The engine's loss model draws claim frequency and severity from normal distributions and multiplies them per trial. The shipped workbook carries 400 pre-seeded trials it cannot regenerate. Here you can replay those exact draws — or change the parameters and re-simulate as many times as you like.
Loss per trial = max(N(μf, σf), 0) × max(N(μs, σs), 0). Seeded draws make every run reproducible — same seed, same answer, audit-grade.
Decision weights against sleeve return coefficients · objective = Σ weight × coefficient
—