any strategy I develop for extricating myself from a foul mood ends up only working within the context for which it was designed.

Have you tried designing strategies specifically so that they wouldn't work in the context where you're designing them, and then running tests on those? Say, leave a post-it note somewhere visible saying "you are in a bad mood, and will respond to this observation with irrational anger," then updating the last bit recursively until it's accurate enough that the tired, stupid version of you is forced to agree, or is at least thrown off-balance enough to break the behavioral pattern.