End of speech to first answer audio, p50
Test: Latency Lab derived span speech_end_to_first_audio across all 21 journeys (reflex and cognition lanes).
End of speech to first answer audio, p50
Protocol
Test definition
Latency Lab derived span speech_end_to_first_audio across all 21 journeys (reflex and cognition lanes).
Limits
Limitations
- Typed inlet (no microphone/ASR audio); hosted cognition model over the network; 21 turns per run (1 cold, 20 warm), p99 not computable; same-day optimisation loop, 21/21 journeys correct; not a certified release figure.
Log
Recorded values
| Variant | Value | Evidence | Samples | Source |
|---|---|---|---|---|
| Baseline, pre-optimisation 11 Sept 2026 | 2,749 ms | MEASURED | 21 | docs/analysis/latency/lab-20260911T221912Z.json |
| Plus task-relevant tool projection 11 Sept 2026 | 1,107 ms | MEASURED | 21 | docs/analysis/latency/lab-20260911T223508Z.json |
Source paths refer to the private JARVIS repository and runtime records. They are listed for audit and are not published.