trace_success_rate

LLM output quality degrades (hallucinations, incorrect answers, poor user satisfaction) but goes undetected due to lack of continuous feedback monitoring.

Multi-step agent workflows fail at specific tool invocation stages, but lack of granular tracing makes debugging difficult and prolongs incident resolution time.