I think even in software engineering, overengineering can be measurable. For example, let's say an API has an endpoint that responds on average within 10 ms, bu

I think even in software engineering, overengineering can be measurable. For example, let's say an API has an endpoint that responds on average within 10 ms, but it is only called once per hour. Spending resources to decrease that response time, if the latency can be afforded, would be overengineering (or maybe just premature optimization), but you could measure the increase in performance. Also side note, overengineering hardware is often good practice. In the examples you gave (i.e. a bridge), a 2x "factor of safety" could be considered low.