Analyze Monte Carlo

Same feed and steps the simulation posted, replayed through a Monte Carlo uncertainty analysis: every model calibration, economic assumption, feed or fermenter titre, the raw-material price level and the capital estimate are sampled JOINTLY (Latin hypercube, rank-correlated) and the distributions of the objective and the headline metrics are reported.

n_samples draws (100-5000; about 2-20 ms each depending on the route) seed makes the run reproducible variation relative P10/P90 half-width on every sampled input (0.3 = a lognormal whose P90 is 1.3x and P10 1/1.3x the base) capex_p10 / capex_p90 the AACE class band on the capital estimate; the default 0.70 / 1.50 is the class 4/5 band the estimator prints include_capex, correlate switch the capital factor and the default rank correlations off for a pure-marginals run