Guide
Handling incomplete and unsupported findings in an AI application
Show incomplete runs with their reasons, keep supported findings usable, and never relabel an incomplete result as finished. Users trust a system that says what it could not confirm.
For product designers and engineers. Written by the Encevra engine team. Reviewed 2026-10-12.
Statuses you will see
The public API reports queued, running, released, incomplete and failed. Released means the brief passed the release checks. Incomplete means the gate withheld a clean release and returned reasons. Failed means the run did not finish.
Interface patterns
For incomplete runs, show the reasons first, then the findings that were reviewed, clearly marked. Offer a path: add material, narrow the question, or send to a person.
Billing consequence
Under self-serve billing only released runs are charged. Incomplete and failed runs release their credit hold.
Example empty and partial states
Write the copy for these states before launch, because users will see them early.
- Incomplete: 'Some findings could not be confirmed. Here is why, and what was checked.'
- Failed: 'This analysis did not finish. Nothing was charged. Try again or contact support.'
- No findings: 'Nothing was flagged in the text we could read.' Never 'no risks'.
Frequently asked questions
- Am I charged for an incomplete run?
- Not under self-serve billing. Only released runs are charged.
- Should I hide incomplete runs?
- No. Show the reasons and the reviewed findings, clearly labelled.
- Does no findings mean no risk?
- No. It means nothing was flagged in the material that was read.