How it works

1. Attack. From every allow rule, Red-Team takes one of your must-allow calls (or builds one) and replaces one argument with each bypass family for its kind. From every deny rule, it disguises the blocked word or address. Your must-deny examples are attacks too.
2. Detect. A breach is a call your gate ALLOWS on the raw text but your own rules DENY once every argument is read the way the tool reads it (decoded, Unicode-normalised, paths resolved, URLs parsed, shell quotes removed).
3. Patch. The smallest set of canonicalizations that closes the breaches without denying any must-allow call you listed. Never "deny everything".
4. Prove. A second implementation written separately from the same specification, standard library only, regenerates every attack and re-derives every decision and the patch in a sandbox. The record is PROVED only when both agree exactly.
5. Sign. SHA-256 chained, Ed25519-signed record and an offline auditor bundle with both verifiers.

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.