A cartel / collusion, I think is if some of the 12 witnesses would cooperate behind the scenes. That is ground for schism. Because then it would take too long time to change them. If 6 witnesses are in collusion and are actually acting as 1.
All 12 can be replaced, it just takes longer time, exactly to prevent changing all 12 at once to "Great Enterprise Censored Network". This is the point schisms addresses in the whitepaper.
Just a moment Ill find whitepaper reference to how more than 1 witness can be changed - there is a rule, when 1 witness changed becomes stable enough for the others to be allowed to change.
This is indeed an interesting aspect of the system. How a new witness can be introduced - does the whole network (all full nodes) need to accept it (include in its witness list) or 51% (?) and whether remaining 11 witnesses have the veto power to prevent it from happening? This is not clearly explained in the whitepaper. I think tonych would be helpful in clarifying this issue as it is of paramount importance to the security of the system.
Let's imagine that 12 witnesses formed a cartel. They will not change their own witness list. The rest of the network decided to change 1 witness (they are only allowed to change just 1). Now how there can be a second witness changed? If any user tries to change the second witness in its list it will cause its list to differ by 2 with regard to the remaining 11 witness own witness list. Remember that there's a cartel - remaining 11 witnesses won't allow the second witness to be changed (they will not allow their own witness list to be changed). I think this prevents the second witness from being introduced.
Reading whitepaper again.
That seems to be correct, if 6 of the witnesses collude/cartel they can prevent others from changing their witness list, and they can censor transactions. Then schism.
Page 20-21 explains Choosing witnesses.
I expect this kind of schism happening pretty frequently. Even simply as a way to disrupt the system by whoever dislikes byteball. Whitepaper says its possible:
If someone just wants to start a new coin to experiment with another set of protocol rules, he can also use the alt field to inherit everything from the old coin, make the switch comfortable for users, and have a large set of users with balances from day one.
Choosing witnesses should be done carefully.
I expect schisms as frequently as collusion happens between 6 of the witnesses. Not often, but every 10-20 years, depending on how Byteball is used, and maybe it would survive even that if its used very widely. It will be hard to pick the good initial ones, but seeing as people here are crying over their exchanges not having Byteball - trusting a third party with their bitcoins a coin whose whole purpose was that no third party should hold your coins - I am convinced we will find the trustful witnesses, because people are willing to trust. Its the slow collusion thats bad, the "power corrupts" effect.
New altballs as often as someone is willing to experiment.