The Concordance · a test of the tools
The Fabrication Gauntlet
Every AI answer is, in the end, “trust me.” This one hands you a receipt.
Each claim below is the kind of thing you might be told. The engine doesn't guess — it checks the math underneath, returns a verdict, shows the worked trail, and mints a permanent, tamper-evident seal you can re-check yourself. A false claim is refuted as citably as a true one is confirmed.
We built this expecting to catch LLMs failing. Then we tested honestly — and a leading model flagged every one of these correctly. That is the point. A smart model is usually right, on its word, with no proof and no guarantee. The engine is right every time, shows its work, and gives you a seal anyone can re-check — and it has never confirmed a falsehood across its benchmark. Verification isn't about catching a dumb model. It's about not having to trust a clever one.
The claims · engine verdict · re-checkable seal
Loading the sealed record…
Bring your own claim
Paste anything you're unsure about — a statistic, a conversion, a verse, something an AI told you. The engine will prove it true or false and seal the result. Every claim you bring grows the record.
Reproducible. Every verdict here came from the public engine: POST a structured claim to
/derivation/verify (or prose to /derivation/solve) and you get the same verdict, trail, and seal. Open any /seal/<hash> to recompute the hash and confirm it was not altered. The engine verifies a provided claim — it never generates the answer.
← The work area · bring anything