FALSEWORK

A deterministic detector for fabricated code: code that looks finished but is not. A verifier that returns True without checking, a hard-coded score, a random number presented as a measurement, an error counted as success. No AI model, no network: the same input always gets the same answer.

#4 of the CAIN-42 Top 25 0 false alarms in every test today fresh blind set: 3 of 12 caught new engine live after next deploy

The honest summary

FALSEWORK is precise and narrow. When it says a module is fake, it has been right every time we have measured. But it only knows the fabrication patterns it has rules for. On a blind set written after today's changes and run once, it caught 3 of 12 fakes. A clean FALSEWORK grade means "none of the known patterns", not "this code is honest".

That is why it is the first stage of a CAIN-42 audit, not the whole audit: a person reviews what it flags and what it cannot see.

Where it is used

• Stage 1 (DETECT) of KEYSTONE: it decides there is a defect, not the AI model.
• The CAIN-42 fabrication audit ($1,500 / $3,500 / $6,000): every module graded, with a signed evidence bundle.
• KEYSTONE CI: the GitHub Action that grades a change before merge.

On this site

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.