Skip to main content
Use this page when Eval Labs generates responses but something in the run, persistence, or environment looks wrong.

​
Symptom: Failed to fetch / transport failure

Likely cause:
  • CORS
  • wrong endpoint
  • Engine unavailable
  • network block
Capture:
  • resolved request endpoint and expected environment
  • requesting origin
  • preflight status
  • application-request status, if one occurred
  • deployed Eval Labs commit and observation timestamp
A successful preflight proves cross-origin acceptance for that observation only. It does not prove that a POST completed, that Lucia returned a response, or that a model was invoked. If the browser reports a cross-origin failure, preserve the evidence and escalate the policy mismatch through the authorized operator boundary.

​
Symptom: Eval Labs responses differ from Lucia Admin Focus Ops

Likely cause: Eval Labs and Admin are hitting different Engine environments. Resolve the expected Engine target for the active Eval Labs deployment from Current System State, then compare the captured endpoint and environment for each product. Preserve each observation’s deployment identity and timestamp.

​
Symptom: Persistence conflict or save failure

Do not retry blindly or change database constraints from a browser symptom. Capture:
  • the response status and safe error summary
  • the run and item identity shown by the product
  • whether the run remains visible after refresh
  • whether review, finalization, and export states agree
  • the deployed commit and observation timestamp
Escalate with that evidence to the persistence owner. A transport response alone does not prove which database operation or constraint caused the failure.

​
Symptom: Latest deploy seems ignored

Likely cause: Old JS bundle cached in browser. Fix:
  • incognito window
  • hard refresh
  • verify Netlify published commit

​
Symptom: Reviewer attribution is missing

If the item was generated but not reviewed or saved, missing reviewer attribution is expected. If an authenticated reviewer saved the item and attribution is absent, capture the run identity, review state, authenticated-user evidence, deployed commit, and timestamp before escalation.

​
Symptom: Export attribution exists without review attribution

This can mean the run was exported without the prompt review being saved. This is not automatically a bug.

​
Debug discipline

Do not guess. Check:
  1. Resolved endpoint and expected environment
  2. Preflight and application-request status
  3. Safe response or error summary
  4. Current deployed commit and observation timestamp
  5. Stored runtime provenance, if present
  6. Run, review, finalization, and export outcomes
  7. Authenticated reviewer state when attribution is relevant
Do not fill missing runtime provenance from the current default, a later verification, or routing history. Do not treat an OPTIONS 204, a POST 200, or a product commit as model-invocation proof by itself.