Leaky Bottles

Someday I hope to tell a skeptical child that cold bottles sweat because the bottles leak.  The fun part would be explaining away objections and attempts to prove this assertion false.

Leaky bottle of delicious ginger beer

If you lick the bottle, the stuff on the outside doesn't taste like the stuff on the inside.  Of course not, the bottle acts like a filter, letting the little water molecules through, while filtering out the big sugar, alcohol, and flavor molecules.  Sugars are long chains of hydrocarbons, while water is much smaller.

Bottles only sweat once you take them out of the refrigerator.  This is expected.  Most things, like glass, shrink when they get cold, so the holes in the glass get smaller too.  Water is weird though, it gets bigger when it gets cold, like you can see ice does, so the water molecules are getting bigger at the same time the holes in the glass get smaller.  That's why cold bottles don't leak until they start to warm up.

Cold empty bottles sweat just like bottles full of liquid.  That's because when the bottle got cold, and all the little holes closed up, trapping the water that was in there before inside.  When it warms up that trapped water is released.

If the humidity is very low, the bottle doesn't sweat.  Water evaporates very quickly in low humidity environments.  The water is still leaking out of the bottle, but it's evaporating before it can collect on the surface.

And so on.

This leads to some interesting questions.  What makes this assertion hard to disprove?  I think there's something about the close relationship between the correct explanation and the incorrect one.  Because the false mechanism is affected in similar ways to the true mechanism by the temperature of the glass, and the mechanism itself is difficult to observe, it's difficult to distinguish between them.

I wonder, for a given system and hypothesis space, what's the most difficult to disprove false hypothesis.  This seems tightly related to controllability and observability.  Are there systems where the relative likelihood of a true hypothesis can never exceed the most likely false hypothesis by more than some margin?  That sounds like an anonymity framework.