Results by arm

⚠ These numbers are incomplete. The durable store did not answer within its deadline, so this page is showing only what is on local disk — which on the hosted deploy is usually nothing. Do not read an empty table here as “no sessions”. Check SUPABASE_DB_URL (a direct db.*.supabase.co host is IPv6-only; the Supavisor pooler host is the one that works from a function).

Human rating means + mean boot-to-welcome, from ../logs. ← back · listen →

Rater:all

No sessions logged yet.

Flagged issues (BROKEN / N/A)

Recordings flagged unusable on /listen — excluded from all aggregates + the leaderboard. Group by reason to spot patterns (a spike in one reason = a harness or prod bug to chase).

No flagged clips.

Judge (uncalibrated, advisory)

Gemini scores from agent/judge.py — a cheap second opinion, NOT calibrated against human ratings. Treat as directional only.

No judgements yet — run `uv run judge.py` in agent/.

Actor runs (Mode B2)

Scripted AI-actor sessions from agent/actor.py — excluded from the human-rating table above. Latency = actor finishes speaking → Amia starts responding.

No actor runs yet — run `uv run actor.py` in agent/.