As a broad heuristic I think this fails because posts get downvoted or removed for a variety of reasons. One common reason is that it’s a cryptocurrency scam. Are those “more likely to be true?”
Cryptocurrency scams aren’t ideas in the sense of the marketplace of ideas. Vaccine efficacy, election integrity, causes of unrest in France, traditional analysis of gender - those are examples of topics around which we discuss ideas. Try to discuss them in a simulated user forum. As an experiment, vary the viewpoint of your remarks, pro-X vs. anti-X. Observe what happens. You can learn from the censorship profile what the censors are worried about.
I suppose that’s a way of learning about what upsets people. But why they’re upset is another question. Maybe it’s not the idea, but its presentation?
Also, what set of ideas are you drawing from? There are a lot of uncontroversial and true ideas. (Or if you negate them, obviously wrong.) Why aren’t they “more likely?” I don’t know how you decide on the set of ideas to test without bias.
And how do you evaluate the results? To tell whether ideas that are “likely to be true” are more likely to make people upset, you need to already know whether they are true. Otherwise you’re just sorting by controversial.