Run Configuration and Determinism
A valuation run is defined by its configuration; determinism follows from controlling every input.
Run configuration fixes the set of trades, market snapshot, model IDs, and pricer settings (tolerances, seeds, paths).
A unique run ID ties configuration, code version, and outputs into a single reproducible artefact.
Deterministic Monte Carlo requires explicit seeding, not hidden random number generators.
Re-running with the same configuration and artefacts should produce identical results up to numerical tolerances.
This panel compares two valuation runs, A and B, as configurations: a stack of inputs feeding a Monte Carlo estimate. Change the scenario to see when runs agree bit-for-bit, when they differ only by sampling noise, and when they differ because the configuration has changed.
When all configuration fields match (including the seed), Monte Carlo is bitwise deterministic: reruns are identical up to floating-point quirks. If only the seed changes, price differences stay on the scale of a few standard errors and shrink as you increase the number of paths.
Changing snapshot/model or tolerances creates structuraldifferences in the configuration. Those are the differences that should survive into a run ID and into audit trails. The practitioner’s rule: do not blame “Monte Carlo noise” for differences that actually come from changed configuration; control seeds and log everything you vary.