Apparently, as a result of a discussion on migration options it appears that Yannis and myself see different statistics when going to the programmers.SE.com migration stats tool and the stackoverflow.com tool respectively.
while I see:
Why are the rejection rates so radically different? Which is correct? (Or am I simply misinterpreting the data somehow and these are representing different statistics)
I assume it's a bug that counts question closed as duplicates "rejected" on the destination site, but not on the source site.
Not counting them as rejected would be correct behaviour since they don't get locked like actual rejects. Another reason against counting them as rejected is that they're on topic on the destination site and closing those as duplicates is desirable.
We had a related discussion some time ago:
What outgoing migration paths should we have?
"migrations are considered to be rejected by the destination site when migrated questions are closed as anything other than "duplicate" or deleted." That doesn't appear to be true. If you look at the rejection stats for security->crypto they can only be explained if duplicates are considered "rejected". (6/14=42% vs. 2/14=14%) – CodesInChaos Jun 29 at 10:29
@CodesInChaos There is some weirdness in how we calculate the rejection stats. You're seeing a 40% rejection rate on Crypto, but Security is showing 15%. However, that bug aside, closing as a duplicate doesn't reject a migration. When a migration is rejected, the question is locked on the destination site and unlocked/closed as off-topic on the source site. The redirect link between the two is also severed. This doesn't happen with questions closed as dupes on the destination site. – Anna Lear♦ Jun 29 at 15:53
Moderators at source and destination sites need a reliable common ground to discuss migrations. Currently, stats fail to serve this purpose, because inconsistencies in these introduce unnecessary confusion.
Based on current system behavior (which in turn seems to follow long established tradition) it looks incorrect to count duplicates as rejected migration.
For a destination site visitor, non-dupe closures carry clear signs of rejection: question is locked, answers from source site are removed - in-site closures just don't look like that.
As opposed to this, duplicates look the same for in-site and migrated questions: there's no lock, all answers are preserved. Add that migrated duplicates (just as in-site ones) remain eligible for community reopen and aren't targeted for roomba and this becomes looking as close to "accepted" as it gets.
Another thing worth taking care of is stats regarding delayed rejections. If the question is closed a year (two, three...) after migration, this should be reflected consistently in stats at source and destination sites. This is, again, to ensure that moderators at both sites have reliable common ground to discuss migrations.
Side note - comparing migration rejection stats at various (preferably all) pairs of source-destination sites looks like a useful automated test to me. Test pass criteria should be the same values obtained at both sides - meaning moderators at both sites will see consistent stats. If there is a difference, something has gone wrong in stats calculation at least at one of the sides.
