We should have a way to visualize weekly and monthly data.

The following are my asks to make this task a success so I can stop running my own queries every month. These reports currently fulfil the needs of designers, Performance Team, MediaWiki frontend developers and volunteers, and various interested product managers and department heads that routinely use my reports

• Publicly accessible url.
• Aggregation per-month and ability to view data from previous months (going back at least 5 months).
• Percentage breakdown of agent_type=user page view counts by browser family. (without browser version or OS) – e.g. "Chrome: 32%". As raw data (html table or csv file), and perhaps in some visualised way like a pie chart, line chart or some other graph.
• Percentage breakdown of agent_type=user page view counts by browser family and browser version. (without OS) - e.g. "Chrome 46: 23.6%".

There are many other data points one could include that would perhaps be interesting and other aggregations (e.g. weekly) but those are not requirements. After the above requirements, as secondary priority I'd be most interested in:

• Per-country breakdown. Global averages can give a skewed perception (e.g. a browser used globally less than 0.05% might seem unimportant, but if there is a single country where it is used by over 10%, it might change things).
• Percentage breakdown of unique visitors by browser family and browser version. Unique visitors rather than page views, if that is possible?

Other less-important ones:

• by mobile/desktop site.
• by source (desktop, mobile web, mobile app, zero). Especially interesting for zero I think. For informed decisions in that area without clobbering with global averages to really expose any major browsers we should support if they are significantly more common among zero users.
• by mobile/desktop user agents.
• by operating system and version (e.g. os_family and os_major).
• by devices family (e.g. device_family).

From designers, but also developers perspective there's another subset of important data points in my opinion:

• viewport properties like viewport size, device pixel ratio (despite all it's historical bugginess)

When we're talking about possibly looking into inlining CSS for performance reasons, which predominantly targets above the fold content, we should try to get some insight what above the fold means.

@Krinkle, I'm trying to start work on this now, but there may be too many interruptions until after the dev summit.

In T118329#1885357, @Volker_E wrote:

From designers, but also developers perspective there's another subset of important data points in my opinion:

• viewport properties like viewport size, device pixel ratio (despite all it's historical bugginess)

When we're talking about possibly looking into inlining CSS for performance reasons, which predominantly targets above the fold content, we should try to get some insight what above the fold means.

@Volker_E: that data sounds useful. I suggest making an Event Logging schema and logging this information for a small percent of our traffic.

just adding this here: https://www.netmarketshare.com/browser-market-share.aspx?qprid=2&qpcustomd=0

• tabular layout on dashiki (similar to http://reportcard.wmflabs.org ) (design a simple to edit config data schema) [8 pts]

• make a graph to show hierarchy (pie-tree?), two levels (browser and major version) [13 pts]

• textual "visualization" [3]

• bookmarks for individual graphs []

Change 267045 had a related patch set uploaded (by Milimetric):
Implement a Tabular Layout

https://gerrit.wikimedia.org/r/267045

Change 267045 merged by Nuria:
Implement a Tabular Layout

https://gerrit.wikimedia.org/r/267045

• Time series (line graph?) for browser family (going back at least 3-5 months)
• Time series (line graph?) for browser family + version (going back at least 3-5 months).
• Percentage breakdown of browser family – e.g. "Chrome: 32%". As raw data (html table), and visualised.
• Percentage breakdown of browser family + version.- e.g. "Chrome 46: 23.6%".

Per meeting with @Milimetric and @mforns: The percentage breakdowns could be a single pie graph that has an interactive interface starting at browser family, down to browser version, os family, os version ("Sunburst"). And perhaps another one that starts at os_family (down to os_version, browser family, browser version).

I'm sorry, I should've kept this open as a tracking task and made sub-tasks. This had two purposes - the broken down tabular layout by itself, and a de-facto tracking task. Even though it's closed for process purposes, I'll still use it as a tracking task for the other things that need to get done.