5-min clock-aligned slots, refreshed every 5 min by rescanning calls whose updated_at falls in the last 30min. A call is “blank” when telephony duration ≥ 5s but the bot produced fewer than 10 audio bytes. Alert fires when any single slot in the trailing 30 min has ≥ 15 blank calls.

5-min clock-aligned slots, refreshed every 5 min by rescanning calls whose updated_at falls in the last 30min. A call is “high-latency” when call_telemetry.latency.end_to_end.p75≥ 3,000 ms. Alert fires when any single slotin the trailing 30 min has ≥ 5 high-latency calls.

Click a point on the chart to pin its breaching call_ids below.

One line per provider-model used on the call (llm.transcriber for STT, llm.synthesizer for TTS, llm.model_name for LLM). Y = the max component p75 (ms) seen in each slot (max-of-maxes when aggregated). Dashed line = its threshold in per_provider_latency_p75_ms. Toggle provider-models like the Piya pod selector.