As a general rule, absolutely! Unfortunately, if applied too strictly though, it can be weaponized or stifling in ways you may not anticipate (which tends to be true of almost any rule used too universally).

For example: I’ve seen cases where there’s disagreement as to which of two technical paths to pursue, where one side (not liking the way consensus is currently leaning) demands a data driven proof that one of the choices is best. Ordinarily this would be fine, but in this case, it would take no less work/time to gather said data than to entirely implement both solutions and directly compare them.