Skip to main content
Review Lucia by asking whether the response worked for the human situation. Polished language is not enough.

​
Review the response, not the vibe

For each item, read:
  1. the prompt
  2. Lucia’s response
  3. any suggested review selections
Then decide what you believe. Suggestions can help you move faster. They are not truth. Behavioral Observatory labels are a separate owner/admin surface in the verified 2026-08-12 access contract. Use that workflow only after an authorized role/scope change and explicit assignment. See Your Role and Access for the canonical evaluator boundary and Eval Labs Current System State for dated platform truth.

​
The evaluator questions

Ask:
  • Did Lucia understand the user?
  • Did Lucia avoid overclaiming?
  • Did Lucia give a useful next move?
  • Was the answer clear without extra scanning?
  • Was the tone calm and appropriate?
  • Did the answer preserve trust?
If the answer is “not sure,” mark that uncertainty instead of forcing confidence.

​
Saving the review

Use the review controls honestly:
  • score the visible dimensions
  • answer Quick Review questions
  • add Human Guidance Evaluation scores when useful
  • write a short note when context matters
  • flag senior review when uncertain or concerned
  • save each item
Finalize only after every item has been reviewed.

​
Escalate

Ask an owner/admin when Lucia may have overclaimed, created risk, missed a sensitive human moment, or revealed a pattern that should become reusable learning.