WPT: An Overview and History (Bocoup)

Boaz Sender, 2 December 2024. The account of web-platform-tests — what it holds, who maintains it, and how the results get published — written by one of the organizations that maintains it.

T2, with the conflict stated. Bocoup is a participant in the project it describes and in interop-2026, so this is not a disinterested audit. What it supplies is scale and history, both checkable, and it is the only place this spoke’s conformance numbers get a denominator.

The denominator

“Over 56,552 tests and counting (with 1.8 million subtests)” at the time of writing. That is the figure kitesurf‘s “215,000+ Web Platform Tests” should have been quoted against — and the mismatch of units is itself the lesson, since a count in the hundreds of thousands can only be subtests, against a subtest total in the millions. The suite grows continuously, so any pass count without a date and a denominator decays into a marketing number.

Where it came from

WPT began in 2010 out of the W3C HTML Working Group’s test suite and was renamed from html-testsuite in 2013. It is now maintained by standards developers from Apple, Bocoup, Google, Igalia, Microsoft, Mozilla and the W3C, with a Core Team established by contributor consensus in January 2019, membership by invitation.

The mechanism that makes it work is the two-way sync: tests written inside a browser engine are upstreamed to WPT, and tests added to WPT are downstreamed back into each engine. Vendors are not being audited by an outside body — they are pooling the test suites they wrote for themselves. The infrastructure behind wpt.fyi has run the suite in CI every couple of hours since it was built starting in 2017.

The metric was chosen to prevent a headline

The article is explicit that an interoperability metric is computed rather than per-browser pass/fail rates, so that “everyone wins as interoperability increases” and the tech press cannot turn the dashboard into a league table.

Worth holding beside what the spoke already knows about vendor numbers. Every performance and capability claim here — kitesurf‘s benchmarks, chrome-built-in-ai‘s, opera-neon‘s — is the vendor measuring itself on ground of its own choosing. WPT is the one measurement in this domain where the vendors gave up that freedom in advance, and they did it by agreeing on the scoring rule before the scores existed. The cost of that arrangement is visible too: a shared metric designed so nobody loses is also a metric that will not tell you which browser is better, which is the question a reader usually arrives with.

For the current agreement and its focus areas, see interop-2026.