Linear rewards leave selfish upvoters at an advantage. Overall they can engage in the bare minimum amount of work (exclusive self-upvote) maximizing their ROI/influence/wealth, while the rest will inevitably see them diminish. (This "work" can easily be fully automated.)

This statement doesn't seem to be true at all. It's quite evident that giving large stakeholders a larger say under superlinear makes it even more profitable to self upvote. That's primarily why I believe any superlinear proposal needs to not be pure n^x but with a return to linear as the stake increases.

So no, #1 I do not entirely agree with, and funnily enough I haven't really seen a convincing argument for it either. Hell, the curve doesn't appear to matter for maximizers, no matter how you roll. And you need to distinguish between "passive investors" (if you can even call it that), who will choose the laziest way of accumulating tokens regardless of what system you throw at them.

Possibly the only thing I've seen in here that might matter is the statement "less large stakeholders are easy to police" but um.... nope, I don't think this is true. It's not true now, it wasn't true during n^2 (correct if wrong), and it won't be true if we go back.