Same Telemetry, Different Query Interfaces: Results from 504 Agent RCA Runs
Agent RCA Bench holds the model and prompt fixed and changes only the query interface: 504 end-to-end investigations across six models. Querying GreptimeDB instead of Prometheus, Loki, and Tempo produced 40% fewer wrong diagnoses at about half the input tokens.
→
