As opposed to continually revised building and safety codes? I think you're perhaps oversimplifying the construction world.

Regardless, if you've built three dozen screens and forms and API calls for your current application, you should be able to reliably estimate the next form and API call.

Part of the problem, IMO, is that we're too concentrated on being "agile". We have a sprint planing meeting where someone shows a screen and gives you 60 seconds to estimate it. Little time is spent these days on analyzing requirements and creating accurate estimates.

But the real problem is that most software "engineers" are anything but. Few have formal training. Few study anything past what's needed for their current requirements. Most don't even bother to learn the full ins-and-outs of their chosen platform.

Put them together, and is it any wonder that most estimates are wrong?