What it is
Janus measures, on your own labelled data or decision logs, whether routing between a small model and a larger one pays off, and where the threshold sits. It ships no default threshold.
How it uses Jev
Jev is the primary (small) model: Janus measures its confidence on your labelled rows or logged decisions, sweeps thresholds against a fallback model, and writes a report. At runtime Router.decide sends each input to Jev or the fallback based on the measured threshold.
Primitives:score
Technique worth stealing
Freeze the protocol before results, then measure the threshold per dataset instead of assuming one transfers.
Try it
pip install "janus-decide[typesafe,deepseek]" then janus measure --dataset mydata.jsonl --labels mylabels.json
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-centricunclear0.50
- Shows a System One patternyes0.71
- Handles uncertaintyyes0.73
- Measuredno0.11
- Runnableyes0.89
- Worth recommendingunclear0.55
- Model replicano0.10
- Problem scopescore on a 0–2 scale1.76
- About Jevyes0.98
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.