1. All members generate a random value (high entropy).
2. All members submit a hashed value of that random value (the commitment).
3. All commitments are distributed to all members. At this point, no more commitments may be submitted.
4. All members reveal their random values to each other and validate each value against its hash.
5. All values are then used to deterministically calculate the coin flip.
A dishonest member will not receive any values used in the coin flip calculation before step 4. If a member decides to change their random value, the hashed value / commitment will be invalid.
Sorry, I think I may have used the wrong terminology in my post above. I was referring to the random value as the 'commitment' and the hash of that as the 'commitment hash'.
But what I proposed doesn't require anyone to change their random value. It relies on the fact that step #4 does not happen instantaneously, and honest clients will send their random values as soon as they transition into step #4.
So from my above example, the process reaches step #4 as normal - all honest clients send out their random values, and all dishonest clients wait. Once the dishonest clients have all of the honest random values, they can now determine what the result for #5 will be(because they know all other dishonest client random values because they are colluding, along with all honest client random values), whereas the honest clients cannot do so, because they don't know the random values from the dishonest clients.
So at that point(half way through step #4), dishonest clients can then determine if they want that value to be the end result, or if some of them should drop out(never broadcast their random value), in order to modify what everyone determines to be the final value from step #5.
I have implemented an extremely similar protocol for a side project game.
In my implementation, step 4 happens in two phases:
4a. All members submit their secret values to the server.
4b. Once all secret values are submitted, they are all sent to each member at once.
This ensures that members cannot determine the outcome before anyone else.
What you are describing could actually be an issue if there is no server involved, however something like commitment encryption (rather than hashing) utilizing secret sharing could help resolve this issue in a p2p environment.
Another option would be that all possible outcomes are randomly shuffled with their positions encrypted and distributed to all members prior to the game beginning.
My protocol implements the latter... all possible outcomes are encrypted and publicly available on IPFS to members before and after the game.
Only after step #5 is the value valid, i.e., after everybody has broadcasted their value. This includes everybody who broadcasted anything in previous stages. As long as the dishonest clients have not all broadcasted theirs, the procedure is not done and there is simply no consensus yet.
One could argue that the dishonest clients could use this to basically abort the procedure before the value is determined (by not broadcasting their values). But once the procedure is complete, being dishonest does not mean anything.
2. All members submit a hashed value of that random value (the commitment).
3. All commitments are distributed to all members. At this point, no more commitments may be submitted.
4. All members reveal their random values to each other and validate each value against its hash.
5. All values are then used to deterministically calculate the coin flip.
A dishonest member will not receive any values used in the coin flip calculation before step 4. If a member decides to change their random value, the hashed value / commitment will be invalid.