Corpus updates

[JanWMF]

Moin,

I toyed with your current version - thank you for putting it together - and I tried to revert the IP's edit detox identified for me as current and toxic but in reality a user did already detox modestly quickly through another pathway after the IP posted:

https://en.wikipedia.org/w/index.php?title=User_talk:94.119.64.26&action=history

I could reproduce the issue several times, which suggests to me:

1. the ML does an awesome job finding concerning edits.
2. the app then apparently currently can't keep track of whether community later detoxed independently of the app.

which suggests the corpus might not be dynamically updated - aside from the ML finding new issues and feeding them into the interface once. Over time, that means the false positive rate - not for whether the concerning edit was an issue but whether it (still) needs to be detoxed - goes up.