Calibration & Research
Work on the models themselves and on the quality of their probabilities. Open-weight reproductions, inference tricks that read option probabilities from ordinary LLMs, calibration and threshold studies, and papers sit here.
264 public projects · page 9 of 9
Unofficial study reproducing Jev-style parallel typed decisions on small stock models on an Apple Silicon laptop, with benchmarks.
BenchmarkresearchPythonSmall one-pass option scorer with n-gram encoder, rival-aware attention and temperature scaling, benchmarked head-to-head against jevlike.
BenchmarkresearchPythonJev-inspired technique scoring single-token labels with an LLM's head via vLLM to fill typed fields, with a fast Mario demo.
BenchmarkresearchPythonTraining-free local decision layer fusing four readings from two frozen Qwen models, auto-releasing only when readings agree.
BenchmarkresearchPythonRepository on a direct approach to Jev-style models using Reinforcement Learning for Calibrated Decisions (RLCD).
ProjectClaude Code and Codex skills powered by Jev, published with a calibration curve measured on 4,995 real agent decisions.
ProjectPython trust middleware for the Jev API that logs decisions, measures calibration from your recorded outcomes and signs the evidence.
ServicePythonTurinTech GGUF builds of Clef, Clef-flash, pplx-decider and Kev for stock llama.cpp at /v1/systemone, at about 27-30% of full size.
ModelproductionLiquid AI's first decision model for text and images, served at a /v1/systemone endpoint callable with TypeSafe SDKs.
LibraryproductionvLLM Semantic Router's Apache-2.0 decision models (0.6B to 27B) answering up to 64 typed questions per input in one forward pass.
ModelbetaPythonSuperagent's Apache-2.0 security decision model from AutoJev-27B, scoring options for injection, tool calls, code and alerts.
ModelbetaSmall personal model trained in the spirit of Jev; few public details beyond a one-line description.
ModelPaper testing, without labels, whether Jev's probabilities obey probability axioms such as P(X) + P(not X) = 1.
PaperresearchPaper on where Jev and Laya fit in LLM pentest agents: finding adjudication, severity recalibration, pruning and confirmation loops.
PaperresearchPaper introducing an encoder-only System One model for Chinese with domain fine-tunes and the CJ-Bench benchmark.
PaperresearchPaper evaluating zero-shot Jev 1.13 on 37 datasets against Qwen and Gemma option probabilities, with code and raw responses released.
PaperresearchPaper testing Jev, Laya, Decider and Nimble against classifiers and LLM judges on prompt injection and risk screening.
PaperresearchPaper introducing PixelJev, a native-image decision interface on small open multimodal models with few-shot adaptation.
PaperresearchPaper introducing an adversarial benchmark for decision models: one appended opinion flips 12.1% of Jev 1.13 decisions.
PaperresearchPaper auditing 28 early Jev papers and proposing a 14-item checklist for evaluating typed decision models.
PaperresearchIndependent paper testing Jev's self-reported fields against ground truth, finding a collapsed verdict vocabulary and misleading flags.
PaperresearchOfficial TypeSafe docs page explaining how Jev's confidence is derived from the probability distribution and how to set thresholds.
ArticleproductionReddit announcement of a multimodal open decision-model family (0.6B to 27B) reporting first place on both Decision Index boards.
DiscussiondiscussionSocial post about a public Jev-shaped API served by an open Qwen3.6-35B-A3B model so anyone can try typed decisions.
Discussiondiscussion