How anything relates to anything

One graph, several grains, one question. Give it two things and it returns the chain of typed relationships between them — or an honest answer that no recorded path exists.

One model, not four

Concepts, code, anatomy and architecture are very different data, and the tempting design is four viewers over four models. That is the mistake this product is most likely to make: the moment a second model exists, a path means one thing here and another thing there, folding gets written twice, and the two drift apart in ways nothing checks. So a grain does not define relationships. It maps its own onto nine shared kinds — contains, part of, a kind of, depends on, used by, flows to, flows from, related to — or the relationship is dropped and counted.

Nine kinds, on purpose

The question a reader actually asks is coarse: does this contain that, does it depend on it, does something flow between them. A vocabulary of sixty relationship types is one nobody can filter by, and a filter nobody uses is the same as no filter at all.

An empty answer is an answer

When no path is found, that is what you are told. Two things can be genuinely connected in the world and have no recorded chain of edges in a dataset — reporting the gap is the only honest option, and inventing a plausible route would be the single worst thing this product could do.

What it will not pretend

A grain with no dataset says so instead of rendering an empty explorer. A graph too large to draw says it folded and how many nodes it stands for, instead of quietly showing you a smaller graph and letting you believe it is the whole one.

Questions

Do I need an account?

No. Exploring every public grain, tracing paths and exporting a mermaid diagram are free and anonymous. An account is for putting your own repository or your own dataset into it.

Which grains can I explore today?

Architecture is published here in full. Anatomy is traversable but its pages live on body.leumas.tech. Code is a member surface. Concepts needs a licensed dataset first, and the page says so.