Stochastic results, percentiles and validity

Methods and settings

Base is not the expected result

Base is the deterministic output using current assumptions. It can differ from the simulated mean and median, particularly with skewed distributions or nonlinear relationships. The displayed percentiles are P5, P10, P25, P50, P75, P90 and P95 using interpolated sample quantiles. Standard deviation uses the sample denominator n − 1 when defined. Neither a percentile nor a sampled minimum is a guaranteed future limit.

Understand the denominator

Target statistics use valid completed target values only. Invalid runs are counted and their available issue information can be inspected. If 90 of 100 completed targets are valid and 18 valid targets fall below a threshold, the displayed probability is 18 / 90 = 20%, not 18 / 100. This describes the valid-result population; do not ignore why ten cases failed.

Change a question without resampling

Changing the target threshold can recompute its probability from the existing results without a new simulation. This does not update stale economic assumptions. A CDF reads the share at or below a value; use the explicit comparison control when equality matters. Additional constraints have their own valid denominators, described in probability constraints.

Ranking and result lifetime

Input ranking measures Spearman association, not causal contribution or statistical significance. Ranking uses absolute correlation strength; a strong negative association can rank above a weak positive one. A constant input or constant valid target has no defined rank correlation and is shown as unavailable, not as proof of no effect. Expanded results provide more room for charts and tables. Changes to relevant model inputs mark the run stale; re-run before relying on it. Completed results are transient rather than a persistent simulation archive, so retain assumptions and needed output evidence through the supported workflows.