||Regarding: "we think that getting 80% accuracy is going to be hard, let's put the threshold at that level"
Sometimes the contest organizers already have a working solution for the problem and are only interested in solutions obtaining better results than it (I think I've seen a few such MMs in the past). So the threshold wouldn't be arbitrary in these cases.
But, in general, if the problem is really new and the organizers don't have a good idea about the quality of the results that they want (or that can even be obtained), then I don't know how they can choose a good score threshold. Moreover, I don't know if the organizers/sponsors of the match would be interested in spending time to come up with such a threshold, given that it's only useful for TCO'17 points and not necessarily directly useful for the results they would get out of the match.
Nevertheless, if a predefined threshold exists (no matter how it was chosen), I can invest some time to estimate how easy/hard would be to achieve it. And, based on this initial time/work investment, I can decide if I want to invest more time (in order to obtain a score better than the threshold) or not.