C3.4 Simulation that proves itself

Open engineering course [26]

What it requires. Every simulated result reports a convergence test with a standard error, showing that the figure stopped moving as samples were added; the sampling scheme, including whether a Latin hypercube or a low-discrepancy sequence was used; a validation of the engine against a case with a closed-form answer, with the discrepancy stated; a heavy-tail check confirming that the quantity of interest has a finite mean before its percentiles are quoted; a global sensitivity ranking of the inputs by their share of the output's variance; and the seed and model version. A result missing any of these is a draft, not a finding.

Where it comes from. The open engineering course [26], the only teaching source in the set, distills these into habits for students who will have to defend a number before a regulator. The error of a Monte Carlo estimate falls with the square root of the number of samples; variance-reduction schemes converge faster than plain draws; the engine is run on a known problem first; a notebook on a classical paradox shows a sample mean that never settles; variance-based indices answer the question a committee actually asks, which input drives the result.

What it reconciles. Finance is full of simulations and never adopted these habits, because its model governance grew up around documentation and independent review rather than around the arithmetic of the run; chapter 12 traced the transfer. The component joins the engineering standard of evidence to the supervisory one in C3.6: the engineer asks whether the run can be believed, the supervisor whether the model is still right, and a figure needs both. The seed and the version feed C6.3, where a finding becomes reproducible rather than merely documented.