YoY in June it is a 61% drop for rowiki, 53% for huwiki, 42% for frwiki, 34% for dewiki. As mentioned in the mailing list, it is entirely driven by the mobile web. The pattern is the same with agent=user and agent=all so should not be caused by bot misdetection.

Also, the decrease started in early 2016, not this April, and has been pretty linear since then. Since the drop has been gradual and there has been no change in traffic labelled as spider/bot (and its tiny compared to human mainpage traffic anyway), I don't think there is every reason to assume the decrease is real. Maybe these wikis did something to their start page that looks particularly horrible on mobile, or users were motivated to avoid it in some other way (better search? something something handwawe Google?)

Since the drop has been gradual and there has been no change in traffic labelled as spider/bot (and its tiny compared to human mainpage traffic anyway),

There has been ongoing work to better catch self-identified bots. The ones not label as such are still being labelled as user traffic.

The main page is also requested when you query, say, http://ro.wikimedia.org so it receives a lot of "unintentional" traffic

Agreed with @Tgr that decrease started earlier but it is more pronounced from April onwards, I have added es.wikipedia to plot for comparison. Detailed plot shows the ro and hu wikipedia's which are much smaller in size.

These plots INCLUDE bot traffic

On further inspection there is definitely a seasonal pattern here in which pageviews are smaller in the summer months.

See seasonal pattern for hu.wikipedia and ro.wikipedia

@Strainu I really do not see anything major here and if I look at series in its entirety the oscillations do not seem so far fetched.

@Nuria, are the stats you show the pageviews for the whole project or for the main page? They don't look anything like the pageviews tool. There is no seasonal variation there, only decreases from March 2016.

@Strainu : argh i deleted part of my comment but we had an issue with organic but not intentional traffic on the main page due (we believe) to a chrome/win issue https://phabricator.wikimedia.org/T141506

So are you saying that it's just "random" browser traffic that used to be targeted at the MainPage and now it's not? What kind of script could have generated that?

@Strainu: pageviews are for the whole project, equivalent on pageview tool: http://tools.wmflabs.org/siteviews/?platform=all-access&source=pageviews&agent=user&start=2015-07-01&end=2017-07-31&sites=ro.wikipedia.org

@Strainu it was not a script, it is a connection failure from the client when stablishing an https connection (at least that caused a similar problem in other sites on a similar date)

I'm sorry, but I'm a bit skeptic about this explanation for several reasons:

• ro.wp did not have that peak in July 2016; it had some other peaks which probably account for the slight rebounds in monthly pageviews and which could be caused by the same issue
• it seems highly unlikely that low-level errors would only occur on one page. Despite that, the second most visited page, Search, did not see any spike or drop except normal summer vacations. Nither did the most visited article.

Before closing this bug I'd like ask you to do 2 experiments, if you have the time and the data allows:

• drill down siteviews per namespace and see if there was a significant increase in another namespace (Wikipedia or Special are the suspects).
• check out the main landing pages for visitors and see if there was any significant shifts in that list

My hypothesis is that Google or another external tool changed something and now people are sent, either directly to articles, or to some other landing page, instead of seeing the MainPage. Knowing if this is the case is important, as it allows us to concentrate our efforts on that other page. Thanks!

@Strainu:
Sorry for my deletions that make this ticket harder to read. Let me try to summarize.

What I see:

• Traffic overall for those sites is slightly lower that same time last year but it follows the same pattern of lower summer traffic. True of pageviews and unique devices:

https://analytics.wikimedia.org/dashboards/vital-signs/#projects=huwiki,rowiki/metrics=Pageviews
https://analytics.wikimedia.org/dashboards/vital-signs/#projects=huwiki,rowiki/metrics=UniqueDevices

• As @Tgr commented the issues with Main_Page are mostly due to lower pageviews for that page on the mobile sites, desktop site pattern is quite different:
  • http://tools.wmflabs.org/pageviews/?project=ro.wikipedia.org&platform=desktop&agent=user&start=2015-07&end=2017-06&pages=Pagina_principal%C4%83
  • http://tools.wmflabs.org/pageviews/?project=hu.wikipedia.org&platform=desktop&agent=user&start=2016-01&end=2017-06&pages=Kezd%C5%91lap

• As an FYI, we have seen before Main_Page counts being distorted because of connection issues, this might be a factor.

Your comment:

it seems highly unlikely that low-level errors would only occur on one page

I think you misunderstood, connection errors occur on main page because that is the one serve by default on our root domain, there is not such a url as http://en.wikipedia.org, such a request will result on an internal redirect to our Main_Page. So errors on establishing a connection most definitely occur when requesting that page. For details you can read this ticket: https://phabricator.wikimedia.org/T141786

Your request:

check out the main landing pages for visitors and see if there was any significant shifts in that list

Referrer data for that time is no longer available, it is discarded after 90 days,

Aggregated data on traffic coming from search engines is available here: http://discovery.wmflabs.org/external/#traffic_by_engine
The magnitude is large so variations as small as the ones that affect these sites might not show

My hypothesis is that Google or another external tool changed something and now people are sent, either directly to articles, or to some other landing page, instead of seeing the MainPage.
Knowing if this is the case is important, as it allows us to concentrate our efforts on that other page.

"other landing page" seems a little far fetched, we only have 1 landing page and that is the one our root domain redirects to, the "Main_Page".
Now, when looking at traffic for this Main_page I would be very skeptical of numbers and variations, while the traffic is real is not necessarily "intentional" .

You can plot request per article for one of those wikis and see for yourself if any article stands out, data is available here: https://dumps.wikimedia.org/other/pagecounts-ez/

I don't have the time or resources needed to do the investigation myself. Thanks for your help on this, I believe we can now close the ticket.