Your AI tester said PASSED.
Did it prove it?

1. 01The trap

   A test that names two prices, and checks neither

   The test expects a checkout total of $32.39: the backpack at $29.99 plus $2.40 tax. Its only defined check reads Thank you for your order! The agent completes the order and reports PASSED, quoting both prices.

   
2. 02Pass withdrawn

   No check read those prices. In Enforce, the pass is taken back

   The integrity gates compare the reported pass with what the run recorded. Both prices are flagged as unverified, so the pass becomes Incomplete: amber, not red. The run didn't find a bug, it proved nothing, and it never counts as a pass.

   
3. 03Proof in, pass back

   Add the checks that read the numbers. Same test, rerun: green

   We added checks for the item total, the tax and the total. The rerun passes with nothing flagged, and this time the green comes with its proof.

   

• PassedThe run recorded the proof. No gate withheld the pass.
• IncompleteNot fully proven. Never counted as a pass. Fix: add the missing check, not a bug ticket.
• FailedThe run found a failure. In Enforce, a Failed with no failure on record is withdrawn to Incomplete too.

In both modes, the agent can be sent back for a missing check before it stops: at most eight extra steps per run, counted toward your plan's steps. Experimental preview: today it's one switch per project, and per-check switches are in design. How verdict integrity works

• Wopee.io

  1 · Audits the pass against the run

  ✓YesIntegrity gates check every agent pass against the run's own record.

  2 · Downgrades an unproven pass

  ✓YesIn Enforce, an unproven pass becomes Incomplete, with the gate and the evidence named.

  3 · Shadow mode first

  ✓YesReport only, on by default: a shadow verdict that never changes a stored result.

  docs.wopee.io
• TestMu AI · Kane CLI

  1 · Audits the pass against the run

  ≈Close“A separate verification step decides the verdict.” The model's own account is not enough.

  2 · Downgrades an unproven pass

  ≈CloseA July 2026 launch note: strict mode demotes tests that passed against a since-changed spec. Not on the product page today.

  3 · Shadow mode first

  –Not foundNot found in public docs.

  testmuai.com/kane-cliLaunch note, 24 Jul 2026
• QA.tech

  1 · Audits the pass against the run

  ≈CloseA review agent writes one verdict per test plan. “Passed tests count as verified.”

  2 · Downgrades an unproven pass

  ≈ClosePlan level: “Needs attention” when unverifiable tests leave part of the plan unproven. PR comments show “needs review” by confidence tier.

  3 · Shadow mode first

  –Not foundNot found in public docs.

  docs.qa.techChangelog, 22 Apr 2026
• Momentic · Mo

  1 · Audits the pass against the run

  ≈CloseEach verdict lists what the agent verified and “the gaps it could not close”. Step level: an independent reviewer judges the settled app state, the recorded actions and your goal.

  2 · Downgrades an unproven pass

  –Not foundNot found in public docs. A separate agent re-checks reported bugs, not passes.

  3 · Shadow mode first

  –Not foundNot found in public docs.

  momentic.ai/docsmomentic.ai/changelog

Public docs and product pages, read 3-4 Nov 2026. Not exhaustive: something missing from public docs may still exist in a product. We also read mabl, BrowserStack, Applitools, testRigor, Tricentis, Katalon and Playwright's test agents, and found failure explanations and healing, but no documented audit of a pass. Tell us if we're wrong and we'll fix this table.