jevbooks

← All projects

AnyDecisionModel

mattt/AnyDecisionModel

Typed decisions from language models: probabilities, choices, scores.

What it is

A Swift package for typed decisions from language models. It provides yes-or-no probabilities, choices among options, and scores on ordinal scales, with two backends: a local MLX model on Apple silicon and TypeSafe's Jev API.

How it uses Jev

JevDecisionModel sends each batch of questions to POST /v1/systemone. The session holds one immutable state (text or JSON). Questions are independent. Results are typed: binary probability as Double, Choice with value and distribution, Score with expected level and distribution.

Primitives:choicescorenoul

Technique worth stealing

Use rotation debiasing to average probabilities over all option orderings, reducing position bias in choices.

Try it

Add the package with the MLX trait, create a DecisionSession, and call probability, choice, or score.

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-centricno0.13
  • Shows a System One patternno0.20
  • Handles uncertaintyno0.05
  • Measuredno0.06
  • Runnableyes0.96
  • Worth recommendingno0.35
  • Model replicano0.07
  • Problem scopescore on a 0–2 scale1.10
  • About Jevyes0.97

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 22 Sept 2026.