What is the verification control surface?

The verification control surface defines what must be checked before an AI-agent action, claim, or output can be treated as complete, correct, safe, or usable.

What is false-green completion?

False-green completion occurs when an agent declares completion and surrounding checks pass even though the substantive task obligation was not performed.

Verification Control Surface

The Verification Control Surface defines what must be checked before an agent action or output can be treated as complete, correct, safe, or usable. Verification must test the task obligation, not merely the surrounding system.

What question does it answer?

Did verification prove the work that mattered?

What does verification failure look like?

Failures include skipped or fabricated checks, checks run against the wrong state, policy advice not checked against current policy, legal citations not authenticated, and false-green completion: tests pass and the agent reports success even though the requested work is absent.

A green test suite may prove that existing behavior remains intact without proving that a requested feature exists. Verification is sufficient only when its target matches the material obligation.

What evidence should an audit inspect?

the declared verification target;
checks mapped to each material obligation;
exact commands, reviewer gates, or authoritative lookups;
input and source-state identity;
retained results and failure polarity;
whether checks could be skipped or substituted; and
whether failure actually blocked completion or reliance.

What is the decision consequence of a gap?

If verification did not test the material claim, the workflow cannot treat a passing result as evidence of completion. The correct outcome is a limited or not-established finding, not a favorable inference from unrelated green checks.

Related Terms

All Control Surfaces → Evidence →