feat(data): DatasetUpdater campaign — complete Scenarii RU/PT + quality passes

[jsboige]

Objective

Complete multilingual translation coverage for all CSV datasets to unblock v0.9.0 release.

Current State

Dataset	EN	RU	PT	Blocker
Scenarii (167 rows)	✅ complete	❌ 76 empty (~46%)	❌ 76 empty (~46%)	OpenAI quota
Fallacies (1408 rows)	✅ core done	✅ core done	✅ core done	taxonomy gaps only
Virtues (223 rows)	✅ 100%	✅ 100%	✅ 100%	taxonomy gaps only
Rules (24 rows)	✅ 100%	✅ 100%	✅ 100%	none

Plan (validated by coordinator ai-01, approved by jsboige)

P1 — Scenarii RU/PT (bloquant v0.9.0)

• 76 rows × 6 columns per language (title, smoothTalker, drawer, context, issue, suggestion)
• Model: gpt-5.4-mini (or gpt-4.1 as fallback)
• ChunkSize=8, TakeChunkNb=-1, WriteOneTargetFileByField=false
• 1 PR per language, cell-by-cell verification
• BLOCKED: HTTP 429 insufficient_quota on OpenAI API key

P2 — Fallacies taxonomy gaps

• EN: 2 desc_en, 1 example_en, 50-77 taxonomy hierarchy fields
• RU/PT: 6-40 taxonomy fields
• link_ru (1382) / link_pt (1406) = HORS scope (Wikipedia URLs)

P3 — Virtues taxonomy gaps

• subfamily/subsubfamily: 8-30 per language
• links: 27-37 per language

P4 — Quality passes

• Review and improve existing translations
• PR per dataset with coordinator review

Glossary (validated by jsboige)

• PT: "O Embromador" (smooth talker) / "O Puxador" (drawer)
• RU: "Болтун" (already OK)

Config Reminders

• Always disable "Cleanup Fallacies" (Enabled=false) before any other task
• dotnet clean before rebuild
• Copy .keys/ to bin/Debug/net9.0-windows/.keys/
• Restore Mode and Enabled flags after each run

[jsboige]

Status Update — 2026-05-24

P1 — Scenarii RU/PT ✅ COMPLETED

• RU 163/167 via PR data(scenarii): translate 76 RU rows via DatasetUpdater #336, final 4 via PR fix(scenarii): complete RU row 7.2.2 surgically — RU 100% (supersedes #341) #342 (surgical)
• PT 100% via PR data(scenarii): translate 76 PT rows via gpt-5.5 #338, residuals via PR fix(scenarii): PT contamination fix + RU residual translations #339
• 0 contamination verified

P2 — Fallacies taxonomy ✅ COMPLETED

• 2 desc_en filled (PK 371, 607) via PR fix(fallacies): fill 2 empty desc_en (PK 371, 607) surgically — supersedes #343 #344
• Wikipedia links curated: 585 interlanguage links (PR data(fallacies): curate Wikipedia interlanguage links (#333) — rebased #352, pending merge)

P3 — Virtues ✅ CONFIRMED NO-OP

• 0 gaps found in cell-by-cell diagnostic

P4 — Quality passes 🔄 IN PROGRESS

• FR simplification pass 1: 27 desc_fr improved (PR data(fallacies): simplify 27 desc_fr for younger audience #351, pending merge)
• FR simplification pass 2: running (gpt-5.4 on pass 1 base)

Data-Gate v0.9.0 = 100%

All datasets verified, release blocked only on visual inspection #303.

[jsboige]

Phase 4 Campaign Status — COMPLETE (2026-05-29)

Summary

Phase	Description	Status
P1	Scenarii RU+PT 167/167	✅ DONE (PRs #322/#330/#336/#338/#339/#342)
P2	Fallacies 2 desc_en	✅ DONE (PR #344)
P3	Virtues taxonomy gaps	✅ NO-OP (0 gaps confirmed)
P4	Quality passes	✅ DONE (FR clarity #366-#369 + cascade EN-only #376-#380)

Phase 4 cascade EN-only wave summary

Dataset	scale-3	scale-6	Status
Rules (24 rows)	full coverage	n/a	✅ #376
Scenarii (167 rows)	28 cells	49 cells	✅ #377, #380 — FP running
Virtues (223 rows)	16 cells	0 CONVERGED	✅ #378
Fallacies (1408 rows)	16 cells	0 CONVERGED	✅ #379

Cascade pattern reached natural convergence on 3/4 datasets at scale-6. Only Scenarii continues finding narrative improvements; currently testing TakeChunkNb=-1 full-pass.

Closing this campaign issue — data gates met for v0.9.0.