Jev implementation checklist
Implementation checklist — independent, unofficial. Not a clone of TypeSafe’s build narrative. Keys come from their console or a marketplace, never from jev.pro. Canonical: docs.typesafe.ai.
Client and channel
- Pick one production door; record model id + auth in config (matrix, dated weekly).
- First-party: official SDK or
POST /v1/systemonewith your own 429/529 backoff. - Do not mix
boolean/noulorjev-latest/typesafe/jev-1.13across doors. - Env:
TYPESAFE_API_KEY(or the door’s name). Never commit tokens. - Pin
jev-1.13.0(or current versioned id) once thresholds exist; logresponse.model.
Questions module
- Single file for instructions, criteria, and floors (skill review rule).
- Dual-stack smoke test (Python + TS) on the same JSON body if both languages ship.
- Filter
statebefore POST; measure input tokens.
Errors and traces
- Map 401 / 422 / 429 / 529 per error envelope.
- Marketplace 402 / customer_verification handled as callouts, not retries of the body.
- Persist a decision trace on every production call (or sampled + all handoffs).
Eval
- Gold JSONL + frozen questions (offline evaluation).
- Score policy actions, not only argmax.
- No invented 193× in dashboards — quote TypeSafe if you show vendor speed.
Next: launch checklist. Previous: design. Hub: Open.
Sources
Public TypeSafe or adjacent documentation only. No private claims.