What it is
JevBench is Benchmark Heaven's benchmark for Jev-class decision models: it feeds a state plus a bounded rubric and expects a typed answer, ideally with probabilities. It publishes per-task outcomes, scoring code, and aggregate results for 534 frozen decisions per entrant.
How it uses Jev
The README does not describe how Jev is used in this repository's code. It measures Jev 1.13.0 as one entrant among others, and notes that classifier.dev's fast tier is Jev, but gives no integration details.
Primitives:choicescore
Technique worth stealing
Four-axis equal-weight harmonic mean over chance-corrected Intelligence, Calibration, Speed, and Cost, with gates for low intelligence and public-to-sealed gaps.
Evidence
Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.
- Jev-centricunclear0.55
- Shows a System One patternunclear0.50
- Handles uncertaintyno0.08
- Measuredno0.21
- Runnableno0.07
- Worth recommendingno0.23
- Model replicano0.07
- Problem scopescore on a 0–2 scale1.15
- About Jevyes0.94
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 23 Sept 2026.