Basically, yeah. As a general rule, the bigger the change you want to make to existing practice, the better your model for that domain needs to be. (I call that the "if it ain't broke" heuristic.)
Changing the composition of an entire nation's schoolchildren's food supply ... should require something like skyscraper-failure-prediction accuracy.
It's no defense that "science learns and revises". If the learning and revising is still that big, you don't have a basis for telling others to do it your way.
(IMHO, this is, at root, the same mentality as "fake it till you make it" / "all self doubt must be Impostor Syndrome and never an accurate assessment".)
Changing the composition of an entire nation's schoolchildren's food supply ... should require something like skyscraper-failure-prediction accuracy.
It's no defense that "science learns and revises". If the learning and revising is still that big, you don't have a basis for telling others to do it your way.
(IMHO, this is, at root, the same mentality as "fake it till you make it" / "all self doubt must be Impostor Syndrome and never an accurate assessment".)