Jon, you might have missed this version of the query from Joseph, which should be more direct and faster (pasted below). However, I'm not sure you can deduce much about the session length from this result. It seems to me you'd have to group by IPs and track referrers over time. However, that doesn't really tell you much as lots of people show up with the same IP, IPs change over time (pretty quickly), etc. From Oliver's work on sessions, I remember that it's not really possible to find sessions over any significant period of time because the data changed too much. Talking to Oliver about this might help. That said, the percent of unique IPs that might be clicking through the site once or twice, according to this data, does not seem too high. I'd believe it. -- results for query below Total MapReduce CPU Time Spent: *3 minutes 56 seconds 660 msec* OK hits mobile_site is_internal num_pvs 5246 b false 8818 2878 b true 9100 31 a true 118 31 a false 64 SELECT COUNT(DISTINCT ip) AS hits, x_analytics_map['mf-m'] AS mobile_site, (referer_class = "internal") as is_internal, count(*) AS num_pvs FROM wmf.webrequest WHERE TRUE = TRUE AND webrequest_source = 'mobile' AND year = 2015 AND month = 5 AND day = 25 AND hour = 1 AND agent_type <> 'spider' AND is_pageview = TRUESELECT COUNT(DISTINCT ip) AS hits, x_analytics_map['mf-m'] AS mobile_site, (referer_class = "internal") as is_internal, count(*) AS num_pvs FROM wmf.webrequest WHERE TRUE = TRUE AND webrequest_source = 'mobile' AND year = 2015 AND month = 5 AND day = 25 AND hour = 1 AND agent_type <> 'spider' AND is_pageview = TRUE AND x_analytics_map['mf-m'] IS NOT NULL AND access_method IN ('mobile app', 'mobile web') GROUP BY (referer_class = "internal"), x_analytics_map['mf-m'] ORDER BY hits DESC LIMIT 50; On Wed, Jun 3, 2015 at 11:27 PM, Jon Katz &lt;jkatz(a)wikimedia.org&gt; wrote:

By the pigeon hole principle, yes, if we had 1000 total pageviews and 999 internal referrer pageviews, then it'd be one user. But that's not what we're seeing, we're seeing 5k or so distinct IPs hit 9k or so pages from the outside. Then 3k or so distinct IPs hit 9k or so pages with internal referrers. I was trying to make two points: * there could be many users for each IP and many IPs for each user, in fairly unpredictable combinations * Given the numbers we're seeing, it's more like some number wildly hovering around 1/2 of our users are clicking through and checking out a few other articles. This is a little different from normal, and that seems expected since these are beta and alpha users. When looking at mobile users in general, I get a much lower ratio of internal to external referers as I'd expect: unique_ips is_internal num_pvs 3213398 false 6778157 731019 true 2310153 SELECT COUNT(DISTINCT ip) AS unique_ips, (referer_class = "internal") as is_internal, count(*) AS num_pvs FROM wmf.webrequest WHERE TRUE = TRUE AND webrequest_source = 'mobile' AND year = 2015 AND month = 5 AND day = 25 AND hour = 1 AND agent_type <> 'spider' AND is_pageview = TRUE AND x_analytics_map['mf-m'] IS NULL AND access_method IN ('mobile app', 'mobile web') GROUP BY (referer_class = "internal") ORDER BY unique_ips DESC LIMIT 50;