ORIGINAL TESTS · VISIBLE METHODS · HONEST LIMITS

Verification case studies

These are documented tests performed while developing Another AI Lie. Each case shows the exact kind of material examined, the questions asked, what the evidence established, what remained unresolved and how the result affected the product.

Why publish our test record? A trust-checking service should show its own evidence. These pages do not claim that three fixtures prove universal accuracy. They let readers inspect representative tests and the boundaries we required the system to preserve.
Live public-claim test

A viral-style UK heat-pump deadline post

Five claims, eight checked sources and one important distinction between contradiction and absence of evidence.

Controlled image test

A forwarded health screenshot with no original URL

Why readable text and familiar styling cannot authenticate a screenshot or the account it appears to show.

Controlled document test

A fictional CV containing public and private claims

How one false public date can be contradicted while employment and education remain questions for authorised evidence holders.

What these studies contribute

General verification advice is widely available. These case studies add something specific: a record of how this product was challenged, the failure modes we looked for, the output boundaries we enforced and the engineering changes made when a test exposed a defect. We publish the method and limitations so readers can judge the work rather than relying on a marketing claim.