Engineer

Open ZTL Engineer → engineer.ztljudge.com (a public demo on a small server: a few questions an hour per visitor; the heavy tasks — large circuits and proved truss ranges — are off)

A technician asks, in plain words and in their own language, whether a pump or a solar kit will do its job, and attaches what they have: a photo of a nameplate, a datasheet, a price list. A model (Meta's Muse Spark) reads it and writes a formal design problem — quantities with tolerances or ranges, physical laws, requirements. The ZTL stand then checks that problem over the whole ranges: it proves the requirement holds for every value in them (EARNED), refutes it with a concrete failing point (REFUTED), or says it cannot decide either way (OPEN).

The proof is about the problem as the model wrote it. So every value the model read is listed with where it came from, the person can correct any of it, and the problem file runs again on the open-source stand with no AI and gives the same verdict. The guarantee covers the problem as written — the laws the model chose, the values it read — not the physical pump. It is a checkable proof, not an engineer's stamp.

Source: github.com/inventor1975/ztl-engineer. Built for the Meta Global AI Developer Hackathon (2026).