How do you evaluate a prototype when lab studies don't match real use?

I keep running into the same problem: a prototype performs well in a controlled lab study, but once people take it into their own homes or workplaces, the results look very different.

I'm curious how others decide when lab findings are enough to publish and when they need to be tested in the wild.

How do you design studies that capture real-world use without losing the control that makes results credible?