False green has several ordinary causes
A worker may claim completion after running only part of the suite. A dashboard may keep showing the last successful run after a new run failed. A benchmark may score a stub or a cached response rather than the live system. None of these failures requires deliberate deception. They happen when the reporting path can detach from the artifact it describes.
The shared symptom is a green signal with no reproducible chain from task identity to tested artifact, command, environment, output, and acceptance decision. If you cannot walk that chain, the display is telemetry rather than a release gate.
Bind the receipt to the bytes that actually shipped
A verification receipt should record the artifact digest, exact command, environment identity, exit status, relevant output hash, and verifier identity. Completion should reject evidence produced against an older digest.
That rule closes a common race: tests pass, the artifact changes, and the stale green result stays attached to the work. Content-addressed evidence makes the mismatch explicit. A pass recorded against an earlier tree is not a property of the current one.
Worker telemetry is for debugging, not acceptance
Worker-reported tests are useful for fast feedback. They should not be the only authority that moves work into an accepted state. An orchestrator or independent verifier should execute the sealed acceptance checks and record the result.
The separation matters most when the worker can edit tests, fixtures, or dashboards. Ownership scopes should prevent one assignment from changing both the feature and the evidence contract that authorizes it. Otherwise the system can grade its own homework and then update the answer key.
Three mutations a sham dashboard cannot survive
First, replace the production call with a stub; the live-path check should fail. Second, alter the artifact after verification; the digest check should fail. Third, remove the verification hook; completion should become impossible rather than silently trusting the worker.
A dashboard that survives those mutations has a stronger claim to green status. Without them, keep the display if it helps operators, and label it as telemetry rather than acceptance evidence.