Methodology

Before the first ranking

No public ranking methodology has been activated yet. Recent papers remain available in the catalogue while the first ranking is prepared.

Catalogue coverage

Scienced ingests arXiv metadata in artificial intelligence, machine learning, and computation and language. Full-text retention and assessment are limited to papers under CC BY, CC BY-SA, or CC0 licenses. The catalogue is a limited selection of arXiv, not a complete index.

The 7-day and 30-day windows use a paper's first submission date. Recent papers are listed newest first, and search matches their title, authors, or arXiv identifier. Catalogue entries do not imply an assessment or a rank.

How to read rankings

Published rankings compare papers within one category and time window. Each board links to its frozen model, prompt, rubric, comparison rules, and limitations. Those details will appear here when a methodology is published.

AI assessments can miss evidence or make mistakes. Use them to choose what to read. They are not peer review and do not establish whether a paper's conclusions are correct.