A confounder is a hidden factor that affects both of the things being compared, making a link between them look stronger or weaker than it really is. For example, if teams that use one repository layout also happen to be smaller and more experienced, those traits may explain better results and not the layout itself.
Confounders matter whenever someone claims that one practice caused an outcome. Observed differences between teams, projects or periods often come with many other differences, so a correlation alone does not show cause.
They are hard to remove completely. Collecting more data does not fix the problem by itself, and a factor that was not measured cannot be adjusted for. Careful comparisons try to reduce them by matching similar teams, comparing the same team before and after a change, adjusting for known factors or, where possible, running a controlled experiment.
When you read a study or a team's report, ask what else differed between the groups, whether the comparison is fair and whether the conclusion goes beyond what the data supports. When you do your own analysis, list the plausible confounders in advance and say which ones you could and could not account for.