I wouldn't call myself an expert on simulators, however I was in the team who first introduced helicopter sim training to the RAF, using an existing commercially owned sim. I later ran the project for the RAF.
We were duty bound to find out how accurate the response of the simulator was. At first, it wasn't at all realistic, not surprising because it was designed around a later mark of aircraft, with a different cockpit layout, different engines, different aerodynamics.
The simulator programmer guru simply plugged in a keyboard, got out his computer language manuals (which he had written from scratch) and rewrote the programme it until it flew like the RAF version. We then had to select our version of aircraft when booting up the sim.
We later found that the sim would go into "crash mode" if certain parameters were exceeded, such as slight yaw or roll rate at touchdown. We were doing double engine failures to EOLs (in the dark, btw) and this was inconvenient because the sim needed to be reset each time, rather than just repositioned. This wasted valuable training time, let alone put the $hits up the average student because it could be quite violent, being a full motion sim. Again, the programmer simply rewrote the software in a couple of minutes, so it wasn't so sensitive. I spent a lot of time talking to this guy, over the years I was in that job. He told me that certain responses could never be assumed to have been programmed correctly. For example, tail rotor drive failures. No-one can be persuaded to deliberately fly an aircraft in this condition to obtain data, so an estimated response of the aircraft is all you can hope for.
As I understand it, these "best guesses" are known as "off model responses".
So, take simulators to be the "complete authority" of actual aircraft response at your own risk!