Font fingerprint test

For each of 107 font names, the test measures a line of text at 72 px set in that font with each of three fallbacks (monospace, sans-serif and serif) and compares it with the fallback alone. A different width under any of the three means the system supplied a real font for the name. Three fallbacks are needed because a font can happen to match one: on Windows, Arial is the default sans-serif, so it only shows up against the other two.

The list mixes the default fonts of Windows, macOS, Linux and Android with fonts common across all of them. Each operating-system group votes for its system; the result names a system only when one clearly leads, and cross-platform fonts are listed without voting.

Installed fonts do not change between two measurements, so the canvas pass runs twice and should find the same list. It then asks the question again through page layout, with hidden <span> elements, and the two methods should agree: they reach the same font system. A made-up font name is measured as a control and must come back absent; if it is "detected", the widths are not real.

Last, the list is compared with the operating system the browser claims. The fonts come from the machine the browser actually runs on, so a Windows user agent with macOS or Linux fonts is a contradiction. The opposite mistake is harder to catch: Linux can install look-alike fonts with the same widths as Windows ones (Carlito for Calibri, for example), which measurement cannot tell apart. The browser fingerprint test has further checks for that, and the research note on the font stack under the user agent explains them.

Fonts are also what the canvas fingerprint draws its text with, which is why the two often change together. WebGL, WebGPU and audio cover the hardware. For background, see what font fingerprinting is.