CL-WA-llm-latency-long-request-counts-and-endpoints
DESCRIPTIVE reproducible from public raw logs
- Metric
client_visible_latency- Analysis rule
Keep schedule capacity, started requests, terminal requests, uncensored completions, and censored/unstarted requests separate; quantiles use completed requests only.- Publication status
STATUS_ONLY- Evidence status
reproducible from public raw logs- Interpretation
Client-visible timing is not provider transport latency, task success, or a Kogen Build efficacy result.- Reported values
{"all_completed_idle_gap_over_90s_n":17,"collector_fault_n":1,"completed_600s_or_1200s_crossings_n":0,"completed_wall_range_s":[112.899,469.428],"first_stdout_max_s":1.223,"graded_n":0,"maximum_planned_n":24,"request_attempts_n":18,"terminal_records_n":18,"uncensored_completed_n":17,"unstarted_n":6}
Round: llm-latency-long