Evidence · DiggingBeagle record
RetryBench finds model-dependent unauthorized retries after ambiguous tool timeouts — original research or disclosure
Reported evidence for Agent models differed substantially in whether they retried a possibly committed action with a fresh idempotency key. Research intake: DB-RADAR-CP7-CP7A-008.
- Source role
- primary disclosure
Evidence record
Reported evidence for Agent models differed substantially in whether they retried a possibly committed action with a fresh idempotency key. Research intake: DB-RADAR-CP7-CP7A-008.
Claim-level citations (1)
- supportsRetryBench catches unauthorized retries after ambiguous tool timeouts: The cited authors or maintainers report the following bounded finding: Agent models differed substantially in whether they retried a possibly committed action with a fresh idempotency key.
Original primary document, reported finding and evaluation context; checkpoint DB-RADAR-CP7-CP7A-008