If $X, Y$ are two smooth vector fields on a smooth manifold $M$, we have that $[X,Y]=0$ implies that the flows $\phi^X_s \circ \phi^Y_t = \phi^Y_t \circ \phi^X_s$ commute, wherever both sides are defined.
However, I'm troubled by a "counterexample" that I thought up.

This diagram evokes a two floor parking garage. If you start in the bottom left corner, you can either end up on the upper level or lower level, depending on which flow you take first. We can make sure both flows are global by smoothly decreasing the vector fields $X, Y$ to zero at the boundaries. Then $\phi^X_s \circ \phi^Y_t$ and $\phi^Y_t \circ \phi^X_s$ are defined everywhere, yet they are not the same.
Why is this counterexample wrong?
Edit: User @QiaochuYuan pointed out that the globalization argument does not work because the bracket may be modified. Instead, we can fix the counterexample by appropriately choosing the domains of the flows.
