...I looked a bit at the UA distribution for action=watch excluding the proxy=Opera rows. Most of that seems to be with higher capability UAs, perhaps from within WebView components or perhaps from non-CTA contexts or perhaps both. Or something.

zcat /a/squid/archive/mobile/mobile-sampled-100.tsv.log-20140409.gz | grep -v 'proxy=Opera' | grep 'action=watch' | cut -f14 | sort | uniq -c
5 Mozilla/5.0 (Android; Mobile; rv:28.0) Gecko/28.0 Firefox/28.0
...<a list>

$ zcat /a/squid/archive/mobile/mobile-sampled-100.tsv.log-20140409.gz | grep -v 'proxy=Opera' | grep 'action=watch' | wc -l
153
# note the grep -v


-Adam


On Tue, Jun 10, 2014 at 10:45 AM, Adam Baso <abaso@wikimedia.org> wrote:
Looking at all non-JS devices is a bit challenging.

That said, Opera proxy market share of pageviews on mobile web is about 5% of mobile web total from what we've seen (X-Analytics of proxy=Opera on text/html responses for certain well-defined "obvious" pageview request paths). Add to Watchlist is the only button on the menubar for such Opera proxy-sourced users, whose experience is about the same as <noscript> or other devices where the RL bootstrapping process bars further JS hooks. Opera proxy-sourced usage of the Add to Watchlist feature is about 0.5% of total non-Opera proxy-sourced access.

$ zcat /a/squid/archive/mobile/mobile-sampled-100.tsv.log-20140409.gz | grep -v 'proxy=Opera' | grep 'returntoquery=article_action%3Dwatch' | wc -l
527
# note the grep -v for exclusion

$ zcat /a/squid/archive/mobile/mobile-sampled-100.tsv.log-20140409.gz | grep 'proxy=Opera' | grep 'action=watch' | wc -l
3
# note this is inclusive grep

No Opera proxy-sourced usage invoked unwatching an article for that particular day from the looks of it.

$ zcat /a/squid/archive/mobile/mobile-sampled-100.tsv.log-20140409.gz | grep 'proxy=Opera' | grep 'action=unwatch' | wc -l
0

I can't say for sure, but I suspect the trends for watch/unwatch are about the same on the other <noscript> and RL module load JS-barred UAs.

I can't remember if that was something that was tracked at the outset, but I wonder if adding EL for the higher-JS Add to Watchlist JSON calls (that is, the ones invoked on tap of the star /post-authentication/) would be useful longer run, as it seems like a kind of interesting metric to gauge interest and regression analysis as items for higher-JS browsers are added/subtracted/reworked on the menubar?  I couldn't tell if Add to Watchlist taps on higher-JS was part of http://mobile-reportcard.wmflabs.org/graphs/watchlist-activity or mobile-reportcard.wmflabs.org/graphs/ui-daily; those seem more related to actual Watchlist maintenance and sidebar usage, respectively, instead. Maybe it's somewhere in a wfDebuglog or backend EL table?

-Adam





On Tue, Jun 10, 2014 at 10:29 AM, Jon Robson <jdlrobson@gmail.com> wrote:
Maryana, I'm still not convinced that these usage stats tell the whole
story. Also when talk/Flow gets added, then there will be 2 buttons in
that bar, so I'm not sure the screen real estate issue is a good
argument. Would we also remove Flow from the menu? Removing the
watchstar from Opera only would actually be messy. The only correct
way to do this programatically would be to remove the page actions bar
for all non-JavaScript users (I refuse to have some nasty Opera
specific hack).

