QL has been very close with most (fairly normal) bottleneck cartridges that I've tested.
With .30-30 and 6x45mm, in particular, QL's predicted velocities have been dead-on. The statement applies across the board, but I've had one specific load for each of those where QL's predictions for each powder charge increment were within 5 fps of each 5-shot string's average.
It has also been fairly close with popular handgun cartridges. Often within 20 fps for popular powders.
But, it doesn't really know what to do when a person starts drawing too far outside the lines.
.458 SOCOM, for example, is pretty much always off by at least 20%. QL will greatly overestimate the powder charge necessary to reach max pressure, while simultaneously underestimating the achievable velocity. (Meaning it's spitting out a max load that's actually 20% or more over the real-world max charge, while suggesting a velocity that's actually slower than the real-world max velocity.)
Some people have been able to fudge case capacity (faking it as lower capacity) and tweak the variables enough to get powder charges to be within 5% or so for a single powder, within a narrow weight range of similar bullets. But, change powders or go from a 300 gr HP to a 405 gr RN, and one must start the process over. ...And velocity predictions are always wrong. I don't know anyone that has gotten QL to correctly predict velocities for .458 SOCOM.
.444 Marlin isn't much different. QL is much closer than with .458 SOCOM, but the predictions are usually still off the mark enough that an experienced reloader can see the prediction as bad, at first glance.
But, as others have said: When used within its capabilities ('normal' bottleneck cartridges), the only real fault is usually bad user input or a bad parameter in the database that the user didn't notice.
Garbage in. Garbage out.
When you select a bullet, check the dimensions, seating depth, etc. Some of the entries in the QL libraries are wrong.