CAIN-42 CAIN Studio2026-11-02 upgrade
We ran the tool as a client engagement on a project where we knew the right answers, read the report as the client would, and fixed four things the client would have found.
| Gap | Fix |
|---|---|
| The report never said which files were clean | "Files with no findings (N)", with the list and what it does not mean; a note when more files have findings than the report shows |
| The Method section listed 5 kinds of check; the engine has 25+ | Rewritten into six accurate groups, plus how severities are set |
| The limitations gave no measured accuracy | Every report now states the benchmark, false-alarm and blind-set results, including 3 of 12 |
| No place for a person's review, so "a reviewer looks at it" was not true of the tool | --notes and --reviewer: notes signed into the report; reviewed_by_person in the signed record; without notes the report says in bold that no person reviewed it; a name without notes is refused |
Also new: audit-code-check.py, so a client can prove the signed report is about their exact code (until today the README promised this, but the client had no tool to check it); a public sample deliverable; and a README correction (each run writes its own one-record ledger, not a chain across a client's audits).
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.