If you don't perturb the initial numbers to see what changes to help construct a richer model (as well as one with tighter bounds on what it predicts or forbids), you're leaving out a lot of the benefits of the exercise. Sometimes the prior doesn't matter that much because you find you already have or can easily collect sufficient data to overcome various arbitrary priors. Sometimes different priors will result in surprisingly different conclusions, some of them you can even be more confident in ruling out because of absence of data in their predictions. (Mathematically, absence of evidence is evidence of absence, though of course it's just an inequality so the proof says nothing on whether it's weak or strong evidence.) And of course some priors are straight from the butt but others are more reasonably estimated even if still quite uncertain; in any case much like unit testing of boundary conditions you can still work through a handful of choices to see the effects.

Someone mentioned fermi calculations, a related fun exercise in this sort of logic is the work on grabby aliens: https://grabbyaliens.com/

> a consistent set of numbers has carried through the calculation, even if those numbers were wrong at the outset

I kind of see how this might be useful, but what I've actually seen is an illusion of certainty from looking at numbers and thinking that means you're being quantitative instead of, you know, pulling things out of your butt. Garbage in, garbage out still applies.

9/12

3 months have passed, 9 to go :)