Skip to content
Schematic
Docs Get access

How it works

Proof,
not prediction.

AI can search. Schematic certifies only results that pass a deterministic mathematical check.

What Verified means

What Verified means

Every guarantee
comes with proof.

Search may be probabilistic. Certification is not. Every Verified result records exactly what was checked, under which assumptions, and against which revision.

Lean checks the proof.

Lean is a programming language built to express and check mathematical proofs. It accepts a proof only when every step follows from explicit rules and stated assumptions.

THE BOUNDARY

Search may fail to find a proof. It cannot make Lean accept one that does not check.

VERIFICATION CERTIFICATE Verified
Restricted actions require approval
VERIFICATION RECORD cert_a31 CHECKED
Coverage
All entry points
Revision
9f72c1
Assumptions
4 recorded
Checked by
Lean 4
Mathematical proof accepted by Lean Revision 9f72c1

The certificate records exactly what was proved, under which assumptions, and for which software revision. The result stays precise, reviewable, and tied to the system that was checked.

The result never drifts away from the software.

Each certificate names the revision it covers. If relevant code or assumptions change, Schematic re-establishes the claim for the new revision and issues a new certificate.

Don’t take our
word for it.

Real software, concrete counterexamples, and results you can inspect.

OPEN-SOURCE SUPERTEST STUDY

Nine libraries.
135 generalized laws.

Across pinned versions of open-source C, Rust, and Python libraries, repeated test families were consolidated into stronger laws while boundary and regression tests stayed explicit.

PINNED LIBRARIES9C · Rust · Python
GENERALIZED LAWS135Across selected test families
FEWER TEST LINES17,87235,468 → 17,596
MEASURED REDUCTION50.4%With boundaries retained
WHAT WE ARE MEASURING NEXT Evidence that tests the promise.
REPRODUCIBILITY

Results tied to what was checked.

Measure whether every issued guarantee remains tied to its recorded claim, assumptions, and software revision.

COUNTEREXAMPLES

Failures missed by existing checks.

Measure exact counterexamples found beyond unit tests and AI-only review.

ACROSS REVISIONS

Guarantees maintained through change.

Track claims re-established and regressions caught as the software evolves.

RESEARCH RESULTS

Published work, formally checked.

Connect results to their original paper, exact statement, and verification record.

MEASURED, NOT ESTIMATED Benchmarks include the dataset, baselines, and method alongside the result.

What would you prove?

Explore Quill Explore Pup