ProtoStar puts scientific data reuse to a practical test
The initiative links open research publishing with challenges designed to reveal whether scientific datasets, models, and code are actually reusable.
Project ProtoStar puts a specific hypothesis about open science under examination: publishing datasets, models and code in AI-usable formats may make research easier to reuse, but availability alone does not establish that reuse works. Announced on October 29, 2026, the initiative brings together universities, research organizations and open science partners, with Hugging Face as its infrastructure partner.
Open challenges are intended to reveal where reuse succeeds and where documentation or compatibility gets in the way. Carnegie Mellon University hosts the opening climate challenge from November 2–13, 2026. Participants predict urban weather conditions using historical records, with performance assessed against straightforward statistical approaches, including historical averages and linear trends. Final forecasting results are scheduled for early 2027, after the required weather records become available.
The climate challenge can test whether an approach improves forecasts over historical averages and linear trends, provided that provenance, documentation and the division between training and evaluation observations are clear. That comparison concerns predictive utility on the challenge data. Until the forecasting results are available, the announcement leaves both the advantage over simpler methods and the practical reusability of the materials unresolved.
Source: Avijit Ghosh and coauthors, “Announcing Project ProtoStar: Making Publicly Funded Science AI-Ready”.