Precision and accuracy.
Models are quite accurate, but they are rarely precise anymore. You aren't going to see run after run of the same solution given both the frequency of runs and model resolution.
When your grid scale was 80 km, subtle shifts made little difference to the outcome. When you drill that down to 13 km and start focusing on QPF/snow maps you are going to see some big moves from haves to have nots. There is a reason why the best mets are using the forecast funnel approach (large scale to small, top down).
Like mid level magic. It isn't just a term we made up for the picnic tables, it's that modeling frequently overdoes QPF near the low center (with latent heat release) and underestimates forcing via deformation processes (in Canada that reads "DEE-formation PRO-sess-ees").