Translate this page

Pages are written in English and translated in your browser by machine translation. Test results and your data are never sent anywhere.

How we measure

What each test calculates, how accurate it can be in a browser, and why every threshold is set where it is.

Last updated

7 chapters · 5 min read

Principles

1 min
  • Measure on your device. All processing happens in your browser. Nothing you test is uploaded.
  • Use event timestamps. Times come from the moment the browser received the input, not from when the page handled it.
  • No animation in measurement paths. Cues switch instantly, smooth scrolling is off in test areas, and ads are hidden while a test runs.
  • Say what we cannot measure. Each test notes its limits, and thresholds are reasoned defaults or sourced, never invented.

Browser limits that affect every test

1 min
  • Timer precision. Browsers may round high-resolution timers, often to around 0.1 ms or coarser, to protect privacy. Some browsers add jitter.
  • Display refresh. Anything visual can only change once per frame: every 16.7 ms at 60 Hz, 6.9 ms at 144 Hz.
  • Input devices and drivers. USB polling, wireless links and operating system processing add delay before the browser sees an event.
  • Background tabs. Browsers slow down hidden tabs, so tests pause or end when you switch away.
  • Permissions. Camera and microphone are only available on secure pages after you allow them, and are released when you stop.

How each measurement works

1 min
Keyboard keys
Read from KeyboardEvent.code, the physical key position, so any language layout works. Auto-repeat is ignored.
Key chatter
A second press of the same key within the chatter threshold after release, or two presses without a release.
Rollover
The largest number of keys held at once, as reported by the browser.
CPS and spacebar
Presses counted from the first press for the chosen time, divided by the seconds. Runs with almost perfectly even intervals are flagged as automated.
Reaction time
The cue is switched inside a display frame callback and timed from that frame; the response uses the input event's timestamp. Responses faster than the floor below are voided. Audio cues are scheduled on the audio clock with reported output latency added.
Mic level
Peak and RMS of the signal from an AnalyserNode, converted to dBFS (0 dBFS is full scale). The verdict uses the loudest moment in the last four seconds.
Background noise
Median RMS level over 1.5 seconds while you stay silent.
Webcam frame rate
Frames actually presented to the page, counted with requestVideoFrameCallback where supported.
Lighting
Average luminance of a downscaled frame using Rec. 709 weights, on a 0 to 255 scale.
Refresh rate
Median interval between display frames over up to 300 frames, snapped to a common rate when within 3%.
Polling rate
Median gap between mouse reports from raw pointer updates or coalesced events during continuous movement.
Mouse DPI
Raw movement counts under pointer lock (unadjusted where supported) divided by the distance you moved in inches.
Stick drift
Largest distance from center of each stick over three seconds with the sticks untouched.
Typing WPM
Correct characters divided by five, per minute. Accuracy counts every keystroke, including later corrections.
Hearing screening
Ascending 5 dB steps per ear and frequency, relative to your own calibration tone. Not calibrated dB HL and not a diagnosis.

Every threshold, with its basis

2 min

These values live in one configuration file that the tools read directly, so this page and the tests always agree.

CheckValueBasis
Key chatter
Keyboard Tester
Same key pressed again within 30 ms Mechanical switches settle (debounce) in about 5 ms by maker specification, and keyboards filter that internally. A second press of the same key less than 30 ms later is faster than people can repeat one key, so it points to a switch that bounces through the filter. You can change the threshold in the tester. Source: Cherry MX switch data: bounce time under 5 ms.
Unwanted double click
Mouse Test
Two clicks within 80 ms A deliberate double click takes well over 80 ms between clicks for nearly everyone. Clicks closer than that come from a worn micro switch bouncing. This is a reasoned default, not a published standard; the double click test lets you change it.
Microphone level
Mic Test
Too quiet below -40 dBFS peak, good from -24 dBFS, too loud above -3 dBFS Digital audio clips at 0 dBFS. Speech that peaks between about -24 and -6 dBFS leaves headroom and sits well above typical room noise. These bands are a practical guide for calls and voice chat, not a broadcast standard. Source: AES: dBFS definition (AES17).
Background noise
Mic Test
Quiet below -60 dBFS RMS, moderate to -45 dBFS, loud above Measured as the average (RMS) level over one second while you stay silent. The bands are a reasoned default for voice chat; browser noise suppression can lower the reading.
Lighting hint
Webcam Test
Too dark below 60, overexposed above 200 average luminance (0–255) Average luminance of the frame using Rec. 709 weights. Values near the middle of the range keep face detail. A reasoned default, shown as a hint only. Source: ITU-R BT.709 luma coefficients.
Stick drift
Controller Tester
Warning above 0.08 idle magnitude, likely drift above 0.24 Measured with the stick untouched. Microsoft documents a left-stick dead zone of 7849 out of 32767 (about 0.24) for XInput controllers, so idle readings above that would leak into most games. Readings above 0.08 are worth watching. Source: Microsoft: Getting started with XInput (dead zones).
Frame pacing
Refresh Rate Test
Unstable above 1.5 ms standard deviation At 60 Hz a frame lasts 16.7 ms; a spread above 1.5 ms means frames arrive unevenly enough to be visible as stutter. Background tabs, power saving and browser timer limits all affect it.
Automation flag
CPS Test
Intervals with under 2 ms spread Human clicking varies by several milliseconds from click to click. A run that is almost perfectly regular is flagged as likely automated and is not saved as a personal best.
Too-early response
Reaction Time Test
Under 100 ms after the cue Responses under about 100 ms are generally treated as anticipations in reaction-time research, because the nervous system cannot see, decide and move that fast. Source: Luce, R. D. (1986). Response Times. Oxford University Press.
Flash rate
Stuck Pixel Fixer
Default 2 per second, never above 10 WCAG treats more than three flashes per second as a seizure risk. The fixer starts slow, sits behind a warning and an explicit opt-in, and stops on any key or tap. Source: W3C WCAG 2.2: Three flashes or below threshold.

Fun labels are labels, not statistics

1 min

Rank names such as "Rabbit" or "Cheetah" in the CPS test are the site's own labels for score bands. They do not describe how you compare with other people. We do not publish averages from visitors because we do not collect visitors' results.

Accuracy checks

1 min

Each tool page lists the browsers and devices it was tested on. When we validate a measurement against known references, such as a display with a known refresh rate or a keyboard with known rollover, we record the result here and on the tool's changelog.

Found a measurement problem?

1 min

Please tell us through the contact page, with your browser, operating system and device. We fix and document confirmed issues.