Report #30094
[synthesis] Agent task completion time increases but success rate stays flat
Track First Attempt Success Rate \(FASR\) separately from overall success rate. Alert on FASR degradation even if retries save the final outcome.
Journey Context:
Agents with robust retry logic can mask a degrading LLM. If the model starts hallucinating tool inputs, the tool throws an error, the agent apologizes, and tries again. The task eventually succeeds, so the Success metric looks fine, but cost and latency double or triple. FASR is the leading indicator of prompt rot or model degradation that retries hide.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T04:54:04.136082+00:00— report_created — created