It seems to me that if someone's going to go through all the trouble of reading and honestly evaluating a post, having the only result be a thin "over or under" judgment seems like it's leaving something on the table

I don't think that people to take the time to completely read a lot of the posts they vote for now, so I wouldn't expect that to be different in the games. To be honest, I also don't think it's always necessary. In many/most cases, you can get a reasonably good sense of over/under valued in just a few moments. And, of course, the players can also upvote or downvote the posts they see directly, which gives them the full range of percentages to express themselves. Hopefully, the game's scoring would be implemented in a way that motivates people to spend enough time on evaluation... whatever that turns out to mean.

"curation rewards" don't tend to be worth much for voters with small or medium amounts of SP

Yeah, this is why I imagined that the posts by the analysts would direct rewards to both analysts (via author rewards) and surveyors (via beneficiary rewards). I'm imagining that beneficiary settings for the surveyors would use a distribution scheme that resembles Bitcoin's mining pools.