I've run a quick simulation of the first-pass automatic conflict resolution for the future single login migration:
Total registered usernames: 2,097,231 No edits on any wiki: 1,475,146 (70.3%) Only registered on one wiki: 1,918,547 (91.5%) Automatically resolved: 139,466 (6.6%) Potential conflicts: 39,218 (1.9%)
Accounts can be automatically resolved ahead of time where all of the instances of a name either have the same e-mail address listed (in which case the owner could reset the password to the master account's) or have no listed contributions (in which case we consider them fair game for reclaiming).
Potential conflicts are where one or more accounts exist for a name which have some contributions listed, but the e-mail address doesn't match that of the primary account. Many of these will be in fact owned by the same user, but just didn't get the mail setting filled out.
Due to the salting of passwords, we can't currently perform an offline check for matching passwords; however when the user first logs in after migration, the passwords can be checked and any matching accounts can be automatically resolved at that time.
In remaining cases where the passwords don't match, users will have to manually select merging (if it's their account) or renaming (if it's someone else's) to clear all conflicts.
Most such conflicts should be resolvable without too much administrator intervention, at least in theory.
I'm running another pass to gather more statistics on how many potentially conflicting accounts are barely-used versus often-used, will report in the morning.
(I've put some notes and the demo code I'm using for building these stats in the CentralAuth project dir in SVN.)
-- brion vibber (brion @ pobox.com)