A score is a claim until it can be repeated.
What belongs beside a benchmark result, and why a submitted number is only the start of the record.
The archive journal
Read beyond the result.
Editorial notes on models, measurement and what the evidence supports. News will be published with sources and corrections.
The number beside VRAM needs a footnote.
Weight size, context length and runtime overhead belong in the same deployment record.
A catalogue with room for uncertainty.
The editorial commitments behind an independent record of open-weight AI.
This first edition contains notes on our methods. No news reporting has been published yet.