H3 — Scalability report

Does per-step runtime grow faster than linear with the number of SKUs? We fit a power law t = a · nb to per-step runtime measurements across cohort sizes (50, 100, 200, 500, 1000 SKUs, 10 seeds each) for each policy, then classify the exponent b: < 1 sub-linear, ≈ 1 linear, > 1 super-linear. The per-policy table reports b, the coefficient a, the fit's R², and the resulting classification; the scatter below overlays raw points with each fit line.