User:Dapete/Report on Flagged Revisions, December 14, 2008

Report originally written by User:P. Birken on the wikide-l mailing list on December 14, 2008

Translation (and a bit of wikification) by User:Dapete. It's a bit shaky, but I tried to stay close to the original. Feel free to improve it.

Software and Project Status
The software is now quite stable, with small changes every now and then. Almost no bugs are still found. Feedback to improve especially usability are still welcome.

The extension has been enabled for all German projects for some time so it can be activated after a Meinungsbild. The biggest projects who did this are the Russian, Polish, Hungarian and Esparanto Wikipedias as well as smaller projects like German Wiktionary or Allemanic Wikipedia. A full list can be found at de:Hilfe:Gesichtete und geprüfte Versionen.

Of 850,947 articles (not counting redirects) on German Wikipedia, 771,385 have at least one sighted revision. That's 90.65 percent. So there's only about 80,000 articles missing, by the end of January all articles should have been flagged at least once. There are 5,377 users with "sighter" status.

Waiting time of changes
See the first last image on http://toolserver.org/~aka/cgi-bin/reviewcnt.cgi?lang=german&action=images&project=dewiki. Shortly before the end of the Meinungsbild, when it became clear that the extension would be activated on German WP, several users invested a lot of energy (particularly Benutzer:Septembermorgen, who has been very active since then as well, thanks a lot to him!) to reduce the backlog of about 11,000 articles to nearly zero. Shortly after it went up again before it stabalized at 5,000 articles. Then the toolserver failed for a few weeks, which took away many tools. Also Aka's page http://toolserver.org/~aka/cgi-bin/reviewcnt.cgi?lang=german&action=overview seemed to cause a lot of motivation. Without it the backlog went up to 14,000, Since then it has been between 10,000 and 12,000. The median article waiting time listed there is about one week (the median time articles edited by non-sighters wait is unfortunately currently not recorded, but it is of course lower - most edits are checked by RC patrol, watchlists or wikiprojects/portals within minutes to hours).

Contrary to that the waiting time of Articles since the zero point early November grew unhindered. de:Spezial:Seiten mit ungesichteten Versionen did and still does not receive enough attention although it's linked from de:Spezial:Letzte_Änderungen. I noticed this too late, now there is a campaign (de:Wikipedia Diskussion:Gesichtete Versionen/Nachsichtung) to ensure that the waiting time doesn't exceed 21 days, so this has been stable since November 19, partially going down to 19 days. This is, in my opinion, a more important number than the number of articles with waiting revisions and is clearly too high.

Articles poling up there are usually only on less than ten watchlists (or completely unwatched) and from niche topics. The number of articles on places and small municipalities, which are either unsubstantial or replace their homepage, is remarkable. There are also many soccer players and soccer clubs as well as pop musicians. It should be noted that edits which waited for so long are not necessarily OK, about 20 percent were reverted by me and many further improved.

Contrary to that the whole thing works very well for some topics (medicine and maths). They seem to be the most active, and non-sighters receive feedback to the quality of their contributions very quickly.

Summary: Number of articles with unsighted revisions and waiting time of edits are stable or slightly falling on a relatively high level. Polish Wikipedia shows it's possible to do this differently: pl:Specjalna:Statystyki oznaczania, with only 70,000 of their 500,000 articles marked at least once.

Influence on edits and user registrations
Regarding the number of user registrations which dropped steeply some time more than a year ago, I have no current data, just as with the question whether new users stay. The number of edits on WP has significantly gone up since sighted revisions were introduced, the number of IP edits has gone down a bit, which has been the trend for over a year. So sighted revisions seem to have no effect on IP edits. That is, the number of edits from registered users has gone up significantly, I know of no breakdown for user groups and their experience.

Future prospects
Paradox has created interesting statistics showing the number of automatic and manual "sightings" over time. The upper diagram shows that automatic as well as manual flagging has been nearly stable since August and that there are only about 3,000 to 4,000 manual "sightings" per day. During the same time period the number of users with "sighter" permissions and the number of articles sighted at least once has about doubled. On the one hand this means that the next 80,000 arcitles still to be "inserted" into the system won't change whether the system scales or not. On the other hand it means that that a minor raise in number of flaggings per day (a few hundred) would significantly lower the backlog. That is, if all of the 700 readers of this list clicked "Sight" once more often each week or the seven most active users without "sighter" permissions got these permissions, it would have a significant effect. If done once a day the backlog would quickly go down.

Criteria for sighted revisions
On de:Wikipedia Diskussion:Gesichtete Versionen and the threads around it there is currently a discussion to more clearly specify the criteria for sighted revisions. How do the most active "sighters" interpret the diffuse phrasing that "an experienced Wikipedia author has looked at it and the revision is free from obvious vandalism." The goal is first to counter the criticism that sighted revisions are just "free of vandalism" and, for that reason, not very useful; and second to give users who don't want to do anything wrong but don't know exactly what to do some guidelines, by giving a more precise description of what is expected [from sighted revisions].

Criteria for "sighter" permissions
Following the discussion already mentioned, an assessment of the current practices to give users "sighter" permission is useful. (Too restrictive, too generous?)