I didn’t simulate until the end of the game.
During simulation phase (rollout) I just play one more move or not at all:
<<According to the last action played during the selection-expansion:
-for a PUSH, evaluate the current position (no additional action)
-for a MOVE, play one additional push for each player according to the following heuristics: try 5 pushes at random, keep the most promising >>
(promising = a good score with the Evaluation function for a player)
For many people, what I did was not an MCTS.
It is an UCT (Upper Confidence applies to Tree).
For me, there is enough randomness (during expansion, for additional action after a move, for new quests) to talk about MCTS (Monte Carlo TS)