It's just a status quo bias to choose "act to preserve without prior of belief > 0.95". It's a novelty bias to choose "act to change without prior of belief > 0.95". But there are many modes of action in between. Knowledge is necessarily imperfect (cf. Conjectures and Refutations). The engineer/scientist difference isn't all that useful in reality. Even disciplines dominated by engineering rapidly see advances over time purely through process improvements which often manifest across the field as the hypothesis-test-apply loop of knowledge acquisition.

Public policy, like all decision making, should work with a gestalt view of utility and certainly most organizations attempt to do that irrespective of whether their spokespeople claim otherwise. It's like any intervention: if the patient will almost certainly die if you do nothing, you try the most unlikely things (assuming consent) but if the patient will most certainly recover with no ill effects you are not going to try an experimental intervention. And naturally you do account for change fatigue, etc. etc. Put simply, the science of policy is not as puerile as you think it is, but the nature of its execution is necessarily a complex interweave of the power centres available and their objectives and influence on the policymaking body.

After all, we don't all have the same objectives.