Slides of the talk given at the FORRT AI in Metascience Online Conference (session T1, Reproducing the record), 29 September 2026. Reproducible is not robust: rerunning the authors' code on their data returns their answer. We replicate the specific claim we rely on, with independent data and software, using AI agents constrained by guardrails: a FORRT replication template, MCP tools (replication-radar, forrt-research-mcp), FAIR vocabularies and research-software standards. Every step, from the verbatim quote to the verdict and the typed citation, is published as a signed nanopublication and can be searched on the Science Live platform. Worked example: Oliver et al. 2018, marine heatwaves (validated for the satellite era). Scale: 47 claims from 19 papers. Limits: computational work only. The last slides list every link used in the talk.
No takes yet. Share an insight, caveat, or question.
Fouilloux et al. (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: