Skip to content

data(scenarii): EN cascade pass gpt-5.5 — 8 'issue' field edits (smoke) - #373

Merged
jsboige merged 1 commit into
masterfrom
data/scenarii-cascade-en-smoke
May 28, 2026
Merged

data(scenarii): EN cascade pass gpt-5.5 — 8 'issue' field edits (smoke)#373
jsboige merged 1 commit into
masterfrom
data/scenarii-cascade-en-smoke

Conversation

@jsboige

@jsboige jsboige commented May 28, 2026

Copy link
Copy Markdown
Contributor

Summary

Second smoke test of the new cascade EN-only pattern (PR #371). Validates the approach on Scenarii (flat PKs, ChunkSize=12).

Parameters: ChunkSize=12, TakeChunkNb=1, processed scenarios 1.1.1 – 1.2.5.
Cost: ~$0.30, 19s, 0 errors.

Changes (8 surgical UpdateRecord on issue field only)

Scenario path Type of fix
Caesar & Cleopatra 1.1.1 "the Queen" → "the queen" + missing period
Caligula 1.1.3 "in the lions" → "to the lions" + article
Trojan Horse 1.1.5 Replaced context-summary EN with proper issue narrative matching FR
Joan of Arc 1.2.1 Lazy "Try to convince him..." → formal "She must convince him to entrust her with an army to liberate Orléans..."
Slavery 1.2.2 "Try to convince..." → "The smooth talker must convince..."
Crusades 1.2.3 "Trying to talk him out of it" → "The smooth talker must try to dissuade him."
Women's vote 1.2.4 "Try to justify..." → "The smooth talker must justify..."
Socrates trial 1.2.5 "Try to convince him..." → "The smooth talker must convince the tribunal..."

Validation

  • Field scope: only issue edited (out of 4 available in FieldsToUpdate: title, context, issue, suggestion_en)
  • The LLM honored the prompt-level priority enjeu/issue > contexte/context > suggestion > titre/title
  • Other-lang columns untouched (verified via column-level diff in commit message)
  • Style aligned with refined FR (PR data(scenarii): FR clarity pass gpt-5.5 — 162/167 records #368)
  • BOM: Scenarii has no BOM convention → preserved as-is

Test plan

  • Smoke ran cleanly, 8 UpdateRecord, 0 errors
  • Non-EN columns untouched (column-level diff)
  • EN style alignment matches FR clarity pass
  • Visual inspection of generated EN Scenarii cards (jsboige to schedule)

🤖 Generated with Claude Code

…2 records)

Smoke test of "Scenarii cascade EN-only gpt-5.5" task, validating the
max-context narrow-edit approach (FR readonly anchor, EN editable target).

ChunkSize=12, TakeChunkNb=1 → processed scenarios 1.1.1 through 1.2.5,
LLM made 8 surgical edits on the `issue` field (EN), aligning style and
structure to the refined FR (PR #368):

- 1.1.1 (Caesar/Cleopatra): article casing + missing period
- 1.1.3 (Caligula/lions): preposition + article ("in the lions" → "to the lions")
- 1.1.5 (Trojan Horse): replaced context-summary EN with proper issue narrative
- 1.2.1-1.2.5 (5 records): consistent third-person "The smooth talker must..."
  formal voice mirroring FR, replacing lazy second-person "Try to..." stubs

API cost: ~$0.30, duration 19s, 0 errors, 0 quota issues.
Field scope verified: only `issue` column edited (4 EN fields in FieldsToUpdate
were available: title/context/issue/suggestion_en; LLM chose issue exclusively
per the prompt priority order).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant