Curate a daily science edition with typed labels

01 · Role in the system

What Jev does here

Each morning the ingest pipeline pulls new papers from arXiv, bioRxiv, medRxiv and 58 leading journals, and a batch Jev call scores them: one Noul question per paper against the reader-interest state, model jev-latest, with the ranking rule published as method.json. The edition applies editorial constraints in code - at most three papers per field, preprints marked as not yet peer-reviewed, abstracts that claim more than they show flagged and excluded - and every entry quotes the one abstract sentence stating the main result exactly as the authors wrote it. Because Jev only chooses among options and returns calibrated probabilities, the site argues, it has nothing to hallucinate: the machine picks and labels, scientists speak. The web app exposes per-paper questions and probabilities, an open-data JSON feed, a Spanish edition, and a CI-backed codebase.

02 · Control boundary

Where Jev sits

Jev as the labeler and ranker of an editorial pipeline: code enforces the publication rules (diversity seats, preprint flags, verbatim quotes), Jev's calibrated Nouls decide what makes the edition, and every question and probability is published for audit.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

• Editorial constraints (field caps, overclaim flags) are the author's policy choices, not benchmarked.
• Journal list and ingest scope are fixed by the operator; coverage depends on source APIs.
• Single-operator publication; labels are not independently evaluated against human editors.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

• ferced ↗Community · github · public · checked 2026-09-28