RE: Wholesale witness voting changes.
This "economic" fix seems to be completely ignoring any thought about what the optimum economic behavior actually is under the proposed changes, as opposed to what they wish it is.
Superlinear payouts don't reward quality. It rewards correctly predicting what other people will vote for, which is a coalitional game that can be played by spammers and scammers alike.
Increased curation rewards don't make it more rewarding to find good content. They make it cheaper to run a bidbot service. Today if it costs $X to buy a vote worth $0.75Y and the service gets $0.25Y, the bidbot profit is the difference between $X+$0.25Y and the opportunity cost of self-voting on a spam vote.
Before: X + 0.25Y > Y
After: X + 0.50Y > Y
Result: price ($X) goes down.
The hope that combining the two will magically result in a game that rewards quality instead is foolish.
In fact, I suspect some simple spreadsheet math would show the thresholds at which two spammers can vote for each other in order to increase rewards, which gets easier with 50% rewards.
It certainly discourages spreading votes around. "Curators" are always incentivized to generate a 100% upvote.
A metric that rewards quality would measure whether the result of an upvote (or resteem) was to bring the article to the attention of others who also found it valuable. (I've been thinking on writing an article about Bayesian surprise, but I haven't been able to measure it well on Steem data yet.)
Somewhat valid point. I think there is some effort to think about that, but it is far from clear especially once one goes beyond simple (e.g. hyperrational, risk-neutral) models and considers actual behaviors.
Let's be clear on one thing. Superlinear is discussed in the post but very few major stakeholders nor witnesses support it (as somewhat indicated by the chart). So in that sense criticizing it here is somewhat straw manning.
Support for 50% curation rewards is higher.
I don't think the curation game is anywhere near as simple as you suggest here. Too much vote power applied to a piece of content discourages future votes (and may draw downvotes). It also means (given limited total vote power) that a curator is restricted to putting eggs in fewer baskets, which increases risk.
That is what curation rewards do, assuming the 'others' also vote accordingly. In reality that works a lot better if the 'others' downvote when something of low value is brought up to their attention (which is part of why cheaper downvotes is part of the discussion)
For the categories I am interested in, "Trending" and "Hot" do precisely zero. They're too low-traffic to be worth bothering with. So, voting doesn't bring posts to my attention.
This is in my mind the big failure of understanding in the Steem designers. Upvotes are not curation, unless they affect what people pay attention to. Systems that provide personalized recommendations do this. Steem and the Steemit front end do this... for very popular categories, maybe.
Actual curation on Steemit takes the form of posts with links and resteems. The "proof of brain" treats the former as just another article and the second as nothing. Instead of being rewarded directly, you have to hope that enough people's upvotes on the articles will increase your curation share enough to be worth it--- but if you're being front-run by a voting bot, odds are your increase will be marginal. I bet hardly anybody doing this is in it for the net curation increase over their initial upvote.
My earlier study on curation efficiency suggested that curation went disproportionately to people gaming the system: https://steemit.com/econometrics/@markgritter/steem-econometrics-efficient-upvoting
Steem was designed mostly to find and reward viral content, which means either general interest or at least a large subculture. Some of this is carried over into design of the UI (e.g. lack of much support for categories aside from the lousy tags stuff), some not.
Now it certainly isn't doing much of that and it is debatable whether the user base is actually large enough for viral to have much meaning. It is also debatable whether finding and rewarding viral content is worthwhile or a useful goal. The upcoming SMT upgrade is intended to focus more on smaller subcommunities with their own (interconnected) token economies and individual distribution and voting models. Whether that will actually work out usefully is anyone's guess at this point.
Certainly no one claims that the sort of voting that takes place on Steem is intended as personalization. It isn't that at all. As you say perhaps a personalization engine could be created but there isn't one afaik and it wouldn't be an on chain function anyway. The data is public, you (meant generally to include app developers) are free to do with it as you like, including personalization.
The voting bot which someone developed for that purpose is (though to be clear many of them are quite dumb, but not all), and there are certainly manual curators also trying to do that. The scale of the available rewards limits human time investment though, at least in high cost regions.
Just looked at. Measuring curation as a third party is very difficult. For example (and this is only one of many ways measurements can be misled), I could guarantee a huge return on a small account by routinely voting with that small account and then voting with a big account behind it. From your perspective as an observer the small account would look like excellent curator but from my perspective it would be just sort of value transfer from the big account to the small one.
In short I don't know how to interpret your results although some aspects of it in the aggregate such as limited alpha are clearly correct.