Then the most important question for any model would be what domains it's good at.

For example: one model approximates the population as infinite, so it gets decent predictions when the number of agents in each category exceeds five (this is rare).

These requirements to apply the model should be the first thing taught about the model.