jevbooks

← All projects

jev-orderby-bench

yodablocks/jev-orderby-bench

Independent measurement of whether ORDER BY over Jev probabilities is defensible.

Scoring & Ranking53%Runner-up: Evaluation & Benchmarking

What it is

A benchmark harness measuring whether SQL ORDER BY over Jev probabilities yields a defensible order, for developers building semantic sort on Jev. It reports pairwise inversion, Score ordinality, calibration, wording invariants and sort-key ties against pre-registered gates.

How it uses Jev

Jev supplies the probabilities that are sorted: jev_bool, jev_choice and jev_score are called on 360 labeled rows and on Amazon ESCI query-product pairs, and the returned values are measured for inversion, ordinality, calibration and ties rather than used to make a product decision.

Primitives:choicescorenoul

Technique worth stealing

Pre-register gate thresholds, then add a paraphrase control so a negation result cannot be overclaimed as negation-specific.

Try it

python3 harness/run_calibration.py with a key; --analyze-only recomputes from cache.

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-centricyes0.81
  • Shows a System One patternno0.26
  • Handles uncertaintyno0.04
  • Measuredyes0.94
  • Runnableno0.09
  • Worth recommendingno0.29
  • Model replicano0.05
  • Problem scopescore on a 0–2 scale1.09
  • About Jevyes0.99

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.