If you're trying to compile a set of facts, you very likely want a system which is incomplete, but correct. ie false positives have a significant cost to your models, where you essentially have a technical debt of needing to weed out false facts
Policy should follow risk analysis, but research which is not time sensitive has compounding factors which favor building certainty before skipping ahead
Think of it like software development: planning/testing/etc have high rewards throughout development, but sometimes you need to sell a rushed MVP
If you're trying to compile a set of facts, you very likely want a system which is incomplete, but correct. ie false positives have a significant cost to your models, where you essentially have a technical debt of needing to weed out false facts
Policy should follow risk analysis, but research which is not time sensitive has compounding factors which favor building certainty before skipping ahead
Think of it like software development: planning/testing/etc have high rewards throughout development, but sometimes you need to sell a rushed MVP