#3 makes it look like things predicted as "very likely not" (10-20%) and "very likely so" (80-90%) tend to be biased towards the contrarian view. Does Metaculus have limits that make it unprofitable to make bets things that would push them outside the window? Anecdotally I've seen quotes from prediction marktets that show a surprisingly high chance for things that are incredibly improbable (to the point that I would call them "impossible" in casual conversation).