Last modified: 2009-04-23 05:57:46 UTC
For example on meta: # 03:26, 25 March 2009: Krims (Talk | contribs | block) triggered filter 17, performing the action "edit" on User talk:Krims. Actions taken: Block autopromote; Filter description: Vandalistic edit summaries (details) (examine) But if you look at http://meta.wikimedia.org/w/index.php?title=Special:AbuseLog&details=139 there is nothing that should match filter 17. Furthermore, testing filter 17 at http://meta.wikimedia.org/wiki/Special:AbuseFilter/examine/log/139 shows that in fact the edit *doesn't* match filter 17 - so why is it logged?
quite serious if true
http://en.wikipedia.org/w/index.php?title=Wikipedia:Abuse_filter/False_positives&diff=prev&oldid=284509845 could be another one.
Given test case matches for me, it's because of a buggy filter. Remember to test against the version of the filter which was in effect at the time, in this case, http://meta.wikimedia.org/wiki/Special:AbuseFilter/history/17/item/90 ("X" rlike summary) will *always* match for empty summaries regardless of what "X" is. For the expected result, reverse "X" and summary.
An chance we can get some indication when the log entry applies to an old version of the filter rather than the current one? This is *very* confusing otherwise. Either the revision number or just top/not top would be excellent.