← Bramble

Bramble ✦ Dev Dashboard

loading…

Improvement rounds — one hypothesis per round, newest first

–

Voice-to-voice latency by eval run (ms, warm turns)

content p50 — first real reply audio content p95 perceived p50 — first sound of any kind dashed = round targets (2000 / 700 ms)

Suite pass rate by run (%)

safety persona conversation benign

Quality (1–5) — Claude judge gates, llama3.1:8b trends

Eval runs

Deployed

–

Improvement backlog

–