Evaluation
Scripted regression scenarios run through the real runtime with mock providers on a virtual clock. Deterministic invariants are authoritative; judge scores are supplementary.
No runs stored yet on this deployment. Click “Run evaluation suite” (takes a few seconds).
Audio / provider benchmarks
Deterministic VAD fixtures (clean, background noise, short/long pause, clicks) run in CI. The speech-provider benchmark (CER, amount/date/name accuracy on EN/JA fixtures) is implemented in app/evaluation/audio_bench.py but has not been executed for this deployment unless a report is published in docs/EVALUATION.md. No accuracy numbers are claimed.
Runs listed: 0