Verification Before Completion

Success is claimed after running the proof, never before. Fresh evidence against the current state of the code — not confidence, memory, extrapolation, or a subagent's word. This covers paraphrases too: any wording that implies things work is a claim.

The gate

Before any statement implying success:

1. Name the command that would prove the claim.
2. Run it fresh, complete for the claimed scope, against the current state.
3. Read the output — exit code, failure count, warnings — not just its tail.
4. Output confirms: make the claim, citing the evidence. Output refutes: report the actual state with the output, no softening.

Claims and their proof