State the target outcome. Anchor the loop to the user's symptom, performance goal, failing test, UX defect, regression, or code quality concern. If the user gave no concrete anchor, create the fastest observable feedback loop first.

Run a pass. Complete the evidence loop and record the exact validation signal that will be re-run after the fix. Proceed only when reports Proven — 100%. If it reports Incomplete, stop and report the missing evidence. If the user asked for delegation, use it only for bounded evidence collection inside this pass; keep bottleneck ranking, fix selection, code edits, and final interpretation in the main loop.

Make a ranked fix plan. Prefer one smallest credible fix at a time. The plan must include:

Implement only the top plan item. Keep the edit scoped to the proven fault line. Avoid stacking speculative cleanup, broad refactors, or adjacent improvements unless the diagnosis showed they are part of the same fault.

Verify with the original signal. Re-run the reproduction path or measurement from step 2, then focused regression coverage, then broader checks when the touched surface warrants them.

Re-run . Diagnose the new state from evidence, not from the intent of the change. Look for:

Decide whether to continue. Continue when evidence identifies a credible next fix and the next action is safe to take. Each new iteration should have a sharper target than the previous one.