Some more questions and points:
* you should look for the unwatch action as well. 'action=watch' is
only for watching articles.
* What is the global usage of the watchstar without JavaScript? What %
of this is from Opera?

e.g.
$ zcat /a/squid/archive/mobile/mobile-sampled-100.tsv.log-20140409.gz
| grep 'action=watch'
$ zcat /a/squid/archive/mobile/mobile-sampled-100.tsv.log-20140409.gz
| grep 'action=unwatch'

On Tue, Jun 10, 2014 at 10:09 AM, Maryana Pinchuk
<mpinchuk@wikimedia.org> wrote:
> Thanks for digging into this, Adam. I don't think the prominence of the
> feature is justified by the usage stats, so I'd be in favor of removing it
> from Opera/lower JS devices.
>
>
> On Tue, Jun 10, 2014 at 5:45 AM, Adam Baso <abaso@wikimedia.org> wrote:
>>
>> Good question, will take a look once I'm at a place with stats cluster
>> access.
>>
>>
>>
>> On Mon, Jun 9, 2014 at 6:33 PM, Jon Robson <jdlrobson@gmail.com> wrote:
>>>
>>> How does this compare to usage of users on say Chrome...?
>>>
>>> On 9 Jun 2014 18:02, "Adam Baso" <abaso@wikimedia.org> wrote:
>>>>
>>>> Juliusz suggested I email out details to mobile-l on the following.
>>>>
>>>> The question arose during an Opera discussion today whether hiding the
>>>> Watchlist icon (which is the case on non-HTTPS supporting UX on Wikipedia
>>>> Zero) on mobile web in the page menubar (not the same as the flyout
>>>> "hamburger" menu) might make sense generally for <noscript> or lower JS
>>>> devices? The Watchlist star on the page menubar takes up a lot of space, and
>>>> as it's the only thing there at the moment (on en.m at least, icons like
>>>> Edit and Add Photo aren't shown), hiding that menubar icon would free up
>>>> some valuable screen real estate.
>>>>
>>>> On <noscript> or lower JS devices (or browsers where RL suppresses JS
>>>> due typically to challenges around timing of Deferreds and the like), using
>>>> the Opera traffic as an example of such a browser, it seems like Watchlist
>>>> usage is sort of low (this is at 1% sampling resolution).
>>>>
>>>> $ zcat /a/squid/archive/mobile/mobile-sampled-100.tsv.log-20140409.gz |
>>>> grep 'proxy=Opera' | grep 'action=watch' | wc -l
>>>> 3
>>>>
>>>> In other words, it seems users make it to the point of using the
>>>> feature, but only about 300 times per day total. Meanwhile, the Watchlist
>>>> start takes up valuable screen real estate for every pageview.
>>>>
>>>> The usage of the feature is about 1/10 of the Opera usage involving
>>>> submission of the login form (a prerequisite of watchlist usage).
>>>>
>>>> $ zcat /a/squid/archive/mobile/mobile-sampled-100.tsv.log-20140409.gz |
>>>> grep 'proxy=Opera' | grep -i 'Special:UserLogin' | grep 'POST' | wc -l
>>>> 31
>>>>
>>>> Which is about 1/10 of Opera usage of the login feature in any capacity
>>>> (GETting the form or POSTing the form)
>>>>
>>>> $ zcat /a/squid/archive/mobile/mobile-sampled-100.tsv.log-20140409.gz |
>>>> grep 'proxy=Opera' | grep -i 'Special:UserLogin' | wc -l
>>>> 331
>>>>
>>>> Which is maybe 1/270 of an oversimplified "pageview" metric on Opera
>>>> Mini, using text/html response types as a rough guide.
>>>>
>>>> $ zcat /a/squid/archive/mobile/mobile-sampled-100.tsv.log-20140409.gz |
>>>> grep 'proxy=Opera' | grep 'text/html' | wc -l
>>>> 89403
>>>>
>>>> The relatively low usage of the Watchlist feature is probably
>>>> symptomatic of the multiscreen flow on such devices.
>>>>
>>>> -Adam
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> Mobile-l mailing list
>>>> Mobile-l@lists.wikimedia.org
>>>> https://lists.wikimedia.org/mailman/listinfo/mobile-l
>>>>
>>
>>
>> _______________________________________________
>> Mobile-l mailing list
>> Mobile-l@lists.wikimedia.org
>> https://lists.wikimedia.org/mailman/listinfo/mobile-l
>>
>
>
>
> --
> Maryana Pinchuk
> Product Manager, Wikimedia Foundation
> wikimediafoundation.org



--
Jon Robson
* http://jonrobson.me.uk
* https://www.facebook.com/jonrobson
* @rakugojon