Answers are grounded in this paper and cite their sources.
Artificial intelligence (AI) is increasingly used to interpret experiments and propose mechanisms, but its mechanistic claims often lack traceable evidence. Here we introduce SKIER (Skills for Evidence-grounded AI Reasoning), a scientific skills AI framework that combines AI orchestration, researcher-defined analytical procedures, deterministic calculations, and versioned records. Using 49 records from a published methanol-oxidation case study, the framework recovered an m/z 48 signal exceeding 60% of the combined m/z 44, 46, and 48 signals in the H218O experiment, which was assigned to C18O18O. This observation supported the previously reported C–O cleavage pathway. AI then used the processed data to propose three competing mechanisms and design CH318OH-labelling, oxygen-exchange, and residence-time experiments to distinguish them. Those follow-up experiments further refined the published mechanism by locating C–O cleavage before formic acid formation. With shared skills, all 11 AI configurations completed data processing, compared with none without them. In the 100 expert-designed questions evaluating mechanism inference and validation-experiment design, GPT-5.6-sol achieved 100% accuracy. Together, these results demonstrate how a scientific skills framework connects reproducible data analysis with experimentally testable mechanistic reasoning across AI models.
No takes yet. Share an insight, caveat, or question.
Meng et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: