What it is
A Python tool that labels every sentence of a document with calibrated probabilities from Jev (TypeSafe AI's System One model) and renders the results as a heatmap for whole-document viewing.
How it uses Jev
Each sentence is sent in one HTTP request carrying five or six typed questions (Noul, Score, Choice) from a preset. Jev evaluates them in parallel against the same state. Answers are written per sentence to JSONL and used by the viewer to color lanes or a wall grid.
Primitives:choicescorenoul
Technique worth stealing
Hiding Jev's prediction during manual validation to avoid anchoring when measuring calibration.
Try it
git clone && cd semantic-microscope && uv sync && cp .env.example .env && uv run microscope --help
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-centricyes0.93
- Shows a System One patternyes0.75
- Handles uncertaintyno0.04
- Measuredyes0.72
- Runnableunclear0.55
- Worth recommendingunclear0.47
- Model replicano0.04
- Problem scopescore on a 0–2 scale1.00
- About Jevyes0.97
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 21 Sept 2026.