Sealed core, public evidence

Most software earns trust by showing its source. We take the opposite route, on purpose. Every instrument in the lab keeps a sealed core and a fully open evidence trail. That is not secrecy for its own sake — it is the only posture that lets a small team make strong, patent-pending claims without handing competitors the mechanism, and without asking buyers to take anything on faith.

The principle is simple: reproduce, don't believe. We do not publish algorithms, formulas, or wire formats. We publish what can be checked — ratios, pass/fail conformance against public test suites, lossless round-trip verification, and the conditions under which each number was produced. A claim you can reproduce on your own data is stronger than a codebase you must read and trust.

That is why access works the way it does. Instead of a public binary, evaluation runs on your corpus, with your baseline, and you verify the result locally. The core never leaves our side; the evidence comes to you. Numbers are labelled, never blended, and limitations are stated on the same page as the strengths.

This costs us the easy credibility of an open repository. We accept that trade. For the buyers who matter — infrastructure, security, research — a precise, reproducible, modest claim ages better than an impressive, vague promise.