[SomeStupidPoint]

The simplest explanation that fits the facts is that it's a mix of both, as both are solidly established to exist.

Occam's razor doesn't really apply to determining an individual account, as there's no model difference in a particular account being either, given many accounts and each being some.

You should respond based on considering it a superposition of both (and more exotic states, like a non-shill blindly repeating shill points), based on the distribution of likelihood given their speech patterns, background prevalence, etc.

That's really the simplest model which lets you incorporate all the information you know in crafting a response -- not a false dichotomy of making assumptions you know will sometimes be erroneous. (...Which is not what Occam's Razor is about.)