jevbooks

← All projects

jev-sec-bench

Gaurav-Gosain/jev-sec-bench

Blind security benchmarks for Jev on prompt injection and vulnerable code.

What it is

A Go benchmark suite that evaluates Jev, TypeSafe's System One model, on two public corpora: 662 prompt-injection messages and 200 matched vulnerable-code pairs. It reports classification metrics, calibration, and a label audit, with a TUI dashboard.

How it uses Jev

Jev answers a yes/no injection question and a severity Score in one request per sample, using the deployment context as state. The returned probabilities are thresholded at 0.50 for the reported metrics; the README suggests a graded policy with 0.70/0.35 cuts.

Primitives:choicescore

Technique worth stealing

Pass the deployment context as state so the same model separates traffic at a fixed threshold.

Try it

export TYPESAFE_API_KEY=...; go run ./cmd/jev-sec-bench -bench all

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-centricyes0.92
  • Shows a System One patternno0.18
  • Handles uncertaintyyes0.64
  • Measuredno0.28
  • Runnableno0.19
  • Worth recommendingunclear0.47
  • Model replicano0.04
  • Problem scopescore on a 0–2 scale1.62
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 21 Sept 2026.