Engineering Feedback Loop

Use this checklist to keep AI assisted work tied to evidence and decisions.

Frame

1. State the question and expected outcome.
2. Identify the owner of the decision.
3. Select the sources allowed as evidence.

Inspect

1. Check current executable behaviour.
2. Read relevant decisions and contracts.
3. Separate current behaviour from proposed direction.

Change

1. Make the smallest useful change.
2. Record assumptions that affect the result.
3. Avoid unrelated changes.

Validate

1. Test the code behaviour.
2. Test the data at the expected grain.
3. Review failures as evidence, not just obstacles.

Decide

1. Release when the evidence supports the outcome.
2. Return to implementation when the behaviour is wrong.
3. Return to design when the model or ownership is wrong.
4. Stop when the value no longer justifies the work.

Share

1. Keep implementation detail with the code.
2. Store rationale with architecture decisions.
3. Add reusable, reviewed guidance to the knowledge base.
4. Record what remains uncertain.

Back to all examples