What it is
A broad, honest map of where a decision model holds up.
What it does: Runs Jev across many public datasets and stress tests, then publishes per-domain results.
Why it belongs in the showcase: Helps you pick thresholds and spot weak areas before you build.
Caveats: One model version; small samples per dataset.
Evidence
Public MIT repo created 2026-10-19 with a live GitHub Pages site. README states 63 public datasets in 27 domains, 33,817 requests, stress tests for option count, numbers, injected text, padding, wording and batching, and a mapping of TypeSafe's nine documented weak spots to measured results. Only aggregate results are published; every number on the site is generated from stored answers.
Caveats
- Independent and unofficial; one model version (September 2026), 300 items per dataset
- Public datasets may have been seen in training
- The harness itself is private; provenance scripts are for reading only
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.