Two-way door decision: reversible. It's ok to just do it, and learn whether you were right after the fact.
But there is a third kind of decision this framework misses.
You can make the decision, but results are uncertain. You can undo it/redo it, but the outcome will be different.
In this way, decisions become about bandwidth.
So experiments become more about bandwidth: how much can we try before we change the initial conditions too much, and pollute the space?
Parallel experiments permit revolving-door decisions.