I trust the machines.
Seriously though, these have been pretty helpful at IDing time periods to watch for. Because of the random forecast creation of these probabilities, it's not like the SREF where large probabilities may be dominated by one variable (like CAPE). This is what it is using for inputs.
"These include surface-based CAPE and CIN, 10-m winds (U10, V10, UV10); surface temperature and specific humidity (T2M, Q2M), precipitable water (PWAT), accumulated precipitation (APCP), wind shear from the surface to 850 and to 500 hPa (SHR850, SHR500), and mean sea level pressure (MSLP). For Day 1, three additional predictors are supplied: surface relative humidity (RH2M), lifting condensation level height above ground (ZLCL), and surface to 850 hPa storm relative helicity (SRH)"