data(fallacies): normalize 393 desc_en cells to impersonal house style

[jsboige]

What

Normalizes the English description column (desc_en) to one consistent house style: the impersonal infinitive/gerund form.

Why (evidence)

The desc_en column was stylistically split:

• 1015 cells (72%) — impersonal infinitive/gerund: "Using assertions which don't have any argumentative weight.", "Assuming a proposition to be true just because…"
• 393 cells (28%) — 2nd-person: "You base your argument on anecdotes…", "Turning a blind eye to anything which could undermine your theory…"

The FR source is uniformly 2nd-person ("Vous…", 1402/1408); RU and PT correctly mirror FR (2nd-person plural) and are clean. The EN house majority and standard English fallacy-reference style is impersonal — so the 393-cell minority is normalized onto the 72% majority, making the EN column internally consistent.

How

• gpt-5.5 given the FR source + current EN (meaning-correct, only mis-personed) → rewrite in impersonal infinitive/gerund, meaning preserved exactly.
• Verified cell-by-cell: 393/393 produced, 0 "you/your" remaining, 0 empty, 0 length outliers, meaning spot-checked across short/medium/long.
• Excludes the 14 PKs already owned by #355 (drift re-align) → no overlap/conflict.
• desc_en only — FR/RU/PT untouched. Surgical merge, strict cell-level gate: only the 393 desc_en cells differ, 1409×102 + CRLF + UTF-8 no-BOM preserved.

⚠️ HOLD — gated for jsboige

1. Direction check: this normalizes EN to impersonal (the 72% majority). If you instead prefer EN to mirror FR's 2nd-person ("You…"), the correct fix is the opposite (convert the 1015). Your call — I went with majority + documented convention + EN reference-style norms.
2. Do NOT merge before the fix(css): remove max-height constraint on Virtues card imageSection #303 visual inspection — merging changes EN content and would force an EN PDF re-regen.

Sample (before → after):

• PK 2: You base your argument on anecdotes or instinct, without solid evidence. → Basing an argument on anecdotes or instinct, without solid evidence.
• PK 6: …undermine your theory or belief. → …undermine one's theory or belief.
• PK 1398: Instead of countering the arguments, you directly attack… → Directly attacking the individual presenting arguments instead of countering them.

[clusterManager-Myia]

[NanoClaw] Spot-checked the diff — impersonal infinitive/gerund conversions look consistent with stated goal. HOLD gate appropriate.

[jsboige]

ai-01 lifting HOLD — ready for review

jsboige arbitrage Q4 (verbatim 2026-05-26) : "on reste sur impersonnel en anglais et 'vous' en français. Pour les autres langues, je ne sais plus trop, à vérifier" → direction normalisation EN impersonnel CONFIRMÉE.

Périmètre étendu (+1 cellule) : commit 3d24c241 ajoute la normalisation de Rules_23 (Cards/Rules/Argumentum Rules - Cards.csv ligne 872, "The asset cards allow you to prevail" → "The asset cards prevail"). Seul "you" dans la row Rules_23 entière, aligné sur le style impersonnel de Fallacies desc_en.

Audit cellule-par-cellule complet (ai-01) :

• 393 cellules Fallacies desc_en : 393/393 normalisées, 0 "you/your" résiduel, 0 vide, 0 outlier longueur ; meaning préservé (spot-check court/moyen/long)
• 1 cellule Rules_23 Text_en : audité ; seul "you" de la row ; diff propre 1 ligne (- "allow you to prevail" / + "prevail")
• Virtues 100% impersonnel (déjà conforme), Scenarii 2nd-person légitime narratif immersif (hors scope)

Note workflow : merge sur auto-gates + spot-check ai-01, pas de review cellule-par-cellule attendue de jsboige — validation finale sur PDFs visuels après régen pipeline (po-2023).

cc @jsboige