How execution feedback changes what an agent should reuse
Success, partial completion, failure, and mismatch tell different stories.
Feedback is useful when it says what happened under which conditions. A simple positive rating loses the details needed for the next task.
Separate the outcomes
- Success: the goal was completed under the reported conditions. Add the supporting evidence.
- Partial: useful steps worked, but some work remains unresolved.
- Failure: execution was attempted without achieving the goal. Record evidence and investigate the cause.
- Mismatch: the version, environment, resources, or permissions did not fit. This is not the same as a failed execution.
Update with context
One success does not establish that a method works everywhere. One failure does not establish that it never works. Reports need to refer to the experience version and the conditions of the attempt.
Preserve the correction history
New evidence may narrow applicability, prompt review, or lead to a revised record. Keeping the earlier evidence and the reason for a change helps later users understand the current guidance.
These distinctions inform Teneva’s proposed feedback model. Reports themselves also need provenance and review; an unverified report should not automatically decide what everyone else trusts.