JevCode /カバレッジに戻る / ベンチマークと研究

ベンチマークと研究

このカテゴリの 16 件。各項目に具体的な判断と開ける根拠リンクがあります。

• Decision Models as Guardrails

  Treats typed decisions as guardrails rather than scores, using fixed options to constrain the action set downstream can execute.

  根拠https://github.com/raxITlabs/decision-models-as-guardrails/blob/main/infra/gcp/laya_server.py#L118

  ラベルpublic-sourceauthor-reported-metriccall-site-verified

• anandi1989/awesome-jev-usecases

  README states: ts2.tech: "TypeSafe AI Raises $40 Million for Jev, but Its 445× Cost Claim Is Still Self-Tested": *The skeptical audit of self-tested benchmarks, plus funding verification.*

  根拠https://github.com/anandi1989/awesome-jev-usecases/blob/main/README.md#L234

  ラベルreadme-claim

• caiovicentino/eikos

  README states: evaluation/eval_jev_suite.py and evaluation/eval_laya_suite.py run the same suites through Jev (API, evaluation

  根拠https://github.com/caiovicentino/eikos/blob/main/eikos/serve.py#L343

  ラベルcall-site-verified

• codeitlikemiley/system-one-adapter-rust

  README states: A drop-in replacement for TypeSafe's system_one evaluation API, backed by LLM

  根拠https://github.com/codeitlikemiley/system-one-adapter-rust/blob/main/src/client.rs#L15

  ラベルcall-site-verified

• erboland/jev-fund

  README states: jev-fund: Open-Source Paper Hedge Fund

  根拠https://github.com/erboland/jev-fund/blob/main/README.md#L32

  ラベルreadme-claim

• frankda/jev-poly-crypto-demo

  README states: This version implements a live market-data adapter, Jev integration, paper trading and a dashboard. It does not connect a wallet or submit real orders. TRADING_MODE=live is rejected at startup.

  根拠https://github.com/frankda/jev-poly-crypto-demo/blob/main/scripts/check-jev.ts#L2

  ラベルcall-site-verified

• fstandhartinger/jevbench

  README states: Jev is TypeSafe AI's decision model; JevBench also evaluates Jev-compatible systems. The canonical JevBench board is https://benchmarkheaven.com/jev-models.

  根拠https://github.com/fstandhartinger/jevbench/blob/main/README.md#L5

  ラベルreadme-claim

• grmkris/robo-harness

  README states: Decision runs (bun run jev) let an evaluation model or rules baseline pick bounded joint steps; chat and a decision run never move the arm at the same time.

  根拠https://github.com/grmkris/robo-harness/blob/main/apps/server/src/decision/strategies.ts#L273

  ラベルcall-site-verified

• kylemclaren/jevpdf

  README states: Cheap enough to ask about every line. Jev costs $0.042 per million input tokens, and output is free. Checking every line of a 15-page paper costs well under a cent.

  根拠https://github.com/kylemclaren/jevpdf/blob/main/server/jev-upstream.ts#L5

  ラベルcall-site-verified

• micytao/vllm-playground

  README states: Decision Models (Experimental) — A brand-new sandbox tab for structured-read decisions powered by DiffusionGemma and Jev / System One. Gallery of 12 use cases across 4 tabbed categories (Noul, Choice, Score, Combination)

  根拠https://github.com/micytao/vllm-playground/blob/main/README.md#L37

  ラベルreadme-claim

• nexibeo/jev-cookbook

  README states: Practical, tested recipes for TypeSafe's Jev, the fast decision model on OpenRouter. Fifteen real-world jobs, each with a runnable script, a small labelled dataset and measured results: support triage, database indexing,

  根拠https://github.com/nexibeo/jev-cookbook/blob/main/README.md#L5

  ラベルreadme-claim

• nickthompson480/typesafe-ai-playground

  README states: You need Python 3.10 or newer, a modern browser, and a TypeSafe API key to run evaluations. There are no packages to install and no frontend build step.

  根拠https://github.com/nickthompson480/typesafe-ai-playground/blob/main/server.py#L17

  ラベルcall-site-verified

• pedronauck/skills

  README states: jev-engineering — Design and build use cases, POCs and production systems with TypeSafe's Jev decision model (System One): design modes and use-case archetypes, decision state, typed questions, Jev combined with generati

  根拠https://github.com/pedronauck/skills/blob/main/README.md#L80

  ラベルreadme-claim

• theSekyi/jevusecases

  README states: A directory of real projects, filtered by what they replace, whether they've been benchmarked, and whether there's code to copy — not a marketing site for Jev itself.

  根拠https://github.com/theSekyi/jevusecases/blob/main/README.md#L5

  ラベルreadme-claim

• unicodeveloper/jevocks

  README states: Jevinik uses Jev, TypeSafe AI's System One evaluation model, to make typed, probabilistic decisions without calling a general-purpose generative LLM.

  根拠https://github.com/unicodeveloper/jevocks/blob/main/README.md#L11

  ラベルreadme-claim

• willkelly/jev-evaluation

  README states: An adversarial evaluation of jev, the decision model sold by TypeSafe. It

  根拠https://github.com/willkelly/jev-evaluation/blob/main/jeveval/config.py#L18

  ラベルcall-site-verified