@MWang (WMF) -- thank you for doing this analysis. I'm somewhat surprised to see that substantial portions of accounts are created in the apps (~10%) even though a much lower share of first edits are done in the apps. I think that sign ups for reading lists might explain this.
But my only real question is about this graph. How can registrations occur through an "other" app? Since we only have Android and iOS apps?
Thanks! I agree that signing up in order to enable reading lists (or other apps features, I am not very familiar with their capabilities) can explain why we see about 10% of creations happening that way.
The way I identify "apps" in the analysis is simplistic, it only looks at the status of the isApi flag, because my exploratory analysis suggested that apps use the API for account creation (while the mobile and desktop sites use a different process). This means that if the API is used for account creation, e.g. to create several accounts for an event, this will also fall into that category. When I then split it by looking at the os_family part of the user-agent string, those API account creations are neither iOS nor Android and fall into that "other" category.
So it's not the case that those accounts are created in apps, but just look that way because of how the categorization works. I'll make a note to update the page to clarify that as I can see that being very confusing.
