jevbooks

← All projects

decider

Mapika/decider

One-pass typed decisions with calibrated probabilities, no text generation.

What it is

decider is an open reproduction of the System One model class (TypeSafe AI's Jev). It reads a state and typed questions, returning calibrated probability distributions for each question in one forward pass. Released as 2B and 35B MoE models, for developers needing fast, structured decisions.

How it uses Jev

Jev makes decisions by answering typed questions (Choice, Score, Noul) over a given state. The model reads the state and questions, then outputs probabilities for each option. In code, these probabilities are used for routing, scoring, or hierarchical beam search, as shown in examples.

Primitives:choicescorenoul

Technique worth stealing

Schema-first layout with cached prefix for fixed question sets, enabling fast one-pass decisions.

Try it

pip install git+https://github.com/Mapika/decider, then use Decider class as in README.

View on GitHub

judged by Jevjev-1.13.0

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-centricyes0.88
  • Shows a System One patternyes0.68
  • Handles uncertaintyno0.05
  • Measuredno0.09
  • Runnableno0.34
  • Worth recommendingunclear0.41
  • Model replicayes0.96
  • Problem scopescore on a 0–2 scale0.45
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.