Model companies aren't super-human either. And they don't have huge secret stashes of research information, just the same stuff we have. It doesn't help either that sometimes even reference material is wrong (e.g Sherman IIC Fireflys, symmetrical Tiger turrets and the 'T34/76C'. Or if you model the 'real thing' the preserved tank you look at has been partially mocked up (like Italeri's Crusader) It does take a long time to become an expert in one vehicle type, let alone the range of subjects any company may end up issuing over time. Except they usually don't have the time to gather the material we do (as Jim points out), because they're in a commercial environment, when you're only as good as your next product.
At least Dragon are sending out trial shots of models to recognised experts to check before putting them into full production. Which is a start, but there are only so many experts out there.
A lot of the errors can be chalked up to corner-cutting (re-using old components) and the kind of assumptions that we all make from time to time (like the rear end of the Sherman is always the same).
I only rivet counted once, on the Airfix WW1 tank, and it was wrong, so I gave up ! On the other hand, whoever was responsible for Trumpeter's Hotchkiss Frankenstein's monster would be advised to stay out of my way for a while.
Not sure about the OS and deliberate errors, the accidental ones are quite enough........