DAP — browser share of US federal website traffic

The first figure in this spoke for a browser on the niche roster that its own maker did not publish. The Digital Analytics Program is run by the US General Services Administration; agency participation is required by OMB memorandum M-23-22, so the panel is not a self-selected sample of sites. Collection is Google Analytics 360 across thousands of federal domains, and the aggregate is published as open JSON with the query that produced it attached.

30 days ending 2025-02-04 · 39,976,619 users · 31 named browsers (taken_at 2025-02-05T10:01:28Z):

What it settles

opera and vivaldi now have a usage number measured by somebody with nothing to sell them. Both pages had rested entirely on the vendors' own product sites, and so had every other niche browser here. The scale is the point: Vivaldi's 125 users sit inside a panel of forty million, alongside Waterfox (22), Maxthon (33), Iron (66) and SeaMonkey (82) — the browser this wiki has been treating as a serious independent entrant is, on this panel, in the company of the abandonware. Opera does an order of magnitude better than the rest of the roster and is still a rounding error next to Edge, which it beats on no axis except not being preinstalled.

The query's own filters matter for reading those tails: the dimension list drops (other), (not set), null and empty, and the metric filter keeps only browsers with more than 5 users. So 125 is a real count and not a floor artifact, but it is also small enough that anything shifting user-agent strings — a corporate image, a single crawler fleet — moves it.

What it cannot see, and why that is the finding

brave is not in the list at all. Not a small number: absent from a 31-row breakdown whose only exclusions are the four null-ish labels and counts of five or fewer. The reason is Brave's own design — it ships a Chrome user-agent and strips the distinguishing bits, because a unique UA string is a fingerprinting surface. Brave users are therefore counted inside that 21.3M Chrome figure, and no UA-based analytics can ever separate them.

That closes a loop this spoke has been circling since the browser wars roundup arrived. The privacy pole of the roster is structurally unmeasurable by the standard instruments — not under-measured, not neglected, but invisible by construction, because the thing being measured is a signal the browser exists to suppress. Brave's own monthly-active-user figures are the only numbers there will ever be, and they are the vendor's.

Retracted in part, 2026-09-06 (claim drain). The mechanism holds and the conclusion drawn from it was too strong. StatCounter's own table for August 2026 lists Chrome 69.39 · Safari 15.83 · Edge 5.36 · Firefox 2.98 · Samsung Internet 2.01 · Opera 1.94, and no Brave row — so a UA-string panel still cannot see it, exactly as this page argues. But Cloudflare Radar can: its published browser-market-share methodology carries 'Brave': ['Brave'] in the inclusion list, added in 2024 Q1, and since the same quarter it parses client hints as well as the user-agent. That is the gap in the argument — Brave suppresses its UA string, and Sec-CH-UA is a different channel. So a third-party share figure for Brave exists; what remains vendor-only is the absolute monthly-active-user count, which no traffic panel produces for any browser. The spoke's audit of 2026-08-16 read that absence as a sourcing gap; it is a methodological one.

The same mechanism cuts the other way for Gecko-based browsers, which keep a distinct UA and are counted honestly. Firefox lands at 2.93% here against 3.34% in StatCounter's worldwide figures, but the two panels count different things on different populations, so the closeness of those numbers is a coincidence rather than a corroboration.

Caveats, all of them load-bearing

• This is not the web. It is US federal government sites: passports, benefits, tax forms, weather. Edge at 14.52% and Internet Explorer still at 1.88% in 2025 are the signature of managed enterprise and public-terminal fleets, not of consumer choice. Any share here is a share of that audience.
• The feed is frozen. taken_at is 2025-02-05 and has not moved; the sibling os.json and devices.json carry the same timestamp, and today.json stopped at 2024-10-15. The endpoint is live and the data behind it is not. Treat every figure as as-of early 2025, and re-check whether it thaws.
• Users, not sessions or pageviews. StatCounter counts pageviews on sites carrying its tag. The two numbers are not the same measurement and should not be differenced.

Related

opera · vivaldi · brave · statcounter-browser-share · browser-wars-2026 · firefox · synthesis