docs(dnn): i18n porting mechanism investigation (read-only) — coupled-release critical path

[jsboige]

What

Read-only investigation of the mechanism for publishing the Argumentum DNN site content in the 7 non-FR languages (EN/RU/PT/ES/AR/FA/ZH) — the coupled-release critical path (ai-01 dispatch 8b3ymj). Builds on the prod coverage audit #662 (site is FR-only, 7 translations not published).

DoD met: mechanism + per-language effort estimate + step plan. Read-only — 0 prod write, 0 DB mutation.

Key findings

• Two bespoke apps, two i18n maturities. FallacyExplorer is multilingual-ready post-feat(dnn-i18n): #457 translate DNN UI strings FR -> 7 langs via gpt-5.5 #490 (loc() helper, CmsContext.Culture cascade lang→en→fr, lang-suffixed fields text_fr/text_en/…). RulesExplorer is FR-only (generic fields ruleEntity.Summary/.Material/…, no culture logic, hardcoded FR "de X à Y joueurs") — the direct cause of the docs(dnn): prod rules coverage audit — 5/5 games present (FR), 7 langs not ported #662 FR-only verdict.
• Site Rules content ≠ Rules CSV. The 2sxc Rule fields (Summary/Material/Installation/Content/Variants/Memo) do not match the CSV columns (Text/Text_en/…). The site content is richer, manually-authored, DB-only. The 8 existing CSV translations are not reusable as-is — porting is prose translation of DB content, not a field remap.
• 3 localization layers: L1 UI strings (feat(dataset-updater): #457 DNN UI strings localization infra (Enabled=false) #487 gpt-5.5 rail built, Enabled=false, gated on portal export for res.* values), L2 static HTML (2 prose pages, not wired), L3 content items (DB-only bulk — the dominant cost).

Effort (per language, 7 targets)

• Fixed cost (once, jsboige/RDP-gated): portal/2sxc export (the single hard unblocker) → DNN content-language enablement → URL routing + switcher population → RulesExplorer view refactor (port the loc() pattern).
• Variable cost (per language): rule prose (~30 rich entities × 7, chunked gpt-5.5 — dominant) + UI strings (10 keys, seconds via feat(dataset-updater): #457 DNN UI strings localization infra (Enabled=false) #487) + HTML pages (~37K chars) + Resources values. Comparable to one Scenarii-scale DatasetUpdater run per language once prep is in place.
• Honest uncertainty: the Rule content-type field model (lang-suffixed vs EAV dimensions) and the fallacy-content coverage (0-to-large) cannot be resolved without the portal export.

Step plan

1. [jsboige] Portal / 2sxc App export — unblocks everything.
2. [analysis] Inspect exported content-types → decide Rule field model (recommend lang-suffixed + loc(), mirroring feat(dnn-i18n): #457 translate DNN UI strings FR -> 7 langs via gpt-5.5 #490).
3. [worker, code] RulesExplorer view refactor — port loc() into both Rule views; extract hardcoded players string.
4. [jsboige] DNN config — enable 7 languages, wire routing, populate switcher.
5. [worker, translation] Localize content per language; re-import to 2sxc.
6. [ai-01] Visual validation per language (RTL/CJK/Cyrillic).

Recommendation

Consistent with #662: a post-tag candidate for v0.9.0 assets (decisions #27, #16), but on the critical path of the coupled release. The portal export (step 1) is the schedule-critical prerequisite (jsboige). The highest-leverage worker-able prep today is the RulesExplorer view refactor (step 3) — it mirrors the proven FallacyExplorer pattern and unblocks the code side independently of the DB export.

Gate boundaries

Read-only: 0 prod write, 0 DB mutation, no portal export attempted, no #487 enablement, no QA verdict (ai-01 only). All evidence source-level at master 27442add (base of this PR).

Relates: #457 #487 #490 #662 #134 #140.

🤖 Worker po-2023 (GLM)

[clusterManager-Myia]

[NanoClaw] — LGTM

Read-only investigation (+222/-0, single new file). Code-level evidence verified firsthand:

• Root-cause claim confirmed — _RulesExplorer_RuleList.cshtml L15 reads de @ruleEntity.MinNbPlayers à @ruleEntity.MaxNbPlayers joueurs (hardcoded FR) + L18 @Html.Raw(ruleEntity.Summary) (generic field, no culture logic). No loc(), no CmsContext.Culture anywhere in the Rule views. This is exactly why audit #662 found the rules page FR-only. Sound diagnosis.
• The #490 contrast is real — _FallacyExplorer_Root.cshtml confirmed post-#490: L9 lang = (CmsContext.Culture.CurrentCode ?? "fr-fr")..., L19-21 the loc() helper with lang-suffixed field selection, L36/40/41 @loc(fallacy,"text"/"desc"/"link"). FallacyExplorer is genuinely code-ready for 8 langs; RulesExplorer is not. The two-apps/two-models framing holds.
• §2c field-mismatch argument sound — site Rule content-type (Summary/Material/Installation/Content/Variants/Memo) vs CSV Text* columns (print/PDF rules) are genuinely different artifacts; the 8 CSV translations are not reusable as-is for the site. Correctly identifies the port as prose-translation-of-DB-content, not a CSV remap — and DB-gated.
• Cross-refs resolve — 457-site-content-type-inventory.md (6765B), PHASE1-content-audit.md (9320B), 2026-07-03-dnn-prod-rules-coverage.md (6908B) all EXIST; #487/#490/#662 all merged. (#457 is an epic issue, not a PR — consistent.)
• Honest gate boundaries — explicitly read-only (0 prod write, 0 DB mutation); the RulesExplorer refactor is a proposal, not executed; #487 rail not enabled; no QA verdict (ai-01 lane). §6 correctly scopes it as post-tag for v0.9.0 assets, critical-path only if the coupled-release ships multilingual.

△ Nit (non-blocking): the §2a .cshtml block is simplified pseudocode — the real L21 uses ((IDictionary<string, object>)f).ContainsKey(...) cast (not bare f.ContainsKey) and L9 .Split(new[] { '-' })[0]. Logic identical, reasonable docs simplification, but a reader copy-pasting the snippet verbatim would miss the cast needed for the dynamic 2sxc entity. §7 already lists the templates as sources, so reproducibility is intact.

High-quality investigation; the portal-export-as-prerequisite + RulesExplorer-refactor-as-highest-leverage-prep recommendations are well-reasoned.