Does a TypeSafe Jev rerank beat embedding search? Graded relevance eval (9,831 pairs, 164 zh/en queries) over the Agent Skills Hub catalog, with the judge-circularity bias measured.
The link points at the commit we read, so the line number still holds.
This one cannot run here
Its question set is assembled at runtime, or never written out literally in the code, so there is nothing to lift. The source link above will show you.
Other projects in this category
jev-playground ▶ — TypeSafe AI の System One モデル Jev を MoonBit から触るためのプレイグラウンド。
eutrya ▶ — Jev-native AI security harness for autonomous research, multi-agent swarms, persistent hunt boards, and long-running agent workflows. CLI-first, open source, and built for authorized security research.
typesafe-local ▶ — Inspired by TypeSafe Ai, Ask a local LLM typed questions, get calibrated probabilities instead of text. Structured output without generation or parsing. MLX / Apple Silicon.
jev-mcp ▶ — Connect JEV to MCP clients and compare its judgments against general-purpose LLMs using shared datasets and measurable accuracy.
omp-jev ▶ — TypeSafe Jev routing for Oh My Pi, with an opt-in checkpoint orchestrator and editable XDG configuration. Requires Bun = 1.3.14 and OMP = 18.2.3.
jev-chat ▶ — A chatbot from typed Jev decisions: hierarchical speculative decoding over System One probabilities.
jev-browse ▶ — jev-browse is an unofficial project and isn't affiliated with TypeSafe or Vercel.