How the tests work
Every test runs inside your browser. This page explains where each number comes from and how much weight to give it before a call.
Camera
Frame timing comes from capture timestamps of each frame where the browser offers them (MediaStreamTrackProcessor). Elsewhere only displayed frames are visible, so the frame rate is shown as a lower bound.
Real resolution is found by asking the camera for each common size and comparing what arrives. Picture readings (brightness, sharpness, noise, colour cast) are relative to the scene.
A browser receives finished frames, not hardware details, so the camera's sensor model and size and the delay the camera itself adds never reach the page. The tests judge the camera by the picture it actually delivers.
Microphone
Levels are computed from every sample in an audio worklet, in dBFS: 0 is the digital ceiling and a full-scale sine reads −3 dBFS RMS.
Background noise is A-weighted from the spectrum of an 8-second quiet recording. Mains hum is detected as peaks at 50 or 60 Hz and their harmonics.
Because microphones are not calibrated, the levels describe the recording, not the loudness of the room in dB SPL. The live spectrum shows what is in the sound; a microphone's frequency response would need a reference source.
Speakers and headphones
Test sounds are synthesised in the browser and capped at −12 dBFS with a limiter. Left and right use hard channels, not panning.
The sweep and the left/right cues rely on your ears and your answers, so they show what you hear on your setup rather than a speaker's frequency-response curve. Bluetooth delay is lined up by ear with a flash and a click, to about ±20 ms.