Introducing Atlas.
Describe what you are stuck on. The answer has two halves, your own literature and the work from outside your field that runs on the same mechanism.
Join the waitlistYour fieldwhat you already have
Outside itwhat you would never have searched
Your field has a way of doing this. So does a field you have never read.Atlas gives you the difference.
Built for people who already know their own literature.
What is indexed
One index, whatever the record is.
A paper, a patent and a person’s history describe the same mechanism in three unrelated vocabularies. All of them come apart the same way before anything is stored.
Papers
113,394 abstracts, and full text where we can reach it.
Indexing
Patents
Claims, and the prior art cited against them.
Planned
People
Authorship and employment history, including the names that appear in two fields at once. There are never many, and they are not usually looking.
Planned
News
Trade and industry reporting. Most of it reports a result and never says why it worked.
Under review
What is searched
Every source is decomposed along two domain-neutral axes before it is stored. A query is decomposed the same way, and the match happens between those decompositions.
What comes back
One result, in full.
Each result is a mechanism, the conditions it ran under, and the distance from those to yours. The engine does not tell you which to try. This is the result the screen above came back with, in full.
Mechanism
Small perturbations applied at the moment a cycle repeats hold an unstable orbit on a regular one, instead of forcing the whole system at once.
Demonstrated in
A gravitationally buckled magnetoelastic ribbon, driven at laboratory scale, at room temperature, with the timing computed from a measured return map.
Gap to your conditions
Yours is cardiac tissue in a living animal at body temperature, where the timing has to come from the interval between beats rather than from a return map you can measure in advance.
Evidence
Physical experiment, published 1990, with the effect observed rather than modelled.
No candidate is recommended for adoption.
That rank is a measurement. Given the problem above, as a cardiologist would have written it before anyone had connected it to physics, the engine returns this paper at rank 7. A leading AI search returns it at no rank at all, as it did on all 43 cases we scored, although a search on its title pulls it out of that index. Two years after that paper, Garfinkel, Spano, Ditto and Weiss did the same thing in a rabbit heart. Spano and Ditto are on both papers. Two people carried it across by hand.
Using it
Four things it does that a search box cannot.
Write out what is not working
Paste a paper, or describe what is failing in your own words.
Results split by distance
Matches from your own field and matches from far outside it are shown apart.
Every claim traceable
Each connection carries the passage it came from. Follow it back to the source paper in one click, or dismiss it.
Your problem stays yours
For classified and proprietary work, only the abstract causal signature travels. The application, the context and the data stay behind your firewall.
Who uses it
Three ways to read the same index.
ReadsPapersPeople
Chemicals and materials
R&D directors and program leads
A process that will not scale. A coating that fails in the field. A reaction that needs a catalyst nobody in the industry has.
Defense and the national labs
Program managers and technical directors
Requirements with no answer inside the domain, on work that cannot leave the network. Only the abstract signature is sent.
Pharmaceuticals
Discovery, and search and evaluation
A molecule that failed for reasons unrelated to its mechanism. A delivery problem another field solved decades ago.
Work is scoped one problem at a time.