Adds a tenth language to all 1,324 exercises: `instructions.fr` and
`instruction_steps.fr`, plus the schema, README, viewer and setup guide.
Register matches the existing Spanish/Italian/Polish translations —
informal second-person imperative (tutoiement):
en: Lie flat on your back with your knees bent and feet flat on the ground.
fr: Allonge-toi sur le dos, les genoux fléchis et les pieds à plat au sol.
How it was produced
-------------------
The 7,710 English step sentences deduplicate to 4,414 unique strings.
Those were translated (LLM-assisted, glossary-constrained, then audited),
and the per-exercise arrays were rebuilt deterministically by mapping each
English step back through that table. `instructions.fr` is the join of
`instruction_steps.fr`, so the two can never drift.
A fixed glossary pinned the recurring terms — répétitions, position de
départ, haltère/barre, sangle abdominale, omoplates, largeur des épaules —
so terminology is consistent across all 1,324 records rather than varying
sentence by sentence.
Verified
--------
- 1,324 records validate against the updated schema; a record missing `fr`
is now rejected.
- fr step count == en step count in every record (the dataset's existing
invariant across all nine prior languages).
- instructions.fr == ' '.join(instruction_steps.fr) in every record.
- Every pre-existing field is byte-identical; `fr` is purely additive.
- No empty strings, none left in English, no vouvoiement across 7,710 steps.
- index.html embed stays in sync with data/exercises.json; the language
picker renders "Français" and the steps display correctly.
Note for review: these are machine translations, reviewed programmatically
rather than by a native speaker. Happy to adjust any wording.
Incorporates the translation content from two external PRs onto current main
(their branches were based on a pre-media-restore history with no common
ancestor, so applied programmatically rather than merged):
- Polish (pl) from #25 (Dawid Matyjasik) — as submitted.
- Korean (ko) from #27 (SeungTaek Lim) — normalized to the dataset convention:
only nested instructions.ko / instruction_steps.ko were kept. The PR's extra
flat top-level *_ko metadata fields (name_ko, target_ko, etc.) were dropped
because they are inconsistent with every other language and would violate
data/exercises.schema.json (additionalProperties: false).
Updates data/exercises.json (all 1,324 records), the embedded EXERCISES blob
and SQL export in index.html, the SQL schemas / INSERT builder / LLM prompt in
setup.html, and README language references (7 -> 9 languages). Change is purely
additive: stripping pl+ko reproduces the previous data byte-for-byte.
Co-Authored-By: Dawid Matyjasik <63721806+dawidmatyjasik@users.noreply.github.com>
Co-Authored-By: SeungTaek Lim <61047602+namest504@users.noreply.github.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
umutcakirai