There's a standard process we have for measuring each item. The goal is to flatten the shirt without pulling on the material.

Measuring clothing is really, really hard to do in a consistent way. In my original plan for the project, I was going to hire 5 task rabbits to measure everything, then throw out the min and max and take the median of the remaining three numbers. It didn't work. The average taskrabbit had a standard deviation of error of 0.3" for each shirt. That might have been OK if the errors were purely random, but there was a lot of systematic error as well.

Eventually I was able to find two people who could provide consistent measurements. They were within 0.1" of my measurement for each item. When there was a difference of more than 0.1" we would remeasure the shirt. In general I feel good about the data we published.