Metaphors -- in this context I'd call them "rules of thumb" -- are computable by humans in real time. Reality generally isn't.
Technically, things don't fall down. When I hold a brick above your head and then let go, you shouldn't just assume that you have to dodge. You should take note of all the massive objects in the vicinity and do a calculation using the law of gravitation. You might also do some energetic calculations to make sure that the brick won't just bounce harmlessly off of your head, or come to rest on it like a falling feather.
Or, you know, you can avoid thinking at all and just allow the circuits in your brain and spinal cord to get you out of the way fast so that you can live to argue another day. Those circuits embody a relatively crude theory of physics, and their design process didn't make use of a theory of physics at all, but they work to keep you alive.
What Krugman is saying is that, while economists would love to develop the economic equivalent of the law of gravitation (and some are even foolish enough to believe that they have) their first priority is to develop some rules of thumb that can tell us when to dodge, when to stand still, and when to wear a hard hat. And that can be used to explain, in simple terms, where the bump on our head came from. ("You stood under a metaphorical brick and failed to dodge in the Keynesian manner.")
Your reflex/calculation analogy makes me wonder if new neural network designs ought to be tested against economic data...
* Joe can make 10 A widgets in an hour, and 5 B widgets in an hour.
* Bob can make 5 A widgets in an hour, and 4 B widgets in an hour.
Joe and Bob need 10 A widgets and 10 B widgets each day to live. They are both lazy, and don't want to work more than they have to.
This is obviously a totally contrived model of an economy. Yet, you can work out that if Bob and Joe negotiate, each only pursing his own self-interest, they will eventually decide that Joe should concentrate on making A's while Bob should concentrate on making B's. Bob and Joe both wind up better off than they would be alone, even though Joe is better at everything than Bob! (Work out the math yourself, it's fun.) Thus, the silly assumptions teach us something very serious about trade and comparative advantage.
I don't get it - how helpful are "metaphors for things we think happen in the real world" if they aren't based in reality? Is that not like saying "let x be an odd number divisible by 2 [...] hence, taxes should go up!". Or am I misunderstanding the relationship between "proof" and "modeling".
Here's how it looks to me: economists believed that people are utility optimizers, and form a lot of conclusions based on the idea. When it is found untrue, economists modify what their theory is explaining (they say it's explaining "market psychology" or some nonsense) so that their conclusions still hold. It's not so obvious when the complicated workings of academics and their institutions behave this day, but if a single individual ever behaved like this, it would just seem stubborn.