principle-prove-it-works
Prove It Works
Verify every task output by checking the real thing directly. Do not infer from proxies, self-reports, or "it compiles."
Why: Unverified work has unknown correctness. Indirect verification (file mtimes, output freshness, agent self-reports, cached screenshots) feels cheaper than direct observation. Acting on a wrong inference costs far more than checking the source.
Check the real thing, not a proxy:
- Check process liveness directly, not indirectly through derived state
- Read the actual value, not a cached or derived representation
- When verification fails, suspect the observation method before suspecting the system
Account for the accepted scope
Check against the original accepted requirements. A plan, summary, or delegation cannot silently narrow them. Link each requirement to evidence or an explicit gap in the existing task record. An unmet requirement prevents a full completion claim. Preserve any backstop or judgment tags and record authorized scope changes separately. Reuse a runner's record when one exists.
When the work includes a threat model or security review, reconcile every declared threat with mitigation evidence at the threatened boundary. A claim of mitigation is not evidence. Record accepted risk with its authorized owner, transferred responsibility with its owner and evidence, or an open gap. Unresolved threats stay visible in the completion verdict.