GRT is after all a simulation. Expecting a simulation to be spot on, or at least 99.5% accurate without calibration1), in every instance, is to be discouraged, and not realistic. If the user fails to verify all of the parameters for accuracy, and just relies on accepting the default values of the provided databases2), a poor result is likely.
For rifle cartridges, in general, a deviation prior to calibration3) of e.g. 3% or less, such a simulation would be rated as “excellent”. For up to 5% would be “very good”, and below 10% as “good/satisfactory”. Even though the variation is often less than 3% with a highly rated powder model, quality components and good loading practices, GRT is not a Magic Ball!
For pistol cartridges the story is somewhat different. Since pistol cartridges have much less powder charge in comparison to rifle cartridges, the available energy is much smaller. Therefore, all of the component variations/influences have a much higher impact on the deviation/error rate. With pistol cartridges the ratings are 6% or lower as “excellent”, up to 10% would be “very good” and below 20% as “good/satisfactory”.
GRT is able to predict rifle cartridges often below 3%. That is amazing enough when you look at the possible influencing factors and their variations (given reasonably good components):
If you now add up several of these variations in the worst case scenario, you know why it is recommended to start charge development at -25% Pmax or below.
Hopefully this explanation gives you a perspective on the simulation possibilities and how to rate the limits and capabilities of GRT's simulation for yourself. That a simulation can be predominantly below 5% remains a mystery that amazes us as well. It is important to recognize that these variations do exist in reloading, the components/tools used, how to minimize them and how to best use the (additional) tool “Simulation”.