Web app manifests on 40 major sites: 10 installable, 19 absent, and 7 that only refused the probe

First-hand measurement of <link rel="manifest"> adoption and manifest validity across 40 major sites, 2026-08-24. The manifest URL is read from the served HTML and resolved after redirects, never guessed: probing /manifest.json would miss Spotify, which serves it from a CDN under a hashed filename.

Result: 10 installable, 19 with no manifest link, 3 incomplete (airbnb missing display, figma missing start_url, linkedin missing all three of start_url, display and any icon of 192 px or more), 1 unparsable, and 7 that simply refused the probe.

The seven refusals are recorded separately from the nineteen absences, because they are different facts and only one of them is about the site. The first pass of this measurement did not follow homepage redirects and reported nine inaccessible domains; following redirects turned two of them into real results (uber.com genuinely has no manifest, flipkart.com is fully installable). Two of nine were an artefact of the probe. Any survey reporting a single "not supported" bucket mixes the measured with the unmeasurable.

Method, criteria and limits are documented in the README. Measured for HTML5 Advent.