Caliper takes a drug/adverse-event claim and puts it through an adversarial review — a proposer, a falsifier, and a referee, checked against public safety data (FAERS and related sources). Every finding is pre-registered before evidence is pulled, so the hypothesis can't be quietly reshaped to fit what was found.

Every published finding is sha1-locked (the finding's identity is fixed to its content — you can't silently edit a finding after publication without changing its address) and Ed25519-signed (a cryptographic signature over the canonical evidence, so tampering after signing is detectable). Signed findings chain into a public, append-only ledger — each entry references the hash of the entry before it, so the whole history can be verified as unmodified, not just any single entry.

How to verify a finding yourself

1. Open any finding's ledger entry and click "verify" — or go directly to /caliper/verify/<sha1>.
2. Download the audit packet (/caliper/audit/<sha1>) — it contains the canonical evidence, the signature, and the exact steps to check it offline.
3. The signing key used is published at /.well-known/jwks.json — a standard JWKS document listing the active public key for every RocSite product (Caliper, Discovery, and the AI Governor). One document, every product's signature checkable against it, so you never have to trust RocSite's own database to confirm a finding hasn't been altered.

Relationship to the AI Governor

The AI Governor applies the same trust model — signed, independently verifiable evaluations — to medical-AI-vs-ground-truth checks (e.g. imaging device validation). Caliper and the Governor currently sign with different keys (each scoped to its own product), but both keys are published in the same JWKS document above, so a single public registry covers every signed claim this platform makes.

Live ledger integrity check: /caliper/api/ledger/verify