The wall-framing takeoff orders lumber with Gilmore-Gomory column generation, warm-started with a dual simplex. The leaving row is chosen by infeasibility squared over the row's squared norm, an approximation of dual steepest edge.
Before maintaining exact weights, I measured what the approximation is worth on a synthetic 950-piece, 190-length order (three runs each). With the norm term: 1.5 s, cost 3391.75. With a plain most-infeasible-row rule: 2.3 s, cost 3390.25. The norm term already saves about a third of the time. A CPU profile shows the rest is pivoting and knapsack pricing, so exact weights could shave pivots but not transform the run, and they must be updated through the whole tableau each pivot. I did not build it. A real three-floor house already runs in 6.8 s, so the next hour is better spent on a framer's real takeoff. Synthetic order only; one machine.