Now, it's true that the lucky monopolist who ended up as the Bill Gates of data feeds probably doesn't have an incentive to screw it up on purpose, and it probably won't be too bad in practice. Often the winners of natural monopolies are pretty good at their jobs - Windows isn't all that bad... But I doubt this is what the designers intend. You have a lot of cost and complexity with all the voting stuff to make the decisions about outcomes decentralized, but then you end up with a de-facto centralized structure anyway, and worse a centralized structure that's hard to fix if it goes bad. If you knew you were going to end up with authorities anyhow I think there would be simpler, more robust ways to do settlement. (For example, skip the voting and let people specify a list of settlement authorities in the contract and a protocol to choose between them if they diverge. That puts your settlement authorities in self-correcting free-market competition as you describe for contracts, rather than lumbering you with the one who won the battle to be the de-facto go-to guy.)
That sounds like my scenario. The voters either notice the bad data coming in and ignore it when voting (similar to how people stopped paying attention to MtGox as it kept fucking up, and a scenario which doesn't require such pervasive incompetence on the part of voters as to enable the data source to do a 51% attack by proxy) or they do a bad vote and then people subsequently issue better markets worded to forbid use of the untrustworthy data source.
Right, as long as the parties are able to make _individual_ judgements on which authority (or better, combination of authorities) to trust (eg they can choose when they make their contracts), it doesn't matter too much that there are authorities involved. You have a nice competitive mechanism that should tend towards good data, at a reasonable price. Handle the contracts on a p2p network and I think you have a very nice simple, robust solution.
The problem is when as well as handling the contracts on the p2p network, you try to take the decision about what authority to trust away from the individual parties and give that to the p2p network as well. (The proposal doesn't think it does that because it doesn't think it has trusted authorities, but I reckon in the real world it may get them whether it wants them or not...)
Trying to make the network do this via voting is a very interesting experiment and I'd love to see it tried, but I think it has some interesting potential pathologies...