jevbooks

← All projects

jev-fanout-bench

blowxian/jev-fanout-bench · Homepage

Measured: batching Jev questions bills state once, saving up to 86%.

Evaluation & Benchmarking81%Runner-up: Calibration & ResearchQuestionnaire

What it is

A reproducible billing study of TypeSafe Jev, measuring token costs and answer stability for batched versus separate questions across state sizes and question counts, using synthetic support tickets.

How it uses Jev

Jev answers Noul, Choice, and Score questions on synthetic support ticket states. The study compares billing and answers when N questions are asked in one call versus N separate calls, using usage.input_tokens and response values.

Primitives:choicescorenoul

Technique worth stealing

Derive fixed per-request overhead F independently at each N and question subset to verify exact linear billing.

Try it

export TYPESAFE_API_KEY=...; python bench.py run; python bench.py report

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-centricyes0.62
  • Shows a System One patternunclear0.59
  • Handles uncertaintyno0.03
  • Measuredyes0.63
  • Runnableno0.07
  • Worth recommendingno0.34
  • Model replicano0.02
  • Problem scopescore on a 0–2 scale1.01
  • About Jevyes0.99

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