jevbooks

← All projects

SemIf-OpenJev

TheoLeeCJ/SemIf-OpenJev · Homepage

Semantic ifs from open models, on a 3090 at home.

Agent Decisions95%Runner-up: Calibration & ResearchCheck-up

What it is

SemIf (formerly OpenJev) is an independent replication of Jev's typed-decision interface using open models. It reads option probabilities directly from a 4B model, avoiding text generation, for runtime-defined semantic decisions.

How it uses Jev

Jev's interface pattern is reproduced: given a state and typed options, the model returns calibrated probabilities for each option. The result is used as a direct if condition in code, without parsing generated text.

Primitives:choice

Technique worth stealing

Direct logit readout from a frozen model for typed decisions, avoiding autoregressive generation.

Try it

Run the browser demo or use the Python scorer with a 4B model on CUDA.

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-centricyes0.83
  • Shows a System One patternyes0.76
  • Handles uncertaintyno0.05
  • Measuredno0.15
  • Runnableunclear0.54
  • Worth recommendingno0.27
  • Model replicayes0.93
  • Problem scopescore on a 0–2 scale1.15
  • About Jevyes0.95

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 23 Sept 2026.