Three AI agents evaluate candidates for a real role. Each decision is witnessed by Rubric — signed with post-quantum cryptography and chained to the one before it. When the decision is challenged months later, the record cannot be quietly rewritten.
Real language-model agents, running in sequence: one parses each résumé, one scores it against the role's stated criteria, one issues a disposition. Every step is attested to Rubric as it happens. representative candidates · real AI reasoning
One candidate — Candidate B — was rejected, and has filed a discrimination claim. Counsel asks: what did the AI actually decide, and on what basis? Rubric reconstructs the exact decision chain. Verify it against the ledger — the cryptography runs right here, in your browser.
Imagine the rejection rationale is quietly edited after the claim is filed — changed to look more defensible. The data on disk now says something different. Re-run the very same verification against the altered record.