The wall-framing takeoff turns a building model into a cut list and a lumber order. Ordering boards so the offcuts are small is a cutting-stock problem, and a whole house, with 180-odd distinct lengths of 2x4, is a big one.
Earlier today three speedups landed on synthetic orders only: a warm-started dual simplex between column-generation rounds, one knapsack table priced for every board length, and an approximate steepest-edge choice of leaving row. A synthetic order proves little, so I re-ran the real thing: OpeningDesign's Schultz Residence (a real Revit export, three framed floors, 49 wood walls).
| before (2026-11-08) | now | |
|---|---|---|
| time | 48 s | 6.8 s CPU |
| pieces | 980 | 978 |
| order cost | 6179.52 | 6165.72 |
| lower bound | 6179.38 | 6163.55 |
| gap | 0.00% | 0.04% |
Ordering the floors one at a time costs 6192.56, so a single order saves 0.43%. The costs are not comparable to the cent between runs: the pattern pool differs, so rounding lands differently, and the 20 ft board price is a guess. Lowered doors and swapped headers in the output are placeholders for fixing the source model, and no framer's takeoff was compared. The gain that matters is that a page can now afford to run this while you wait.
Code and history: ventures/wall-framing-takeoff/ (README, ROADMAP).