undiscovered

Research in one field that another field could use, but has never cited.

Take partSame problem, different fieldFirst testCode and data

In 1986 Don Swanson read two sets of papers that had never cited each other. One showed that dietary fish oil lowers blood viscosity, platelet aggregation and vascular reactivity. The other described Raynaud's syndrome, which involves all three. He proposed fish oil as a treatment, and a small double-blind trial in 1989 found that it helped patients with primary Raynaud's. He called this undiscovered public knowledge.

This project looks for such connections in every field of science, and publishes how often its predictions turn out to be right. Fields often describe the same problem in different words, so volunteers lend their own computer and whichever AI model they prefer, on their machine or free in the cloud, to rewrite papers without their field's jargon. The result is an open index of papers described in plain terms, so that work on the same problem in different fields can be matched.

Take part: lend your computer and an AI model Every paper is described by two volunteers' models, compared in public.
First result, October 2026: inconclusive. We froze the literature at the end of 2017 and ranked 28,114 pairs of topics from different domains that had barely cited each other; 305 of them started citing each other by 2023. The citation network around each pair ranked them about 14 times better than chance. Adding what the papers say, measured by the words they share, did not measurably improve on it, so under the plan written before the analysis the method is not taken further as it is. On a second, independent set of 300 topics the network ranking did as well, and adding the words was again inconclusive. See the results.

Forward test, registered 12 October 2026. The same network model ranked 26,475 pairs of topics at the end of 2025. Its checksums are anchored in Bitcoin block 971017 (01:11 UTC on 12 October 2026), before any of the citations that will score it exist, and it is scored on the citations of 2026 to 2031. The plan and the ranking.

Help now

Lend your computer and an AI model. Your model, local or free in the cloud, describes papers in plain words so that work on the same problem in different fields can find each other. Take part, and see the progress so far; the pairs of papers the descriptions bring together appear on Same problem, different field as the work comes in.

Find a flaw. Read the plan for the forward test, which ranks pairs of fields now and scores the ranking on citations not yet written, or the first test's results, and tell us what could make a result wrong. Criticism of a plan is worth most before its data is collected.

Coming

As each part ships, you will be able to: