Jev 101
Start Guides
Search
中文 EN
Platform heat 1 · Quiet★ 0heat incomplete (no X)

jev-calibration-lab

A MIT independent calibration harness that compares TypeSafe Jev probabilities to human-labeled public datasets versus Claude and a TF-IDF baseline, with a GitHub Pages leaderboard; Jev/Claude numbers in-repo are still synthetic dry-run until keys run.

Repo: veermshah/jev-calibration-lab ↗

What it is

Ask whether Jev probabilities match humans, not only other models.

What it does: Human-label calibration harness with Pages leaderboard.

Why it belongs in the showcase: Clear independent calibration angle with honest dry-run labeling.

Caveats: Real Jev runs still pending keys.

Evidence

Public MIT repo, created 2026-11-01, 0 stars as of 2026-11-02. README and CI badge document five human-labeled datasets, dry-run generators labeled synthetic, real logreg baseline, and https://veermshah.github.io/jev-calibration-lab/. Distinct human-label focus vs jev-eval / jev-threshold-lab.

Caveats

  • Published Jev/Claude leaderboard arms are dry-run/synthetic until API keys run
  • Independent lab; not affiliated with TypeSafe
  • Do not treat synthetic numbers as model measurements

Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.