A somewhat cheeky subject line, I know, but I wanted to get people's attention. Let me explain:
My team's job is to run experiments that tell us why Wikipedia editors stay or leave.[1] One of the conclusions we've come to is that we have zero data on how site performance impacts editor retention (IP editors included).
We could throw all our energy into building cool new features, but if people still have a frustrating experience because contributing is appreciably slower than reading (for purely technical reasons), we have no idea what the net loss is. We really need to know what the numbers here are, not just assume that slower is bad in an unquantifiable way.
We could get a measure of this by artificially slowing down the site for a subset of users, but we'd rather not do that for obvious reasons. So my proposal is this: if you're going to deploy anything that you think might effect English Wikipedia site performance, positive or negative, tell us beforehand and we'll measure its impact on users for you.
-- Steven Walling https://wikimediafoundation.org/