Some of the most valuable problems share one shape: many local sources — each partial, each competent in its own corner — that must combine into one global answer you can trust. Sheaf is the engine for exactly that, and it tells you where the sources agree and where they don't.
Wherever you gather independent views — analysts, models, agents, documents, datasets — you face the same two questions: what is the one coherent answer, and where do the sources quietly contradict each other? Most systems give you the first and hide the second. Sheaf measures both — a single answer, plus the confluence (where they cohere) and the disparity (where they don't), made explicit and scored.
| Local sections | each source's own view — a model, an analyst, a document, a dataset |
| Gluing | assembling those local views into one coherent global answer |
| H⁰ — the confluence | what the sources agree on — your high-conviction core |
| H¹ — the disparity | the irreducible disagreement on what they share — scored, not hidden |
Finance is the first vertical we've built out — proven and live. The same engine is general by design: it already runs in production today, and the same machinery extends to the most demanding professional domains next. Here's what's live versus what's on the way.
Live today
Many models answer → one synthesized answer plus a number for how much they actually agree. Already live across a family of Sheaf products.
The engine already powers products you can use now — Sheaf Search, Sheaf Mind, Six Opinions, and more. Proof it travels.
On the roadmap
Many papers, sources, or experts → one defensible finding, with every contradiction surfaced.
Clauses, precedents, reviewers, policies → a coherent position and the exact points of conflict.
Multiple reports and sources → one assessment, with confidence and dissent kept visible.
Disparate documents and signals → one verdict, with the disagreements that deserve a second look.
Local stores that should agree → a unified whole and a map of exactly where they don't.
Our flagship is a five-discipline markets desk — Fundamental, Macro, Quant, Technical, Risk — that answers a market question and scores exactly where the analysts agree and where they split. Finance is the most demanding version of the problem: high stakes, genuine disagreement, and a hard requirement to show your work. See the desk run live →
Sheaf is built on sheaf theory — the branch of mathematics for how local data glues into a consistent global whole, and how to detect when it can't. It isn't a metaphor we reached for; it's the founder's own field. Mathematical rigor isn't our marketing — it's our training. Read the manifesto →
Where others route, we glue — and we measure the seams.
Sheaf