Registry

The exercise registry is the interop anchor that resolves an ExerciseRef (§6) to a stable base movement plus structured facets. It is a separate, independently-versioned, fetchable artifact, decoupled from the spec — the spec defines the mechanism; the registry ships content on its own semver cadence. The data is CC0.

There is a parallel measurement-type registry (§4.5) for canonical Measurement type tokens and channel conventions, using the same mechanism.

Hand-curated canonical + crosswalk

The model is deliberately two-layered:

• Canonical entries (data/exercises.json) are hand-curated for quality and stable identity — each a deliberate, reviewed movement with a stable id. This is the authority.
• Crosswalks (crosswalk/<source>.json) map a third-party source’s movement id → canonical id (or null). A crosswalk is both (1) a resolution table for refs that arrive named in an incumbent system and (2) the curation worklist — every null is a movement not yet curated.

Third-party data is crosswalked, never imported

Third-party datasets (e.g. free-exercise-db) are used only as crosswalk sources — never imported as canonical entries. Facets are always hand-authored against §6.3, regardless of source. That’s how the registry’s quality and id stability are owned, not inherited.

Entry model (§6.3)

{
  "id": "squat.barbell.high-bar",        // canonical, unprefixed, dot-segmented
  "names": ["High-Bar Back Squat"],      // first is the preferred label
  "facets": {                            // classification (intrinsic) + variation (distinguishing)
    "modality": "strength", "movementPattern": "squat", "mechanic": "compound",
    "anatomy": { "primary": ["quadriceps", "glutes"] },
    "equipment": "barbell", "barPosition": "high-bar", "laterality": "bilateral"
  },
  "attributes": { "met": 6.0 },          // registry attribute (NOT identity), e.g. MET
  "coded": { "wger": 73 },               // advisory crosswalk to incumbents (§6.4)
  "source": "curated"
}

Ids follow ^[a-z0-9]+(?:-[a-z0-9]+)*(?:\.[a-z0-9]+(?:-[a-z0-9]+)*)*$. Consumers MUST NOT infer semantics from segment structure — the registry, not the string, is authoritative. A movement’s attributes (e.g. MET intensity) are registry attributes of the base movement, not identity codes and not part of ExerciseRef.

How to contribute

The canonical set starts small and grows by reviewed addition (PRs). Summarizing the registry’s CONTRIBUTING.md:

1. Decide the base movement. Lowercase dot-segments, base[.variation…]; the base is the movement pattern (squat, bench-press, curl), variation segments narrow it. Reuse an existing base where one fits — don’t invent synonyms.
2. Order variation segments consistently: base . equipment . position . grip . laterality. Add a segment only when it distinguishes this movement from a sibling. Atomic movements stay single-segment (plank, run).
3. Fill facets so they agree with the id — classification (modality, movementPattern, mechanic, anatomy.primary[]) and variation (equipment, grip, stance/barPosition, laterality, rangeOfMotion).
4. Attributes (optional): met (cite the Compendium edition).
5. Crosswalk codes (optional): coded may carry incumbent ids for this entry; for bulk source mappings, prefer the crosswalk file.
6. source: "curated", "compendium" (MET-sourced attribute), or "free-exercise-db" (a crosswalked movement seeded the entry — facets still hand-verified).
7. Validate: npm run check (id format + uniqueness + facet conformance + crosswalk integrity). Green is required.

Id stability

Pre-v1.0 ids may change. From v1.0 they are stable — a published id is never repurposed. That is the whole point of hand-curation: get the id right once.

Licensing & sources

Registry data is CC0 — the interop registry must be completely unencumbered. The v1 seed is free-exercise-db (The Unlicense / public domain), supplemented by Compendium MET values; share-alike (wger, CC-BY-SA) and proprietary (ExerciseDB) sources are not used. See the registry’s SOURCES.md for the per-source evaluation.

Status (per the registry repo)

The model, schema, validator, crosswalk, and CI are in place and green; canonical curation is ongoing (the free-exercise-db crosswalk is the worklist). The registry lives in its own repository, openbody-registry.