(Go: >> BACK << -|- >> HOME <<)

Page MenuHomePhabricator

CX2: Abuse filter unexpectedly triggered
Open, MediumPublic

Description

When trying to translate Auri (band) from English to Portuguese, adding the first content paragraph after the infobox results in an abuse filter which is totally unexpected: "Página nova com uso excessivo de maiúsculas" ("New page with excessive upper case usage").

As the screenshot below shows, the content added to the translation is far from being full of uppercase characters. So this suggests that the content evaluated by the abuse filter may be something different from what we see, and we may want to polish the workflow for abuse filter checking based on the underlying issue.

The is abuse filter uses the following regular expression:

!'confirmed' in user_groups
	& page_namespace == 0
	& page_recent_contributors == ''
	& new_size > 50
	& !new_text rlike '[a-z]'

Screen Shot 2018-10-22 at 15.38.04.png (656×1 px, 251 KB)

Event Timeline

I initially thought that this rule is completely broken, as it seems to match any new page which contains letters other than a-z. But on further thinking, this must be an unanchored regular expression, which means that it there isn't any letter a-z on the text, then it fires.

Nothing in Logstash. Debugging these is quite hard as validation is not doing on every save.