What we measure. What we don’t.
A browser observes the connection between hardware, drivers, the operating system and a web page. It cannot isolate every layer of that connection.
How results are assigned
- PASS: the stated observation was obtained, or the user confirmed expected behavior. It applies only to the reported check.
- WARNING: something warrants another check, including access failures and uncertain observations.
- FAIL: reserved for a clearly documented failed check. Current automated checks use WARNING when the cause is ambiguous.
- NOT TESTED: no observation recorded, or the user skipped the test.
- UNSUPPORTED: the needed capability is not exposed in this context.
- USER REVIEW REQUIRED: the data needs listening, visual inspection or physical input to interpret.
Microphone
Web Audio time-domain samples provide RMS, converted to dBFS using 20 × log10(RMS). A window above −60 dBFS indicates a signal; peak amplitude at or above 0.99 flags possible clipping. These are disclosed heuristics, not calibrated sound pressure or audio-quality standards. Gain control, echo cancellation and noise suppression may change the signal. A quiet interval is not a reliable room-noise measurement.
Camera and sound
Camera resolution comes from MediaStreamTrack settings. Delivered FPS uses requestVideoFrameCallback where exposed. Neither establishes maximum sensor performance. Web Audio generates short output tones with a gain ramp and optional stereo panning. Only a listener can confirm physical sound or distortion. This is not a hearing test.
Keyboard and mouse
Keyboard events show received codes and simultaneous presses while the surface has focus. Reserved shortcuts and lost focus limit coverage. Close mouse click pairs use a user-selected interval and do not prove switch bounce. Event rate counts delivered pointermove events over a rolling five-second window; it is not USB polling frequency.
Display
Patterns support visual inspection; the browser cannot see defects or calibrate color. Refresh estimates use 1000 divided by the median requestAnimationFrame interval over eight visible seconds. Standard deviation divided by mean below 10% indicates a relatively consistent sample, not a statistical confidence interval. Background interruptions invalidate sampling.
Controllers and touch
Gamepad stick magnitude is sqrt(x² + y²). A five-second resting sample requires released sticks; the chosen dead zone is a comparison threshold. Drivers may filter readings. Touch coverage records visited grid cells and maximum simultaneous touches, then asks you to judge continuity.
System information
We read basic browser hints, CSS dimensions and pixel ratio without hashing them or collecting a graphics fingerprint. Battery charge is not battery health. Network values are estimates, not a bandwidth test.
Browser support
The implementation targets current Chrome, Edge, Firefox and Safari, plus Chrome Android and Safari iOS. Feature detection controls availability. Battery, network, haptics, fullscreen and recording vary. Embedded browsers may restrict capture. See the repository QA report for versions actually tested.