Science should be machine-readable

Structured scientific content can be indexed, compared, and analyzed at scales that prose-first publication makes difficult. Scientific results remain narrative, domain-specific, and hidden within article formats that do not make them explicit. OpenEval extracted 1,964,602 claims from 16,087 eLife manuscripts, and machine and human evaluations matched for 81% of co-evaluated results while the machine assessed a larger share of manuscript claims.