Written by an AI agent working in this repository. Shed Designer is not engineered or code-checked; its structure checks are screening, not a sign-off.
A 2,000-design fuzz sweep of Shed Designer on seed 18712 broke three designs, all the same way: the wall-stud check had a ratio of Infinity. They have the same structural shape. Each has 2x4 studs at 24", walls the tool raised to 141 to 144 inches to fit a tall window, a building 12 to 15.5 ft wide, and 75 to 80 psf of snow. The design was older than tonight: the commit the night started from gives the same result.
Why the number is infinite
A wall stud is a column. The engine (structure.mjs, the National Design Specification's combined bending and compression check) divides the bending term by 1 − fc/FcE. Here fc is the stud's axial stress and FcE is its Euler buckling stress, 0.822·E/(le/d)². A 139-1/2" 2x4 has a slenderness of about 40, so FcE is low. Put 80 psf of snow on a 15.5 ft wide roof and fc passes FcE. Past that point the stud does not have a utilisation; it buckles. The engine returns Infinity on purpose, and the row was marked as failing, which is right.
The bug was in the words. The Structure tab, the utilisation list and the printed heat-map plan sheets all printed Math.round(ratio × 100) + '%', so a person saw "Wall studs Infinity%".
Two lanes, one hour
I wrote a fix: a formatter that names the case instead of printing a number, used by every view. When I merged main into it, the roof-structures lane had already landed a fix of its own, as part of its truss work (PR #11589). That fix caps the stud's ratio at 9.99, so it shows as 999%, and adds "the load reaches the stud's buckling load" to the row's detail. Either fix works. I dropped mine rather than spread a second formatter over eight files other lanes were editing.
What I kept closes the gap that let the bug through. The fuzz invariant used to check only that every ratio is a finite number. The ratio was the one thing a reader never sees: they see the text. The invariant now also renders the utilisation report and every heat-map plan sheet and fails on Infinity, NaN or undefined anywhere in them. On the code from before tonight's fix, that check catches the three designs. A pinned test rebuilds the smallest case I found: the 8x12 classic widened to 15.5 ft, with 12 ft walls, 2x4 studs at 24", 80 psf snow and 30 psf wind.
What fixes that shed, by the tool's own check: 2x6 studs at 24" (53%), or 2x4 studs at 12" (97%, just under the limit). 2x4 studs at 16" are not enough (184%).