At the moment - tasks are manually approved.
This is bound to change as escrow system gets developed.
You can find development roadmap found about.cryptotask.org
Feedback is added with after a task is successfully completed, which over time would create high quality freelancers from ones who don't perform so well.
As for the dispute system - I'll copy previously provided answer taken from Forbes article:
"With a review system like that of CryptoTask, reviewers are token holders that stake an amount equal to the task value that they are putting themselves forward for as a potential reviewer. It means that being a stakeholder, you already have a chance of being selected. That chance is directly proportional to that individual's stake, so this prevents sybil attacks.
If they do, however, get randomly selected to be a reviewer and the task goes into dispute, they are required to cast a vote on whether or not they think the task was actually completed. If a reviewer votes against the consensus or does not vote at all, they do not lose any certain percentage of their staked amount.
Some of the benefits of such a system include:
Selection happens in secret to prevent reviewers from influencing the consensus through collusion.
A two-stage voting process (secret commit and then reveal) means reviewers can’t wait to see how other reviewers voted and then just go with the consensus.
Reviewers would generally be professionals with a relatively large token stake and therefore in a good position to escalate and vote on a dispute."
This would sum up dispute system with incentive for reviewers being 10% of the task value split among them.
Very cool. Although there's still challenge on the human side on /seeding/ the right high quality initial network, I can see something like this really being able to scale. The question of course is how well this conceptually scales with task complexity -- obviously, something like an Amazon MTurk task is one thing and probably the appropriate existing system for a system like this to supplant.
With that said, how does this scale to handling more complex tasks that require more back and forth? There's the "fuzzy" area of problem solving as tasks get larger and less clearly defined. Is that out of scope for now to keep focus is on the low hanging fruit?
I would love to see something like this completely deprecate MTurk and low wage Upwork.
As for the dispute system - I'll copy previously provided answer taken from Forbes article: "With a review system like that of CryptoTask, reviewers are token holders that stake an amount equal to the task value that they are putting themselves forward for as a potential reviewer. It means that being a stakeholder, you already have a chance of being selected. That chance is directly proportional to that individual's stake, so this prevents sybil attacks.
If they do, however, get randomly selected to be a reviewer and the task goes into dispute, they are required to cast a vote on whether or not they think the task was actually completed. If a reviewer votes against the consensus or does not vote at all, they do not lose any certain percentage of their staked amount.
Some of the benefits of such a system include:
This would sum up dispute system with incentive for reviewers being 10% of the task value split among them.