If you were to run TCEC how would you define/make the opening rules?
My opinion is that random openings of length X are useful for testing, but trying to splice this into a "league" concept is not so palatable. Choosing 12 moves from a large PGN file can lead to funky play, especially when strange moves are made at the opening part in the game. Maybe if you stuck to "top GM" games it would work better, but from what I've seen, a non-negligible percentage of the games have "quirks" in the first few moves. [Computers shouldn't have to put up with openings from dodgy humans
]. Also, especially with 3:1:0 scoring, if engine X happens to get 4-5 drawish opening from the 7 choices, and engine Y receives two less, it could be a burden.
Ideally, the makers of the engines would have some input, perhaps via a submitted opening book (similar to SSDF). I'm personally somewhat against the separation of book/engine, though I know many developers prefer it (and UCI almost tends to enforce it). However, given that some of the developers are probably now "boycotting" TCEC due to the inclusion of IvanHoe/Houdini, this is unlikely to get anywhere in practise.
Another option would be to query outside expertise. There's got to be some quality book-makers around who would be willing to spend the time (3-5 hours?) to collate 56 (times 3, plus a few more for the Elite match) playable opening positions for a season's worth of games, especially now that TCEC has gone big-time. I don't see how this could fail to improve on random PGN selections. You could also open it up to the public via a poll somehow.