ZCode PreToolUse guardrail built on the shared jev-evaluator core, combining deterministic rules and TypeSafe Jev for allow, ask, and deny decisions.
Research and benchmarks — page 14
Benchmarks, papers and independent reproductions. Read these when you need to understand behaviour rather than adopt a library.
537 active projects in this collection. All collections →
Use Jev instead of an LLM for RAG decisions: relevance grading, query routing, hallucination checks and LLM-as-a-judge evals. Faster, cheaper, measured first.
Version-aware evidence decisions for RAG assistants using Jev, with inspectable routing, bounded recovery, and reproducible evaluation.
A quality gate for AI-generated requirements, powered by Jev (TypeSafe System One). Routes each requirement to PASS / REVIEW / BLOCK with calibrated probabilities. Real benchmark vs LLM: 0 false blocks, 9x cheaper.
Run only the tests your change touched: import graph + ownership rules first, Jev decides the ambiguous rest, fail-safe to the full suite. Replay benchmark with real ground truth.
0
Open emulator of TypeSafe's Jev API on open models, plus a paired benchmark of local and hosted models against Jev
0
Bobcat: a typed-decision model. State in, Choice / Noul / Score out, with a probability for every answer you name; no generated text. Compiler, server, training and evaluation code for Bobcat 1.1 and Bobcat Flash 1.1.
0
JEV Studio is a small Next.js evaluation lab for turning natural-language input into structured signals with the Typesafe SystemOne API.
0
Java 21 client for TypeSafe's Jev evaluator: typed questions, typed answers, no decisions made for you
Two-stage session and passage retrieval with optional Jev reranking
RAG evaluation with typed decisions: sentence-level hallucination verdicts with offsets and citations, passage and answer relevance, in one Jev call per answer. Benchmarked against Ragas, DeepEval, an LLM judge and HHEM.
Response validation, observation and evaluation utilities for the official TypeSafe SDK.
0
kblam: the knowledge base for LLM-assisted mereology. it's a flat-file research/knowledge storage system with enforced organization (using heuristics, semantic similarity, and Jev evaluation) to prevent document drift, internal contradictions, incorrect data gleaned from reading superseded sections out of context, and other LLM bullshit. not done
0
Sev-X: Kev-descended decision models (DDP, eval context, SRA program)
Measured experiments on decision models. Can an open-weight model re-rank search results? Does a token budget explain a benchmark failure? What does one typed decision cost across model families? Every request and response is recorded, and negative results are published as they came out.
0
A tree housing every closed question humans or machines ask, answered by Jev (TypeSafe System One): capabilities, defaults, and jaggedness. Not a benchmark.
Roteamento automático de incidentes para o time certo usando o modelo Jev (TypeSafe AI) via OpenRouter, com benchmark de 1000 requisições.
Benchmark TypeSafe's Jev (System One model) vs Claude on structured decisions: accuracy, latency, calibration, cost
面向语义决策模型的中文言下之意评测集：100 道伴侣对话 Choice 题，支持本地 ONNX、Jev 官方 API 与内网模型对比
Evaluate cloud spend against a budget and gate CI with Jev (approve, warn, block, or review).
How far can a small model go? Testing TypeSafe's Jev on 1,352 cases against large LLMs: 98% accuracy, ~200x cheaper than Claude Opus 5.5
Jev (TypeSafe) as the decision layer for Claude Code: permission gate, skill routing, conditional instructions, effort-tiered subagents, context packs. Tested PoC with evals.
Reproducible Jev benchmark on Klondike Draw-1 with baselines, decision traces, latency metrics, and replay viewer
0
Test bench for TypeSafe AI's Jev evaluation model on Vercel AI Gateway