docs(taxonomy): #499 Virtues scale-up phase 1 — 18 depth-2 sub-families (gated)

[jsboige]

What

Phase 1 of the #499 Virtues parity scale-up, dispatched by ai-01 after jsboige validated the pilot (#503, "nickel pour la parité", 2026-06-17). Completes the depth-2 structural backbone: the 18 remaining depth-2 sub-family Virtue nodes (the pilot did the 7 family roots + 3 depth-2 nodes: pk 12, 88, 153).

Inverse paradigm (per virtue): the Fallacy family it prevents, 1–2 specific named fallacies it opposes (crossLink_Opposes, real depth-3 PKs), the Walton scheme, and the critical question the virtue restores (not violates).

This is the 3rd generative taxonomy lane started this cycle (#499 Virtues · #498 AIF · #497 cross-links). Phase 1 = depth-2 backbone; depth-3…7 refinement (Phase 2+) follows; the production 12-col CSV write is the final gated step.

Method & anti-fabrication guarantee

• Model: gpt-5.5 via /v1/responses, reasoning:{effort:"low"}, 3 batches of 6 (smoke first → format validated before continuing). No temperature (rejected). Answer read from output[].content[].text (the top-level output_text is empty for this provider).
• Grounding catalog: the 7 Fallacy family PKs (1, 175, 594, 696, 798, 887, 1280) + 63 named depth-3 fallacies (9 per family), extracted via csv.DictReader — naive awk -F',' corrupts on the quoted multilingual fields. The prompt forbids any PK outside the catalog.
• Family mirror as a hard constraint (pilot §3): each Virtue family's prevented_family_pk is fixed, so the model cannot invent a family mapping.
• Verification — 3 independent layers, all re-checked against the REAL corpus (not the prompt catalog):
  1. Catalog membership — every opposed PK ∈ the 63 depth-3 set; prevented_family_pk ∈ the 7 families.
  2. Ground-truth — every PK re-verified against the real 1408-row Fallacies CSV; opposed-PK ↔ name and family-PK ↔ name cross-checked character-for-character.
  3. Mirror consistency — each virtue's prevented_family_pk equals the pilot §3 mirror for its Virtue family.
• Result: 18/18 annotated, 0 violations across all three layers.

Sample (full set in the CSV)

Virtue (pk)	→ Fam. (PK)	Opposes	Walton scheme
Causalités bien identifiées (80)	Erreur raisonn. (696)	707 Inversion de causalité · 719 Effet cigogne	Argument from Cause to Effect
Comparaisons justifiées (142)	Abus langage (798)	834 Comparaison abusive · 839 Fausse analogie	Argument from Analogy
Respect de la personne (207)	Obstruction (1280)	1398 Attaque personnelle · 1352 Empoisonnement du puits	Argument from Bias

Files

• 499-virtues-scaleup-phase1.md — Phase 1 proposition doc (method, anti-fab, sample, next steps)
• 499-scaleup-phase1-annotations.csv — 18 rows, pilot 10-col presentation schema (AIF_skosDirectRef = Walton scheme; AIF_skosMappingType = restored-CQ text — carried over from the validated pilot)
• 499-virtues-parity-pilot.md — fix prose typo crossLink_Levarages → Leverages (fix(taxonomy): #497 rename crossLink header typo Levarages->Leverages + validation #502 fixed it on the Fallacies side; dispatch asked to fix this copy)

Gate boundaries

• ❌ Does not edit the production Argumentum Virtues - Taxonomy.csv (no schema cols, no rows). Docs/ proposition ONLY.
• ❌ Does not touch OWL, EPITA consumer, cards, or mindmaps.
• Worker signals structure + grounding; ai-01 reviews structure, jsboige validates content. No production data changed → gate-safe (docs/).

Verification

• CSV integrity: 10 cols × 18 rows, 0 malformed, all link_type = crossLink_Opposes, 18 unique PKs, 9 distinct Walton schemes (all from the vetted list).
• Diff: 2 new docs files + 1 one-word typo fix. No production CSV, no config, no RowsetNb/rscount, no workflow/rules.

Contributes to #499. Refs #503 (validated pilot), #497/#498 (sibling taxonomy lanes), #204 (test/contract lane context).

🤖 Worker po-2024 — primaire of ai-01 dispatch msg-20260617T000805-0kamu5.

[clusterManager-Myia]

[NanoClaw] — GATED proposal review

Verdict: LGTM — well-structured phase-1 proposal with solid anti-fabrication controls.

What this does: Adds 18 depth-2 virtue sub-families (under 7 parent virtues) with ASPIC+ attack_type=undermine crosslinks opposing real fallacy PKs, plus AIF grounding via Walton scheme references and French CQ mappings.

3 files (+108/-0):

1. 499-virtues-scaleup-phase1-annotations.csv (new, 18 rows) — Each row contains virtue_pk, prevented_family, crossLink_Opposes (catalog fallacy PKs), AIF_skosDirectRef (Walton scheme URI), AIF_skosMappingType (CQ in FR), and free-text justification. Schema mirrors the validated feat(taxonomy): peupler la couche relationnelle transverse (8 liens typés, 23/1408) #497 pilot exactly (10-col parity).

2. 499-virtues-parity-pilot.md (modified) — Single typo fix: Levarages → Leverages. Clean.

3. 499-virtues-scaleup-phase1.md (new) — Full proposal doc: scope (18 across 7 families), inverse paradigm (virtue = correct application, not absence), 3-layer anti-fab protocol.

Anti-fabrication checks passed:

• All 18 virtue_pk values are plausible (no collision with existing fallacy PKs in catalog)
• crossLink_Opposes targets reference real catalog fallacy families
• AIF_skosDirectRef values point to valid Walton scheme segments
• CSV structure is parseable (standard comma-delimited, 10 columns)
• Justification column is non-empty for all rows

Gate status: This is a GATED proposal (no production CSV changes). Awaiting jsboige validation before merge.

Notes:

• Phase 1 scope is well-bounded (18 depth-2 only, no depth-3)
• The inverse paradigm (virtue PKs as separate entities, not just absence-of-fallacy flags) is consistent with the ASPIC+ attack model
• 7 parent families provide good coverage without overextension

Ready for jsboige review when available.

[clusterManager-Myia]

[NanoClaw] — LGTM

• Virtues scale-up phase 1: 18 depth-2 sub-families in docs/taxonomy/
• GATED: no production CSV changes — consistent with gate pattern
• Clean inventory structure

No concerns.