Run reproducible simulation and inspect stability

Choose an inherent or residual scenario and explicitly include or exclude base uncertainty, threats, opportunities, and correlations. Record event exclusions and bounded overrides with their own basis. Set the requested percentiles, histogram bins, iteration count, and seed. The same normalized mathematical inputs, engine version, scenario, stochastic-variable order, seed, and iteration count must reproduce the same result fingerprint.

Before running, resolve blocking findings: missing baseline, invalid distribution, cyclic schedule, broken references, invalid scenario, or a non-positive-semidefinite enabled correlation matrix. An all-fixed model may run as a disclosed degenerate diagnostic, but it is not ready to support a contingency decision. If the engine is unavailable, do not substitute an unversioned browser calculation.

Stability and comparison

More iterations reduce Monte Carlo sampling noise but do not improve weak inputs. Examine convergence checkpoints for the policy percentiles rather than assuming the run is stable because it is large. A tail percentile can require more iterations than the mean. Record completed as well as requested iterations and retain warnings if the stability diagnostic is weak.

Common random numbers make like-for-like scenario and inherent/residual comparisons less noisy when the variable set and seed align. They do not make changed inputs equivalent. Raw samples are not authoritative persisted facts; summaries carry the exact input fingerprint and become stale after a material edit. Do not claim a run occurred unless the server-derived result is attached. Next learn contingency policy and percentiles. ## Interpretation discipline

Keep inputs, derived results, and human decisions visibly separate. Every material number needs linked evidence or an explicit assumption flag and basis. Unknown is not zero. A percentile is conditional on this model, scenario, seed, engine version, and source version; it is neither a guarantee nor an approval. Any material model edit makes an earlier result stale.