This is very interesting.

If I'm reading it right, people vote to settle outcomes, under an incentive system where you want to:
1) participate in a lot of votes
2) avoid falling out of line with the consensus
? 3) avoid damaging your reputation ?

It sounds like the incentive will be for everybody to pull data from the same feeds, because that will be the easiest way to stay in line with the consensus. So you end up with whatever the feeds are that everybody coalesces around becoming the de-facto trusted authority that your network relies on.

This may not be a bad thing - presumably that authority will be motivated to stay reliable, and if they suddenly stop providing feeds, the network will be able to switch. But if the feed just provides sub-optimal data, or even starts providing outright lies, how does the scheme incentivize people to get back on the right track? It feels like once we're all pulling from Feed X, anybody who tries to deviate from Feed X is going to get spanked, even if Feed X is now full of shit.

Doing this with consensus witnesses is a hard problem compared to the problem Bitcoin solves because in Bitcoin all the nodes have to do is execute some extremely cheap scripts and timestamp the results, which is so simple and hard to screw up that there's no point in trying to pull that information from a central authority instead. Just run the software as designed and you'll stay in line with the consensus. (Although even there, miners tend to organize into pools and outsource their "voting" to a small number of pool operators...)

Disclosure: I run Reality Keys, which is a centralized authority for certifying facts that among other things can be used in an (otherwise decentralized) prediction market, albeit one that can be combined with other authorities and settlement mechanisms.