Review and feedback
Use the right judgment record for an asset, statement, identity, write or outcome
Review makes a judgment explicit with its reason and supporting evidence. Different decisions have different effects; a generic approval does not authorize every operation or establish the truth of every output.
Choose the right record
| Decision | Record or workflow |
|---|---|
| Save/apply an AI-authored asset | Review and approve the exact proposed artifact revision/change set |
| Accept or reject a statement | Assertion review lifecycle |
| Review a produced output | Appropriate artifact review |
| Correct which entity a record represents | Reviewed identity correction proposal |
| Permit write semantics | Applicable write-policy proposal/approval |
| Approve a requested action | Governed-action approval where required |
| Record an observed prediction result | Prediction evaluation |
Approving a pipeline definition does not approve all future external writes. Approving a claim does not overwrite every materialized fact. Retrying an execution does not approve its outputs.
Example: the wrong equipment
If an answer uses P-102's records for P-101, investigate subject matching and identity first. If identity is correct but temperature is wrong, inspect the source and binding. If the source is correct but a failure forecast was poor, preserve the original prediction and record the actual outcome.
Choose a correction that addresses the defect. Keep the reason and evidence, update the relevant source/configuration/version deliberately and verify a separate new execution.
What feedback changes
Corrections, reviews and evaluations are durable records for investigation and future authoring. They do not automatically retrain a model, modify every mapping or guarantee that the next answer improves. Inspect how the supported workflow uses the record.
Continue with the relevant workflow
Read the authoring and review loop, assertion review, identity corrections or prediction evaluations, depending on the decision.