The False Pass Checklist
Seven questions to ask any AI testing tool before you trust its green.
A false pass is a test that reports PASSED without proving it. These seven questions tell you whether your AI tester can produce one, and whether you would notice.
Wopee.io's answers describe verdict integrity, which is in experimental preview.
- 1
When the agent says PASSED, what did it record as evidence for each check?
Why it matters: A pass with no record is an opinion.
Wopee.io's answer: Every run keeps the agent's steps, checks and screenshots.
- 2
Does anything check the pass against that record, or only the agent's own account of it?Must answer
Why it matters: An agent can describe a check it never ran.
Wopee.io's answer: Integrity gates compare the reported pass with what the run actually recorded.
- 3
What happens to a pass the record doesn't back up?Must answer
Why it matters: If it stays green, the dashboard is lying.
Wopee.io's answer: In Enforce, an unproven pass becomes Incomplete.
- 4
Is "unproven" a state of its own, separate from passed and failed?
Why it matters: Unproven is not broken: it needs a better check, not a bug ticket.
Wopee.io's answer: Yes: Incomplete (amber), with the reason in the report.
- 5
Can you try the stricter verdict on your own runs before it changes any result?Must answer
Why it matters: Teams adopt what they can trial without risk.
Wopee.io's answer: Report only, on by default: it flags disagreements and never changes a stored result until you switch on Enforce.
- 6
Can a product owner trace a green result to its evidence in one click?
Why it matters: Trust has to reach people who don't read logs.
Wopee.io's answer: The run report shows the Verdict Integrity section with each gate's finding.
- 7
Does the vendor say where the tool is still weak?
Why it matters: Honest limits are a sign the rest is true.
Wopee.io's answer: Yes: one switch per project today, per-check switches in design.
If a tool can't answer 2, 3 and 5, its green is a claim, not a result.
See how Wopee.io answers 2, 3 and 5 on a real run.
Watch “When AI tests pass but prove nothing” on No proof, no pass, or try it on your own runs.