Introduction

Your clinical AI produced an answer. Can you prove how?

clinical-receipt captures the inputs, evidence, models, tools, guardrails, and human decisions behind every clinical AI output — then lets anyone with the disclosed evidence verify the record offline.

The package proves the committed record has not changed. It does not prove the original record was true. That distinction is core, not marketing: every verification report reflects it, every doc page respects it.

What it is not

Not a blockchain. Not a hosted service. Not a compliance product. Not an LLM observability dashboard. Not a replacement for clinical validation.

What sits in the box

• A recorder that turns a workflow into a signed, hash-linked event graph.
• A verifier (@0xsarwagya/clinical-receipt/verify) with no recorder dependencies — the audit tool never sees the recorder's code path.
• A selective-disclosure primitive: reveal only what you need to, prove cryptographically that the shown pieces belong to the original receipt.
• An offline CLI that verifies, inspects, diffs, and discloses receipts with typed exit codes.
• A written specification (spec/1.0/) that any implementation, in any language, can conform to. Test vectors are byte-pinned.