Interop 2026
The browser vendors’ own annual agreement about which parts of the web they will make work the same way, published by the web-platform-tests project and scored in public. Run by representatives of Apple, Bocoup, Google, Igalia, Microsoft and Mozilla — the incumbents of browser-wars-2026 plus two consultancies, at one table.
Its stated aim is to “increase the interoperability of key features of the web platform across browsers,” and its scope rule is narrow on purpose: only technology already specified in web standards. Interop does not decide what the web should have. It decides what the browsers should stop disagreeing about.
The twenty focus areas for 2026
Container style queries · CSS anchor positioning · CSS attr() · CSS contrast-color() · CSS zoom ·
custom highlights · dialogs and popovers · fetch uploads and ranges · IndexedDB · JSPI for Wasm ·
media pseudo-classes · the Navigation API · scoped custom element registries · scroll-driven
animations · scroll snap · CSS shape() · view transitions · web compat · WebRTC · WebTransport.
Four investigation efforts sit beside them, for areas where the tests do not yet exist to score: accessibility testing, JPEG XL, mobile testing and WebVTT.
The scoring rule is the interesting part
The dashboard shows per-area pass percentages and “an overall total score of tests that pass in every browser.” That last clause is the design decision. A browser’s own pass rate is a number it can improve alone; the interop score only moves when the laggard catches up, so no vendor can post a win by shipping something the others do not have.
To be scorable at all, a proposed feature must already be covered by web-platform-tests, the coverage must be enough to “represent a meaningful assessment of implementation quality and interoperability,” and the tests must run fully automated in Chrome, Edge, Firefox and Safari desktop runs on wpt.fyi. That gate quietly excludes a lot: anything mobile-only, anything needing a human, anything without tests. Three of the four investigation efforts are there because of it.
What it settles for this wiki
kitesurf announced “215,000+ Web Platform Tests” passed, and this spoke recorded that a pass count without a denominator is not a conformance rate. Interop is the denominator’s home: wpt-overview-bocoup puts the suite at 56,552 tests / 1.8 million subtests as of late 2024, so a raw six-figure pass count is a fraction of the suite quoted without the fraction.
More usefully, it establishes what a credible conformance claim looks like in this domain, which matters for the from-scratch engines the spoke is tracking. ladybird and Falco are both writing engines against a platform whose incumbents needed a formal, funded, six-organization project to agree with each other. The twenty areas above are not obscure corners — dialogs, IndexedDB, scroll snap, WebRTC — and they are on the list because four mature engines still behave differently. That is the size of the target a new engine is aiming at, stated by the people who own the existing ones.
The 2026 list is also a decent map of where the platform is moving: five of the twenty are CSS layout and styling primitives that used to require JavaScript, and the Navigation API, view transitions and scroll-driven animations are the same trade — behaviour the browser does natively that a framework used to ship. None of it is AI. The spoke’s organizing axis is assistant-versus- agent, and the vendors’ own shared engineering agenda for the year touches neither pole.