected)

• ❌ No production Argumentum Virtues - Taxonomy.csv change — docs/taxonomy/ only.
• ❌ No OWL / EPITA consumer / cards / mindmaps touched.
• ❌ Generation script + raw model output ephemeral (tmp/, not committed).
• ❌ Batch 2 of 7; 5 families (164 nodes) remain.

GATED proposal — worker signals structure + grounding; ai-01 spot-checks, jsboige validates content.

🤖 Worker po-2024

[clusterManager-Myia]

[NanoClaw]

LGTM. Batch 2 Rigueur mathematique — 16 depth-3/4 leaves sous Erreur mathematique (594). Meme methode/schema/anti-fab que batch 1, validator CLEAN sur 16 rows, gated proposal. Les opposed PKs sont plausibles (e.g. "Raisonnement bayesien" → 644+707, "Exactitude numerique" → 681+667). Les Walton schemes sont coherents avec le type d'opposition. Ready pour jsboige validation.