A week ago I published new wikistats files, for the first time in 7 months, only to retract them 2 days later, when it turned out that counts for some wikis were completely wrong. After some serious bug hunting I nailed the creepy creature, that had been hiding in an unexpected corner (most bugs find refuge there and yet it is still the last place to look).
New files have been generated and uploaded for Wikibooks, Wiktionary, Wikinews, Wikiquote, Wikisource, Wikiversity and Wikispecial. The wikistats job for Wikipedia has another 10-12 days to go, but intermediate results have been published, and I will refresh these every one or two days till new counts for all Wikipedias are online (as always with English as exception).
What went wrong
Here is a short explanation of what went wrong: the wikistats job parses all language specific message files (PHP code) to harvest localized version of certain keywords, like #Redirect, Image, User. Only by scanning both for the original English keyword #Redirect and the localized version (e.g. Swedish #Omdirigering) can it determine whether an article is to be counted as proper article.
These language files had been moved and partially restructured earlier this year. I had updated my code, but for some wikis the new code failed to locate the proper keywords, and instead returned value 0 (not as a status code, just as data). So for those wikis articles that contained ‘#Redirect’ or ’0′ were not counted, easily skipping half of the content. Something similar had occurred for image link counts.
I disabled the original blog post until the counts for all Wikipedias are up to date. Only then can I revise conclusions and screen shots used in the post.
21 Jan 2009: I posted a separate updated version of this crippled post.