feat(csv): #499 Phase 1 — add 12 relational/AIF columns to Virtues prod CSV

[jsboige]

What

#499 Phase 1 — prod-write: append 12 relational/AIF columns to the Virtues prod CSV (Cards/Fallacies/Argumentum Virtues - Taxonomy.csv, 66 → 78 cols), populated from the 12 validated annotation CSVs in docs/taxonomy/ (joined by virtue_pk == pk).

De-gated by jsboige (interactive 2026-06-23, "GO" on docs/taxonomy/499-virtues-parity-closure.md); dispatched by ai-01 as Phase 1 of the #499 deep-queue. Data-only (the OWL code propagation = Phase 2, separate PR).

Ground-truth re-verified on current master f432455e (spec #580 claims held exactly)

Measure	Value
Prod Virtues rows / distinct pk / cols	223 / 223 / 66
Annotation CSVs / total rows / distinct virtue_pk / dups	12 / 222 / 222 / 0
In-annotations not in prod	0
In-prod not annotated	1 (pk=0 taxonomy root — correctly left out)
Target cols populated (crossLink_Opposes / skosDirectRef / skosMappingType)	222 / 222 / 222

The 12 columns (verbatim Fallacies names, appended after link_fa col 66)

crossLink_PredatesOn, crossLink_Denounces, crossLink_Leverages, crossLink_Allows,
crossLink_Opposes, crossLink_Inverts, crossLink_Mirrors, crossLink_IsRelatedTo,
AIF_skosDirectRef, AIF_skosExceptionRef, AIF_skosOther, AIF_skosMappingType

• 3 populated for 222 rows: crossLink_Opposes (Fallacy PK semicolon-list), AIF_skosDirectRef (Walton scheme label), AIF_skosMappingType (FR prose critical question).
• 9 structurally empty by design: a Virtue's only cross-taxon link is "Opposes" its prevented Fallacy family; Exceptions/Other are Fallacy attributes.
• pk=0 (root) left empty.

Additive proof (the critical invariant)

Existing 66 columns are byte-identical before/after — 0 value drift across 223 × 66 = 14 518 cells. The write is a line-append (12 new cells per line); existing content untouched. Guard: the script aborts if raw CRLF-line count ≠ 223 (embedded-newline safety).

Verification

• Additive: 0 existing-cell drift ✓ | first-66 col names unchanged ✓
• 66 → 78 cols, 12 new cols in spec order at positions 67–78 ✓
• 3 target cols filled 222/222; 9 structural empties = 0 non-empty; pk=0 empty ✓
• Cross-link integrity (tools(taxonomy): reusable anti-fab validator for scale-up CSVs #518-style): 60 distinct Fallacy PKs in crossLink_Opposes → 0 dangling (all resolve in the Fallacies corpus PK column) ✓
• UTF-8 no-BOM, CRLF preserved, 998061 bytes (+30898) ✓
• Spot-check pk=71 (the "Valid results" Virtue from fix(csv): #202 translate 9 RU Virtues cells 'Valid results' → Корректные результаты #583): crossLink_Opposes=[681;690], AIF_skosDirectRef=[Argument from Sign], AIF_skosMappingType=[Les résultats sont-ils correctement calculés, reproductibles et cohérents… ?] (FR prose comma correctly quoted) ✓
• git diff --stat: 224 ins / 224 del (every line gets 12 cols appended; existing cells unchanged within each line) ✓
• Unit tests: 533 passed / 0 failed / 5 skipped (baseline unchanged — unmapped cols ignored by CsvHelper on read) ✓

Phase 2 (separate code PR — NOT this one)

ArgumentVirtueClassMap mapping of the 12 new columns + OwlAdapter Virtues pass (currently Fallacies-only). No code changed in Phase 1.

🤖 Worker po-2024 · #499 Phase 1 prod-write · 12 additive cols (3 populated ×222) · additive-proof 0 drift · crossLink 0 dangling · 533 tests pass · Phase 2 (OWL code) next

[clusterManager-Myia]

[NanoClaw] — data-integrity review (RFC4180 parse of all 223 rows, old-master vs head blob)

✅ PASS — no data loss.

• Schema 66 → 78 cols; every one of the 223 data rows parses to exactly 78 fields (no ragged rows).
• The existing 66 columns are byte-identical across all rows (verified field-by-field). No unintended edits, no truncation.
• Header prefix preserved; 12 new cols: crossLink_{PredatesOn,Denounces,Leverages,Allows,Opposes,Inverts,Mirrors,IsRelatedTo}, AIF_skos{DirectRef,ExceptionRef,Other}, AIF_skosMappingType.
• Join integrity matches the PR's ground-truth table: populated cols show 222/223 non-empty = the 222 annotation rows; the single orphan prod row (no matching virtue_pk) is correctly left empty.
• No secrets / no control-char corruption in the new blob.

⚠️ One to confirm (non-blocking): only 3/12 new cols carry data — crossLink_Opposes (222, 147 distinct), AIF_skosDirectRef (222, 14 distinct), AIF_skosMappingType (222, 222 distinct). The other 9 are empty across all rows. Given Phase 2 = OWL/crossLink propagation (separate PR) the empty crossLink_* set is plausibly expected — worth an explicit 👀 that these are "source annotations had no data for those relations," not a silent join gap. If only those 3 had source data today, the change is internally consistent.

No merge-blocking issues. Phase-1 prod-write looks structurally sound.