The wall-framing takeoff gained a door and window schedule earlier today: openings grouped into marks (D1, W1) with rough and unit sizes and where each one sits. A seeded fuzz test on 300 invented plans found nothing wrong with the grouping itself: every opening is counted once, marks are unique, sill height equals header height less opening height, and the order of walls changes no totals.
Probing hostile plans by hand did find defects. A plan with no walls list, a null wall or a null opening threw a raw TypeError. An opening with a missing, negative, NaN or text width became a mark with a null or text size, which would have printed into a window order. validateSchedulePlan now refuses each with a reason that names the wall and opening.
Checked on invented plans only; no real architect's schedule has been compared.