FAQ

Isn't the model just grading its own homework?

No. The model's test is recorded, but it never decides the verdict on its own. A VERIFIED_FIX also needs an independent reference test the model never saw, failing on the original and passing on the fix, and a clean re-scan by a deterministic detector.

Why should I trust your key?

You should not trust it blindly. Fetch it from the site, compare its fingerprint with us over a second channel, and pin it. The verifier refuses to say VALID without a pin, and a key shipped inside the zip proves nothing.

Does a valid signature mean the fix is correct?

No. It means the record was not changed after signing. Whether the tests really pass is what the re-run kit checks. Whether the fix is actually right is what your own tests check.

Why are there only 8 records?

Live fixing on the public site has been off since mid-October. We show the runs that happened rather than simulated ones.

Why does record 2 say VERIFIED_FIX if it would not qualify today?

It was signed under the rule in force on 2026-10-14. Evidence is never rewritten. It is disclosed and flagged instead. See the upgrade page.

Can the code under test fake the reference test's result?

Since 2026-10-23 the reference test runs isolated from the code it judges, precisely to stop that. It has not been tested by an independent party yet; that is in scope of the paid review we are commissioning.

Is this certified?

No. There is no certification and no third-party audit. Everything here is built so that anyone can check it.

Which AI model does it use?

Every recorded run used Claude Opus 5.5 (claude-opus-5-5). The verdict rule does not depend on which model wrote the fix: the independent test and the detector decide.

FIRME AI CORPORATION · CAIN-42 is a Byzantine trust fabric. Numbers marked "our own check" were run by us, not by an independent party. Nothing on these pages is a certification.