Simulation verification and validation

Verification asks whether the model was built as intended. Validation asks whether that intended model is good enough for the decision you want to make.

A model can pass verification and still be a poor representation of the real system. It can also match one output by accident while containing implementation errors elsewhere.

• Verification: check logic, routes, units, event order, counters, and calculations.
• Validation: compare model behavior with real data, known limits, and expert knowledge.

Start with cases whose answer is easy to predict. Remove randomness where possible, send one entity through one path, force a queue to form, and check totals at each boundary.

Trace surprising results back to the event or state change that caused them. Do not start by tuning parameters until the result looks plausible.

• Check that every entity is created and disposed exactly once unless duplication is intentional.
• Check units at every conversion: seconds, minutes, hours, rates, and percentages.
• Check extreme cases such as zero arrivals, unlimited capacity, or one service slot.
• Check conservation rules where they apply, such as inventory in = inventory out + inventory remaining.

Choose observations that matter to the intended use of the model. For a patient-flow model, that may include arrivals by hour, queue lengths, waiting times, length of stay, and resource use.

Compare more than one operating condition. A model that matches an ordinary Tuesday but fails during a demand surge may not be valid for surge-planning decisions.

The practical question is whether there is enough evidence to use the model for a stated purpose. Write down that purpose, the tests you ran, the conditions covered by those tests, and the known gaps.