feat(config): cascade context-split + EN-only tasks (max-context approach) - #371
Merged
Conversation
…oach) Refactor cascade drift tasks to give max hierarchical context to LLM while narrowing edits: **Existing cascades — renamed to multi-lang variant:** - Move EN fields from FieldsToUpdate to readonly context - Now LLM sees FR (source clean) + EN (clean) as anchors for the 6 other langs - Editable: RU/PT/ES/AR/FA/ZH only - ChunkSize bumps: Rules 4->8, Virtues 4->6 (richer hierarchical grappes) - Fix pre-existing typo: Scenarii cascade had `suggestion` (FR) duplicated in EN row; corrected to `suggestion_en` per Scenario.cs:30/83 mapping **New EN-only cascades (4 tasks):** - Rules/Virtues/Scenarii/Fallacies cascade EN-only - Context: FR fields only (clean source) - Editable: EN fields only - Smallest-context, fastest-converging variant for calibration smoke tests **Rationale (jsboige directive):** Give max context to LLM (FR clean + EN clean), keep only target languages editable. Larger ChunkSize for richer hierarchical grappes lets the model arbitrate between siblings. **Total cascade tasks now: 8** (4 multi-lang + 4 EN-only) **All Enabled = false** (smoke-test on demand, $8/PR budget guard-rail) **Tests:** 120 pass / 0 fail / 5 skip (GDrive/GUI), build 0 errors Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
4 tasks
jsboige
added a commit
that referenced
this pull request
May 28, 2026
…ules) (#372) Smoke test of "Rules cascade EN-only gpt-5.5" task introduced in PR #371, validating the max-context narrow-edit approach (FR readonly anchor, EN editable target). ChunkSize=8, TakeChunkNb=1 → processed rules 1-8, LLM made 7 surgical edits on rule "L'École des Menteurs" (rule 1 of the deck), aligning EN to the refined FR (PR #366): - Components/Game overview/Setup/Round sequence/The count: section headers rephrased to match FR clarity (was: Game material/Game summary/Installation/ Course of a round/The points count) - "smooth talker" → "embromador" (proper game term, consistent with FR) - Added missing emoji codes in win/tie conditions (🥇👇 🥈👇 ✅🎭 etc.) - Filled in 8 missing "Draws" subsection bullets (winner draws 0, jurors draw 1, embromador's-card-finders draw 1 extra...) - Filled in 6 missing variant bullets (skit interventions, drawer-as-winner, multi-card embromador, final-scenario rule) - Title alignment: "The school of liars" → "Argumentative mixology bingo" (matches FR "Le Bingo mixologie argumentative" — was a leftover from a prior rule variant) API cost: ~$0.50, duration 36s, 0 errors, 0 quota issues. Validates EN-only cascade pattern for follow-up smoke tests on Scenarii/ Virtues/Fallacies. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This was referenced May 28, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Refactor cascade drift tasks to give maximum hierarchical context to the LLM while narrowing edits — implementation of jsboige's directive: "include FR (clean) + EN (clean) as readonly anchors when editing the other 6 languages".
What changed
4 existing cascades renamed to multi-lang variant (Rules / Virtues / Scenarii / Fallacies):
text_en/desc_en/example_en(or equivalent) fromFieldsToUpdateto readonly contextsuggestion(FR field) duplicated in EN row; corrected tosuggestion_enperScenario.cs:30/83mapping4 NEW EN-only cascades:
Total cascade tasks after PR: 8 (4 multi-lang + 4 EN-only).
All
Enabled = false— smoke-test on demand, $8/PR budget guard-rail.Test plan
dotnet build "Argumentum Converters.sln"→ 0 errors, 20 warnings (all pre-existing)dotnet test→ 120 pass / 0 fail / 5 skip (GDrive/GUI), unchanged from masterWhy max-context matters
With FR + EN both visible in the prompt, the LLM has two anchors of "what the cell should mean" before deciding whether each translation has drifted. Solo-FR-anchor (the prior approach) gives less leverage for arbitrating between candidate edits.
🤖 Generated with Claude Code