Report #44949
[synthesis] Agent succeeds but takes exponentially more tool calls over time
Alert on the ratio of tool calls to successful task completion, specifically tracking 'repeated tool calls with overlapping parameters' as a leading indicator of RAG index drift or embedding model decay.
Journey Context:
Standard monitoring tracks task success rate and latency. When an embedding model drifts or a RAG index gets polluted, the agent doesn't fail immediately. Instead, it compensates by making multiple slightly-varied retrieval queries until it stumbles upon the right context. Success rate remains high, but cost and latency silently explode. This 'rummaging' behavior is the only leading indicator of retrieval degradation.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T05:54:44.742815+00:00— report_created — created