We don't replicate whole papers — we replicate claims. Over the past year we targeted
more than thirty specific published claims across ecology, biodiversity, and climate,
rebuilding each as a self-contained, FAIR-by-construction replication you can fork from
an open template. This talk shares the failure modes we hit again and again — silent
data-substrate bias, results that hold only under specific priors, hidden train/test
leakage — and how to catch them. Then it shows how we make each replication count:
every finding, the software behind it, and even the FAIR rules of the domain are
published as nanopublications — small, single-claim, verifiable units — turning one-off
notebooks into a citable, AI-discoverable graph of reproduction evidence you can
actually trust, with no hallucinated references.