jevbooks

← All projects

S1Rank

zaesho/S1Rank

Benchmarking Jev, a System-One decision model, as a reranker.

Scoring & Ranking100%Runner-up: Classification & RoutingQuestionnaire

What it is

S1Rank evaluates whether Jev, a non-generative model answering typed questions with probabilities, can rerank BM25 top-100 candidates for information retrieval. It provides a benchmark, raw responses, and a paper for researchers.

How it uses Jev

Jev answers Noul yes/no questions per document in one request holding all 100 candidates (joint pointwise), producing a probability per document used to rank. Also tested: isolated pointwise, graded Score, Choice-best, Peel, Tournament, and S1Adaptive routing.

Primitives:noulchoicescore

Technique worth stealing

Joint pointwise: one request with all candidates and one Noul question per document.

Try it

python scripts/fetch_data.py && python scripts/analyze.py

View on GitHub

judged by Jevjev-1.13.0

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-centricyes0.94
  • Shows a System One patternyes0.62
  • Handles uncertaintyno0.11
  • Measuredno0.16
  • Runnableno0.06
  • Worth recommendingno0.33
  • Model replicano0.04
  • Problem scopescore on a 0–2 scale1.09
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 23 Sept 2026.