No model runs on this page
A language model handed a lead record will write “your team grew by 240 people this year” when the only thing on file is a size band. Prompting cannot prevent that, because a prompt is a request, not a guarantee. This is the deterministic check that does. Give it an evidence ledger and a drafted email and it answers one question, with reasons: does every factual claim resolve to something you actually know?
Everything below runs in your browser. The validator is the same file the product runs, on the same inputs; nothing is uploaded and nothing is stored.
| Id | Fact | Assertion |
|---|
Observed: collected from a source
Inferred: a model reading of collected facts
Unknown: we looked and found nothing. Never citable.
| Claim | Cites |
|---|
The validator runs as soon as the Python runtime finishes loading.