§16.8 · Technical dimension · 7 metrics
Technical scoreboard
Substrate performance evidence. Call setup latency at P50 and P95/P99 by cell, end-to-end conversational latency including barge-in response, session failure rate, model routing success rate, infrastructure cost per minute, and uptime per cell against contractual SLA. Owned by Engineering; reviewed weekly.
- Owner
- OWN_ENGINEERING
- Review cadence
- 7 days
- Metrics
- 7
- In red
- 0
| Metric | Latest | Status | Trend | Q4 target | Year-2 target | Source |
|---|---|---|---|---|---|---|
Call setup latency (P50, by cell) P50 time from carrier offer to media flowing, per cell (Ghana, US-East, US-West, EU-Frankfurt, UK-London, CA-Central). The median substrate-performance signal. METRIC_CALL_SETUP_LATENCY_P50 | — | unknown | — | 330 ms | 280 ms | prometheus |
Call setup latency (P95 + P99, by cell) P95 and P99 call setup latency per cell. Tail-latency evidence — the percentiles enterprise buyers evaluate against contractual SLA. METRIC_CALL_SETUP_LATENCY_P95_P99 | — | unknown | — | 650 ms | 550 ms | prometheus |
End-to-end conversational latency including barge-in (P95) P95 time from end-of-user-utterance to start-of-agent-response, including barge-in resumption. The single number that determines whether the voice experience feels human. METRIC_E2E_CONVERSATIONAL_LATENCY | — | unknown | — | 470 ms | 400 ms | prometheus |
Session failure rate Proportion of sessions that terminate in a failure mode (unreachable, transcoding error, agent runtime crash, etc.). Trending downward as platform reliability matures. METRIC_SESSION_FAILURE_RATE | — | unknown | — | 0.5% | 0.3% | prometheus |
Model routing success rate Proportion of model-gateway routing decisions that produce successful inference (completion + within-budget latency + no fallback cascade). Evidence of hybrid-intelligence stack operational quality. METRIC_MODEL_ROUTING_SUCCESS_RATE | — | unknown | — | 99.2% | 99.6% | prometheus |
Infrastructure cost per minute Fully-loaded infrastructure cost per voice-minute served. Trending downward as scale efficiency compounds; primary input into unit-economic leadership. METRIC_INFRA_COST_PER_MINUTE | — | unknown | — | $0.026/min | $0.020/min | database |
Uptime per cell and jurisdiction (vs SLA) Rolling 30-day uptime for each operational cell (Ghana, US-East, US-West, EU-Frankfurt, UK-London, CA-Central) measured against the contractual SLA. Evidence of multi-jurisdictional operational reality. METRIC_UPTIME_BY_CELL | — | unknown | — | 100.0% | 100.0% | prometheus |