Prompt sent to Claude Sonnet 4
Without Raptor
Raw model output · no system prompt, no record
No model record. No hash. No governance. No proof.
With Raptor
Governed · policy shaped this answer, then sealed it
6 governance checkpoints · 1 Merkle proof · 0 trust required
Same model. Same prompt. The left is the model's raw default — an unshaped draft with no record it ever ran. The right was shaped by governance policy into an accountable decision, then sealed so it can’t be altered after the fact.
Try to break it

Change what the AI said. Watch Raptor catch it.

This is the exact answer the governed run produced. When Raptor sealed it, it recorded the answer’s SHA-256 fingerprint in the hash chain you just saw. Edit a single character below — a dose, a name, the “Yes” into a “No” — and the fingerprint stops matching. That mismatch is the whole point: with Raptor, no one can quietly rewrite what the AI did.

SHA-256 recomputed live in your browser · nothing is sent anywhere
Sealed in the chain
This text, right now

That check ran against this demo’s receipt, live in your browser. Want to prove it’s not a trick? Verify a real, published Raptor receipt yourself — the client-side verifier recomputes the proof and Raptor’s signature entirely in your browser, with no account and without trusting Raptor. Or paste any receipt hash for a quick server-side check.

Full Evidence Comparison

FieldUngovernedGoverned

Same model. Same prompt. One answer was governed, shaped, and sealed — the other is just text.

Talk to the founder