What it is
A benchmark comparing TypeSafe Jev against Cohere Rerank 4, ZeroEntropy zerank-2, and a chat-model baseline for passage reranking, with raw API responses, scoring code, and uncertainty intervals.
How it uses Jev
Jev ranks 30 BM25 candidates per query using a 4-level rubric, yes/no questions, or a Choice+none setup. Its scores are used to order passages and compute nDCG@10, Top-1, and other metrics.
Primitives:choicescorenoul
Technique worth stealing
Using a decision model with typed questions for reranking, compared against dedicated rerankers.
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-centricno0.28
- Shows a System One patternno0.23
- Handles uncertaintyno0.04
- Measuredyes0.95
- Runnableno0.05
- Worth recommendingno0.30
- Model replicano0.06
- Problem scopescore on a 0–2 scale1.36
- About Jevyes0.98
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.