Before you call a case done

  • The subject is a FinchNode demo subject with a repo fixture, or the package carries fixture/ and you accept offline only.
  • check reports 0 errors, and every warning is fixed or explained.
  • trial passes, or every test it flags is updated to the new case in the same change.
  • The procedure plays in the headset, or the requester knows the case is interview only.
  • The persona matches the chart, and the voice doesn’t collide (or the voice change is made).
  • Every chart risk is covered by a correct pick.
  • install --write is done, and gates is green except steps that can’t run here.
  • Generated files (ScalpalIds.cs, the asset manifest, the bundle) are committed with the change.
  • Someone has played it once, and a clinician has reviewed the content before it’s shown as more than a demo.

Gates

From services/preop, npx tsx $T gates runs these in order: Unity editor validations and a headset run need a person.

Ids that must line up

The headset re-checks at every stage that the case, interview scorecard, coach session and Time-Out check all carry the same subject, procedure and case id.

Places that are still hard-coded

What a new procedure costs

A procedure the headset can’t run yet needs more than data:
  1. Data: the procedure, coaching for every step, step roles, facts for any new anatomy, a procedure site.
  2. Code in TypeScript and C#: ideal actions for every new open-body step, the branches that check for the literal open_appendectomy id, and any new tool verbs. The two engines must stay number-for-number identical.
  3. Unity: the handoff allowlist, anatomy for the region with colliders, the wound and wall layers (hard-coded at McBurney’s point today), organ handling, the tool tray, briefing art and voice, then a new APK.
The skill writes the data for a tier 2 case and hands back this list filled in for that procedure.