==Overview==

When an interwiki import happens, edit history is brought over from one wiki to another. By default, performers of the imported edits appear with an interwiki prefix, like meta>Example. This is documented at https://www.mediawiki.org/wiki/Help:Import.

• See an on-wiki example of an imported history, here.
• See how that example shows up in our core tables data, via this query
• See how that example shows up in Mediawiki History, via this query

(Fyi, we experience the same issue for other cross-operations. The one we found in T425443 relates to users changing user-groups from centralauth – on metawiki – impacting users on possibly all sister projects.)

For the instances of interwikis imports of revisions, the imported revisions show up with usernames presented in a couple of different ways, such as with

• wiki prefix (e.g., en>USERTEXT , de>USERTEXT, meta>USERTEXT , strategywiki>USERTEXT)
• “import” prefix (import>USERTEXT)
• letter prefix (e.g., b>USERTEXT , w>USERTEXT )
• miscellaneous prefix (e.g. regiowiki.at>USERTEXT , *>USERTEXT)

==Implications==

These imported revisions may be inflating anonymous revision counts, as well as revision counts that don't exclude anonymous revisions. Why? These imported revisions all show up in mediawiki_history as event_user_is_anonymous = TRUE

SELECT event_user_is_anonymous, COUNT(*)
FROM wmf.mediawiki_history
WHERE snapshot = '2026-04'
 AND event_user_text LIKE '%>%'
 AND event_entity = 'revision'
 AND event_type = 'create'
GROUP BY event_user_is_anonymous

The imported revisions share the same sha1:

SELECT wiki_db, page_title, page_id, event_timestamp, event_entity, event_type, event_user_text, event_user_is_anonymous, event_user_is_permanent, revision_text_sha1
FROM wmf.mediawiki_history
WHERE snapshot = '2026-04'
AND wiki_db IN ('dewiki', 'enwiki')
AND page_title = 'Battle_for_Dream_Island'
AND event_timestamp < '2026-03-10'
ORDER BY event_timestamp DESC , wiki_db
LIMIT 50

Now that temporary accounts have rolled out, temp account revisions that are imported are labeled as event_user_text_is_anonymous = TRUE and event_user_text_is_temporary = FALSE:

SELECT wiki_db, event_entity, event_type, event_user_is_permanent, event_user_is_anonymous , event_user_text
FROM wmf.mediawiki_history
WHERE snapshot = '2026-04'
 AND wiki_db = 'enwiki'
 AND event_user_text LIKE '%>~%'
 AND event_entity = 'revision'
 AND event_type = 'create'
LIMIT 50

==Impact==

If we look at overall impact, per the following query, dewiki seems to be affected the most, with more than 6 million rows in MWH having a revision done by a user with event_user_text LIKE '%>%’. That's spanning all years.

SELECT wiki_db, event_entity, event_type, COUNT(*) as count
FROM wmf.mediawiki_history
WHERE snapshot = '2026-04'
 AND event_user_text LIKE '%>%'
 AND event_entity = 'revision'
 AND event_type = 'create'
GROUP BY wiki_db, event_entity, event_type
ORDER BY count DESC

Here are the top five rows of that query’s output:

If we look at the monthly impact, we can see e.g. that for January 2026, dewiki was the most impact, with 5,033 rows in MWH having a revision done by a user with event_user_text LIKE '%>%’. That’s spanning all years.

SELECT wiki_db, substr(event_timestamp,1,7) as month, event_entity, event_type, COUNT(*) as count
FROM wmf.mediawiki_history
WHERE snapshot = '2026-04'
 AND event_user_text LIKE '%>%'
 AND event_entity = 'revision'
 AND event_type = 'create'
 AND substr(event_timestamp,1,7) = '2026-01'
GROUP BY wiki_db, substr(event_timestamp,1,7), event_entity, event_type
ORDER BY count DESC

Here are the top five rows of that query’s output:

Impact note: It should be noted that many of these instances may occur on pages that were moved and the original page was deleted; pages where the original page (translated from) gets deleted; or translated pages (i.e. the new page) that subsequently get deleted. This will affect how these edits show up – or don’t show up – in various counts.

==Suggestions==
We might consider

• Adding an event_user_is_cross_wiki field to make this instances explicit, as suggested in T425443
• Whether or not we should exclude instances where event_user_is_cross_wiki = TRUE in downstream tables, including the wikistats tables and analytics tables that have edit counts (e.g. geoeditors monthly, edits hourly, etc.)