Skip to content

Open-source gym & body-weight tracker, now in the App Store (iPhone + Apple Watch) and as an Android APK. Plan routines, log workouts, see which muscles are trained, import from Strong/Hevy/FitNotes. Self-host it or use it on the phone alone.

Merge #266: Russian exercise names (the third name pack) (dima-po)

Duarte Santos683383fcda29
diff --cc frontend/src/lib/i18n-core.js
index d72c5b26f,5ce312858..dc29c51c8
--- a/frontend/src/lib/i18n-core.js
+++ b/frontend/src/lib/i18n-core.js
@@@ -10,7 -10,7 +10,7 @@@ export const LANGS =
ko: '한국어', hi: 'हिन्दी', th: 'ไทย', hu: 'Magyar'
}
export const INSTR_LANGS = ['en', 'es', 'fr', 'it', 'tr', 'ru', 'zh', 'hi', 'pl', 'ko', 'pt-BR', 'hu']
- export const EXERCISE_NAME_LANGS = ['pt-BR', 'hu', 'de', 'es']
-export const EXERCISE_NAME_LANGS = ['pt-BR', 'hu', 'ru']
++export const EXERCISE_NAME_LANGS = ['pt-BR', 'hu', 'de', 'es', 'ru']
export const DATE_LOCALES = {
en: 'en-GB', de: 'de-DE', 'de-CH': 'de-CH', es: 'es-ES', fr: 'fr-FR', it: 'it-IT',
pt: 'pt-PT', 'pt-BR': 'pt-BR',
diff --cc scripts/exercise-name-sources/README.md
index 25960ad7f,0089bc9ca..2e7709b02
--- a/scripts/exercise-name-sources/README.md
+++ b/scripts/exercise-name-sources/README.md
@@@ -22,61 -22,20 +22,81 @@@ assistance and must not be described a
named human reviewer completes that review. They are original translations and
were not copied from another Portuguese exercise dataset.
+---
+
+ # Russian exercise names
+
+ `ru.json` is the same thing for Russian, and `GLOSSARY.ru.md` beside it fixes
+ the terminology — which movements keep an English loanword a Russian lifter
+ actually says (`бёрпи`, `машина Смита`, `EZ-гриф`) and which get a Russian
+ term, so a batch translated next month matches one translated today.
+
+ ```sh
+ node scripts/translate-ru-exercise-names.mjs --apply # fills in what is missing
+ node scripts/build-ru-exercise-names.mjs # regenerates the runtime pack
+ ```
+
+ The translation script only visits IDs the source does not already have, so it
+ is safe to re-run after the catalogue grows, and it checkpoints after every
+ batch. Same caveat as above: these were produced with LLM assistance from the
+ English titles and must not be described as reviewed by a native speaker unless
+ a named human reviewer completes that review.
++
++---
++
+# German exercise names
+
+`de.json` is the editable source for the German exercise-name pack, and it
+covers a stage rather than the whole catalogue: 760 of the 1,324 built-in
+exercises. Left out are the 325 whose equipment is `body weight`, and 239
+equipment exercises whose German name is written but not yet signed off by a
+native speaker (`AWAITING_REVIEW` in `scripts/de-name-rules.mjs` lists them,
+with the reason). Both groups keep their English title — the app falls back
+per exercise, so a partial pack costs nothing beyond the names it does not
+yet carry. Which exercises the stage contains is enforced, not approximate:
+`scripts/de-name-rules.mjs` defines the set, and the builder refuses a source
+that misses one of them or adds an entry from outside it.
+
+Clearing a name from `AWAITING_REVIEW` is what a second German-speaking
+reviewer does: delete its id there, put its name in `de.json`, rebuild. The
+test fails if the two ever disagree, so the list cannot quietly rot.
+
+`de-CH` ships no pack of its own. It derives this one by replacing ß with ss,
+so German names are written with ß here and the Swiss spelling follows
+automatically.
+
+Generate the runtime pack with:
+
+```sh
+node scripts/build-de-exercise-names.mjs
+```
+
+Translate a stage — one equipment family at a time, dry run first, `--apply`
+to write:
+
+```sh
+node scripts/translate-de-exercise-names.mjs --equipment='trap bar'
+node scripts/translate-de-exercise-names.mjs --equipment=dumbbell --limit=999 --apply
+```
+
+`--limit` caps how many names one invocation translates and defaults to 40, so a run
+without it stops early and still exits successfully — pass a limit at least as large as
+the family you are translating. Names already in `de.json` are skipped, so re-running
+after an interrupt costs nothing.
+
+Equipment and identity-changing qualifiers (assisted, weighted, one-arm,
+seated, incline …) are checked against `scripts/de-name-rules.mjs` while the
+batch is still in hand: a violation goes back to the model as a correction
+naming the offending word, and a batch that cannot be fixed is never written.
+The same rules run again in `frontend/src/lib/de-exercise-names.test.js`, so
+CI fails on anything edited in by hand afterwards.
+
+The translations were produced from the English EXDB titles with LLM
+assistance and then reviewed by a native German speaker, who went through the
+pack against the English titles: 405 of the 760 shipped names were accepted
+individually and 2 were corrected by hand. The remaining 353 were not read one
+by one — they contain none of the terms the review left open, which is the
+criterion that put the other 239 equipment names in `AWAITING_REVIEW` instead
+of in the pack. So the pack is native-reviewed in the sense stated here and no
+further: a second reviewer is still worth having. They are original
+translations and were not copied from another German exercise dataset.