Are the Cochrane recall and daily cost figures actually measured?

The digest reports screening 19,447 records across four unseen Cochrane reviews, reproducing 96.9% of included studies while removing 78% of reading for $0.60, with per-run tokens and cost logged to data/runs.jsonl.

Why not stick with keyword alerts or an LLM digest bot?

Keyword alerts miss different wording, embedding recommenders drop dissimilar past reading, and chat LLMs over the full firehose cost more. One judgement per interest costs a fraction of generated text, so it can score every paper.

My interests are vague — will check and calibrate actually help?

paper-radar check warns about multi-idea interests and count or date questions, and crisp single-idea wordings produced confident hits where vague ones never crossed threshold. Calibrate then fits thresholds to your labels.

Do I need a server to read a radar or to run my own?

No server for reading: seven hosted radars expose plain RSS links with no key. Running your own is a fork with radar.toml interests, a stored API key, Pages sourced from Actions, and weekday runs after the arXiv announcement.

What exactly does Jev get to read for each paper?

Only the title, the abstract and the categories. Author names and affiliations are deliberately left out because they do not bear on topic. Every question for a paper goes into a single call, so more interests add tokens but almost no latency, and output tokens are free, which is what makes reading all 1,500 papers in a weekday announcement affordable.

What does a full run over arXiv actually cost?

A measured run on 2026-09-23 judged 50 papers in 5 seconds for 46,584 input tokens and $0.0020, 932 tokens per paper on jev-1.13.0, which works out to roughly $0.06 a day for the whole weekday announcement. Every run logs its own tokens and cost to data/runs.jsonl, and paper-radar check estimates your profile and cost before you spend anything.

Do I need a TypeSafe key, or can I use OpenRouter instead?

Either. Set TYPESAFE_API_KEY from console.typesafe.ai, or OPENROUTER_API_KEY if you are still on the TypeSafe waitlist, in which case radar.toml sets backend = "openrouter" and model = "typesafe/jev-1.13". To see the shape of the output first with no key at all, paper-radar demo writes a sample site from fictional papers using a keyword heuristic in place of Jev.

CommunityJev Tools & CLIs

JEV-Paper-Radar

Forkable weekday radar that asks Jev one yes-or-no question per interest over each new arXiv announcement and publishes a page plus RSS feed.

See highlights

Open github.com

Stars: 30
Language: Python
License: MIT
Published by: Community
Author: Eliot5566
Last verified: Oct 2026

Eliot5566/JEV-Paper-Radar

User repository on GitHub

View on GitHub

Let Jev read every new arXiv paper each morning and surface the few you should read. Plain-English interests, calibrated probabilities, ~$0.06/day, fork and go.

Forks: 11
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

Jev reads every new paper on arXiv each morning. You read the few that matter. Plain-English interests · calibrated probabilities · about 6 cents a day for all of arXiv · fork and go, no server. Quick start · Try the demo · How it works · Benchmark · 繁體中文. A real run: 50 new arXiv papers judged by Jev in 5 seconds for $0.002. No key? paper-radar demo shows the same page offline.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification Jev for Ranking

Used inJev for Software & DevTools

Languages: Python

Highlights

Judgement instead of prose: per-interest probabilities your code thresholds with max or noisy-OR, plus exclusion handling.
Daily site with must-read and maybe lists, feed.xml for readers like Zotero, optional chat digests, and audit trail in data/.
paper-radar check validates radar.toml and estimates cost while calibrate fits thresholds from your own yes or no labels.

Watch out

Requires a TypeSafe or OpenRouter key with GitHub Actions and Pages enabled; the local demo uses a keyword heuristic instead of Jev. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

hermes-nerve

Async supervisory layer for Hermes Agent that locks Definitions of Done, proves mechanical criteria locally, and calls Jev only for unresolved semantic checks under a token budget.