Is there a programmatic way to optimize speedruns for this kind of games? The number of steps/the size of the search space seems too large to apply a MILP solver to it