Really good point about the "free-market response". There's no better way to know how something will work out than to try it! That's also along the lines of what I was thinking about with SMTs. They will allow actually trying different curation strategies in live environments.
In the middle of writing this comment, I had a different, half-baked, idea of how to handle curation. Perhaps instead of paying out a % of the rewards from each post, there could be a separate reward fund specifically for curation. This way it would be possible, in theory, to get a larger curation reward for voting on a post than the author gets from the votes on that post.
The idea is to de-couple curation rewards and author rewards - i.e. not have them fighting for a share of the same pie.
As I said this is a half-baked idea - currently the "quality" of a post is determined by the value of the votes it gets, so this would require some other method of evaluating post quality, which may or may not be possible.
Anyway I think i'm rambling here - very glad that we're starting more discussion about this!
"As I said this is a half-baked idea - currently the "quality" of a post is determined by the value of the votes it gets, so this would require some other method of evaluating post quality, which may or may not be possible."
I love this idea, and I've considered proposing it. Of course, the key problem is subjectively assessing quality with an automated algorithm. I've posted about how to start attacking this problem so it could be possible to suggest this change, but I'm not sure exactly how to do it. Using reputation somehow is an option, and I've also suggested having a second reputation score for curation rep specifically. This might allow the decoupling (experienced curators can earn more via their higher curation rep) of which you speak.
I would like to work with a machine learning expert to train a model on the existing set of Steem posts and see how well it can "learn" to pick out high quality posts.
It would have to have no knowledge of the author of the post (so it doesn't learn just to pick popular authors) but just the content of the post and the final payout.
Then if that is used to evaluate new posts it would be interesting to see how different its evaluation of what the payout "should" be vs what it actually is. This would tell us to some degree how much who the author of a post is affects the payouts, and might allow us to find unknown authors that make high quality posts more easily.
I would bet that machine learning could do a lot with this topic.
An algorithm that actually works to make higher quality appear in trending would be great, but it's quite a task from where I'm sitting.