Ask HN: How does your team prevent “gotchas” during implementation?

Estimating is an art - there will always be unforeseen obstacles when working on projects. There's a wide attack surface for this problem, from "keep your codebase clean in the first place" to "allocate time for investigation" to "get better at rolling with the punches." What's your team's strategy and why does it work?