Duct tape is correct during validation: fast, cheap, unashamed. Load-bearing engineering is correct after validation: built to carry the weight the validated numbers require, with a safety factor.
Most expensive failures come from one of two confusions: engineering before validating (polishing something whose biggest assumption is still standing), or hacking long after the evidence demanded better.
The skill is not choosing a mode once. It is knowing, at any moment, which mode the evidence currently requires, and switching without sentimentality when it changes.