devin-qa-pack

agent assurance — 60-second demo
agent-assurance demo — UNVERIFIED, PARTIAL, PASS verdicts
A deliberately flawed agent session goes in one end; a defensible, reproducible verdict comes out the other.
devin-evals dream synthetic sessions → devin-inspect schema contract → devin-qa-pack claims vs evidence → devin-evals rubric grading
PASSevery claim is backed by tool-call evidence in the session store.
PARTIALsome claims verified, others missing evidence — reported, not guessed.
UNVERIFIEDclaims without evidence — the audit refuses to sign off.