jev-benchmarks v0.1.0

First public research release of jev-benchmarks.

Highlights:

• reproducible Jev and GLiNER2.5 pilot across AG News, DAIR Emotion, and BTZSC Banking77
• accuracy, macro-F1, calibration, selective-risk, latency, paired-bootstrap, and cost reporting
• manifest and prediction identity checks with SHA-256 artifact provenance
• machine-readable aggregate results and an evidence-first README
• typed package, Apache-2.0 licensing, contribution and security policies
• CI-tested on Python 3.11, 3.12, and 3.13 with a 90% coverage gate

See the committed report and JSON artifact for the full result and limitations.