What it does
It reports per-ticket latency with p50 and p95, and accuracy overall against accuracy above a movable confidence threshold, so the abstain rate stays visible.
Maker-reported (not independently measured by JevMade): sample size 25-1000, default confidence threshold 0.90 · estimated cost = total input_tokens / 1e6 x $0.042 (output free) · for long runs the feed renders the latest 200 cards; metrics cover every ticket
- Primitives
- choice
- Platform
- TypeScript
- Added
- Project created
Source checked 2026-09-24 — opened the GitHub repository directly.