And all that should be compared to how much bugzilla search was used.
Points 2 and 3 may be difficult to reconstruct from requests logs, though: (2) requires knowledge of sessions; parameters for (3) are sometimes in POST data, IIRC.

Well, the greater problem is that it requires bugzilla hits/searches
to have been logged in the request logs; at this point we're talking
the sampled logs, and I'm not sure if Bugzilla is handled by the
"misc" varnish machines or not. If it is, I don't know if we'll have
data greater than a couple of weeks old.

In T88863#1099461, @Aklapper wrote:

I'd be also curious about this but I have no idea if anyone plans to commit time...

How would someone who does get access to the data?

I imagine we could either use DB queries (in which case we need to to come up with a very reliable filter to make sure they do not contain emails, password hashes etc), or install EventLogging into Phabricator (either on client or server side). Of course, upstream might know of a better way.

This is definitely an upstream task. I'm pretty sure our already stretched too thin design research/user research team will never put this in their queue; they're busy making wikimedia user sites better :)

I second Andre's suggestion of declining.