There is a very big difference between precision (statistical error or uncertainty) and accuracy (systematic error). Accuracy will depend on the code (bugs, nuclear data, specific treatment, etc.) and on the user (wrong hypothesis, bad use of particular method, etc.). The precision will depend on the number of events in the region concerned and on the statistical method used.

What is a good uncertainty?

If the result of a run is ф with N histories contributing to this quantity, the uncertainty is Аф / 1/vN; thus, to divide this error by 2, N has to be multiplied by 4. But the computer time is proportional to N, so the duration of the run is also multiplied by 4. In general, an MCNP result with an uncertainty less than 10% is sufficient, but only if the entire volume has been visited by particles. In order to check the confidence in a result, MCNP gives the evolution of a number, the figure of merit (FOM) which is defined by FOM = 1/(Аф2 x T) where T is the computer time. This number must be approximately constant (T / N and Аф / 1/VN) for the result to be credible.