What it is
openJev-verdict-2.0 is a 149.6M parameter decision model that evaluates typed decision schemas (Choice, Score, Noul) in non-autoregressive forward passes with calibrated confidence. It targets enterprise financial, security, customer support, and agent trace observability workflows.
How it uses Jev
The model takes a state and typed questions (Choice, Score, Noul) and returns calibrated probabilities. It uses dual-channel calibration: a distribution head for soft labels and a confidence head for correctness. Deterministic software retains control over thresholds and business rules, using the model for bounded semantic classification.
Primitives:choicescorenoul
Technique worth stealing
Symmetric Permutation-KL penalty between option-shuffled batches reduces prompt-order bias.
Try it
Run `cd webgpu-demo && python3 -m http.server 8080` and open http://localhost:8080 in Chrome or Edge.
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-centricyes0.83
- Shows a System One patternno0.25
- Handles uncertaintyno0.05
- Measuredyes0.94
- Runnableno0.18
- Worth recommendingno0.24
- Model replicayes0.93
- Problem scopescore on a 0–2 scale1.35
- About Jevyes0.91
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.