How do I catch silent failures in production?
Silent failures are the sessions that return a 200 and a confident answer that's wrong, empty, looping, or off-policy, no exception, no alert. Flowlines catches them as behavioral signals (hallucination, cascade_failure, low_quality_response, and more) scored on every session, so the failures that never threw an error still surface.
The dangerous failures in agent systems don't crash. The agent says “I've fixed it” without calling the tool. It cites a policy that doesn't exist. A tool returns nothing and the chain keeps going on empty context. Every one is a 200 in your logs.
Flowlines scores behavior, not status codes. Each session is checked against the signal registry, so a confident-but-fabricated answer fires hallucination, a broken tool chain fires cascade_failure, and a generic non-answer fires low_quality_response, none of which your error monitoring would ever see.
Because these are tracked as rates with baselines, a rising silent failure rate becomes an alert, and you can drill from the rate to the exact sessions to see what the agent actually did.
Last updated 2026-05-28