Three kinds of evidence
- Measured: browser-provided samples or events, such as microphone amplitude, video dimensions, touch contacts and animation intervals.
- User observed: what you hear, see or feel. Speaker audibility, display defects and vibration depend on your confirmation.
- Unavailable or incomplete: permission was blocked, an API was missing or the necessary actions were not tested. These are not interchangeable with a hardware failure.
A passed item means the stated check worked in this session. A failed or warning item records an observed problem or mismatch, not its physical cause. The homepage report combines saved outcomes; its grade is a checklist summary, not a laboratory certification. Older saved results may contain fewer measurements than newly completed tests.
Audio capture and playback
The microphone tool reads a 1024-sample time-domain buffer and displays a waveform plus relative RMS and peak amplitude. Values above a quiet threshold are signal evidence, not recognized speech. Near-full-scale samples flag possible clipping. Noise suppression and automatic gain can affect readings.
A short MediaRecorder capture uses the browser-selected MIME type. Input stops when recording finishes. Playback is initiated by the user and clear sound requires a user assessment. The live meter remains useful on browsers that support capture and Web Audio but not recording.
The speaker tool schedules brief panned sine tones. The headphone tool adds stereo, balance and selected frequencies. Neither measures acoustic output. The optional moisture tone schedules a short low-frequency sweep; completion never establishes water removal.
Camera and webcam
The camera diagnostics request a browser-visible device or facing preference. Dimensions and frame-rate settings describe the negotiated stream. Frame callbacks indicate presentation activity, with playback-time observation as a weaker fallback. Snapshots copy the delivered frame locally and release the input.
The webcam utility serves a separate task: mirrored framing, lighting and a downloadable snapshot before a call. Neither tool tests every phone lens, native camera mode or autofocus mechanism.
Touch, keyboard and mouse
The touch grid divides the browser canvas into 96 regions and records positions actually delivered during a slow pass. Uncovered regions need retesting. The multi-touch tool counts concurrent touch pointers and keeps short visible paths. Mouse and pen inputs are excluded from its touch maximum.
The keyboard test records down/up events while its test box is focused and distinguishes held keys from codes already seen. Unobserved codes are not automatically broken. The mouse tool counts button presses, double-click events, wheel directions and movement; these are not sensor DPI or USB polling measurements.
Display appearance and timing
The dead-pixel tool supplies six solid backgrounds for visual inspection, including gray. The screen patterns add gradients and dark/light steps in an embedded area. Neither analyzes physical pixels or calibrates color.
The FPS workload draws fifteen circles for six measured seconds. Mean FPS is interval count divided by total interval duration; minimum and maximum reflect individual intervals. The refresh estimate collects eight seconds after a half-second warm-up and reports 1000 divided by the median interval, plus variation. Both cancel hidden/unfocused runs. Neither reads the panel specification directly.
Other browser capabilities
Orientation requires incoming, changing values before the target exercise can complete. A lack of events is unverified. Vibration requests two pulses and relies on what you feel. Battery status describes exposed charge information, not capacity or health. Device information can be approximate or withheld for privacy.
The network check times five HTTP requests to an external endpoint, with a five-second timeout per request. Mean request duration and adjacent-sample variation include server and browser overhead. Opaque responses cannot confirm HTTP success; this is not bandwidth, ICMP ping or packet-loss measurement.
Repeatability and privacy
Retest under comparable lighting, routing, power and browser conditions. Change one variable at a time. The troubleshooting guides explain practical comparisons and when professional inspection may be needed.
Raw diagnostic media stays in page memory and is not uploaded by the diagnostic code. Saved outcomes and selected metrics persist in browser localStorage until cleared or replaced. Hosting, analytics, and the network test involve separate requests described in the privacy policy. See runtime compatibility before interpreting an unavailable test.