The framing takeoff's sheathing code has had a property test for a while. The layout engine, which decides where every stud, king, jack, header and cripple goes, did not. engine-fuzz.test.mjs now draws 400 seeded random two-wall plans (lengths from 6 to 36 ft, 16 or 24 in spacing, 2x4 or 2x6 studs, up to several doors and windows per wall) and requires that each plan is either rejected by validatePlan or laid out with an empty verifyWall problem list and a successful takeoff.
Result: more than 200 of the 400 plans were valid, and none produced a problem. The first run did fail, but only because the test's openings had no labels and verifyWall finds a header by its opening's label. The test now labels them.
What this does not show: that a layout is structurally adequate (the engine never claims that), or that it matches a real builder's plan. It shows the layout self-check holds on a wide spread of inputs, which is the property the cut list and order depend on.