A research project on whether the literature a paper rests on actually says what the paper claims.
Relit is being built at ETH Zurich, in economics first. The near-term output is twofold: a benchmark for passage-level citation verification, and a system that scores well on it. The benchmark comes first on purpose — a tool of this kind is only worth using if its precision has been measured rather than asserted, and precision is what the design spends its budget on.
Beyond that, the same pipeline supports a question no one has answered at scale: across a field's working papers, how often does a citation fail to support the claim it is attached to?
Relit is the literature-facing member of that set: the others look inside a document or its code, this one looks outward at what it cites.
Elliott Ash, ETH Zurich — [email protected]
Interested in early access, in having a working paper run, or in the benchmark once it is released? Get in touch.