What it is
Explores zero-shot email classification with TypeSafe's pretrained Jev model, comparing ham/spam/phishing decisions against TF-IDF logistic regression baselines on main, fresh, and recent phishing sets.
How it uses Jev
Jev answers a Choice question selecting one of three categories (legitimate, spam, phishing) from an email state; the returned label and probability distribution are used directly by code. Enriched state adds Reply-To, link destinations, and attachment metadata.
Primitives:choice
Technique worth stealing
Preserve relevant email context (link destinations, Reply-To, attachment metadata) while keeping the question and category definitions unchanged.
Try it
Run experiments/jev-context/evaluate.py with an API key; see REPORT.md for details.
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-centricyes0.89
- Shows a System One patternno0.17
- Handles uncertaintyno0.06
- Measuredyes0.94
- Runnableno0.19
- Worth recommendingunclear0.56
- Model replicano0.06
- Problem scopescore on a 0–2 scale1.90
- About Jevyes0.98
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.