It would have been interesting to add in a validation layer, letting the crowd evaluate each others moves, creating a virtual trust score for each player. Those with high trust scores get to meta-moderate others moves before those moves can be applied to the board. You can also try to catch trolls here by presenting them with a known good matching and the system blackholes those who reject it. Similarly, if their overall trust score dips below a certain threshold, whether due to poor skill or malicious activity, they get blackholed then too and any subsequent change they make to the board is only visible to them privately though they think they're causing havoc.

Combine a system like that with a requirement that N (some reasonable number of randomly chosen people) must agree when combining pieces or splitting pieces before the change is applied to the master board. And add the ability of the project admins to scan the board and lock in place obvious good solutions. That might work.