---
title: "How We Test – Methods, Limits and Sources for Each Test"
description: "How each DeviceTests test measures your screen, keyboard, mouse, sound, camera and internet, where our limits come from, and what a browser can’t measure."
url: https://devicetests.com/how-we-test
language: en
---

# How we test

How our tests work, where our limits come from, and what a web browser can and can’t measure on your device.

Last updated November 17, 2026

Every test on DeviceTests runs in your web browser, on your own device. This page explains how the tests get their results, which limits they use to judge them, and where those limits come from. Each test page also has a section called “How this test works” with the full details for that test. DeviceTests is run by a small team at DeviceTests LLC: see [About us](https://devicetests.com/about-us).

## The short version

- **The test only sees what your browser sees.** Your device tells the browser what happened, such as “the A key was pressed”. The test can’t look inside the hardware.
- **Timing numbers are estimates.** Browsers round their clocks, and your computer can delay events. We say so on every test that shows a time.
- **You only get a result for what you tested.** Parts you didn’t try are shown as “not tested”, never as working.
- **Some things need your eyes or ears.** A browser can’t see a dead pixel or hear your speakers. In those tests, you tell the test what you saw or heard.
- **We say where our numbers come from.** A limit comes from a standard, a maker, a study, or our own choice. Below, we say which.

## What each kind of test can and can’t tell you

| Device | What the test can tell you | What it can’t tell you |
| - | - | - |
| Screen | Dots that don’t change colour, when you look for them. How often the browser updates the picture. | Brightness or colour accuracy. A refresh rate the browser doesn’t use, for example in battery saver mode. |
| Keyboard | Which keys reach your computer, by their place on the keyboard. Left and right keys apart. Keys that stay pressed or type twice. | The Fn key and keys your computer keeps for itself. How fast the keyboard is inside. |
| Mouse and touchpad | Which buttons and scroll directions work. Clicks that come in twice. How often the mouse reports its position. | The exact sensor setting (DPI). Polling rates above about 1,000 times per second are rough. |
| Game controllers | Which buttons work, how far triggers go, and where each stick rests. | Touchpads, motion sensors and lights on most controllers. Vibration only works in some browsers. |
| Speakers and headphones | Whether the left and right side play, as you hear it. | Sound quality as a number. How loud the sound is in your ears. |
| Microphone | Whether your microphone picks up sound, how loud it is compared with silence, and what it sounds like when played back. | Exact loudness of your room in decibels (dB). Which sounds your computer already cleaned up, such as background noise. |
| Camera | Whether pictures arrive, and the size and frame rate the browser gets. | How sharp the picture is. Phones and some browsers give a web page a smaller picture. |
| Internet | Your speed, ping and lost data between this device and our test server, at this moment. | The speed to every other website. Your connection at other times of day. |
| Phone sensors and your computer | Touches, position accuracy, motion sensors and what your browser reports about your device. | Satellites, the sensors’ factory accuracy, or hardware the browser keeps hidden. |
| Files | Converting, shrinking and reading photo details, on your device. | Formats your browser and our tools can’t open. |

## How we decide what a result says

Most tests end with a result in words. Behind each result is a set of limits, for example “a stick that rests more than 15% from the centre is likely drifting”.

- **Where a limit comes from:** a published standard (such as WCAG for flashing, or ITU recommendations for calls and video), a maker’s documentation, a research study, a widely used convention from other test sites, or our own choice. When there is no outside source, we say the limit is our own.
- **We judge carefully.** When the evidence is weak, the result says “likely” or “probably”, or gives a warning instead of a fail.
- **We keep four kinds of answer apart:** measured by the test, seen or heard by you, not tested, and not possible in this browser.
- **We never call a device healthy** after a partial check. A test can show that something works. It can’t promise that nothing is wrong.
- **The result comes first, in words.** For example: “2 keys didn’t respond: F7 and Q”. Colour is never the only sign. Then we say what to do next, such as trying another cable, and when a repair shop or the maker can help more.

## Why timing numbers are estimates

A browser gives each event a time stamp. To protect your privacy, browsers round these times. A millisecond (ms) is a thousandth of a second.

| Browser | How finely it rounds times |
| - | - |
| Chrome and Edge on Windows and Mac | About 0.1 ms |
| Chrome on Linux | About 0.1 ms for screen timing, about 1 ms for key and mouse events |
| Firefox and Safari | About 1 ms |
| Privacy settings, such as Firefox’s fingerprinting protection or Tor Browser | 17 ms or more |

A finely rounded clock doesn’t make the whole measurement that exact. Your computer can also hold events back for a moment when it is busy. Browsers may also join several quick mouse movements into one update. So we:

- use the time stamp of each event, not the time the page got round to it
- measure how exact your browser’s clock is, and say so when it is too rough for a result
- show typical values, such as the median (the middle value), rather than a single best or worst one
- round numbers, so they are never more exact than the measurement

## Screens

A web page can’t see your screen. So most screen tests show an exact pattern, and you judge what you see. The tests that measure something, such as the refresh rate, measure the pictures your browser draws.

| Test | How it works | How we judge it |
| - | - | - |
| [Dead pixel test](https://devicetests.com/dead-pixel-test), [backlight bleed test](https://devicetests.com/backlight-bleed-test), [burn-in test](https://devicetests.com/burn-in-test) | Full-screen fields of exact colours and greys. A field counts as seen after 1 second in full view. | Your answer. A dead dot is a fail. Bleed and burn-in get a warning at most: makers judge bleed in normal room light, and none publish a limit (Dell, ASUS). |
| [New monitor test](https://devicetests.com/new-monitor-test) | The dead pixel, bleed, evenness, refresh rate and resolution checks, in five steps. | Each step uses the same rules as its own test. A skipped step means “not fully tested”. |
| [Refresh rate test](https://devicetests.com/refresh-rate-test) | Times every picture your browser draws for 3 seconds. Uses the middle gap, then the average of the gaps close to it. | Named as a known rate, such as 60, 120 or 144 Hz (pictures a second), when within 2.5%. A gap of 2 or 3 normal gaps is a skipped picture. More than 10% skipped is a warning. Our own limits. |
| [FPS test](https://devicetests.com/fps-test), [ghosting test](https://devicetests.com/ghosting-test) | Moving pictures at fixed speeds, from the same clock. The ghosting test’s usual speed is 960 pixels a second. | You judge the trails. We give no response time in ms: that needs a light sensor, as lab reviewers such as RTINGS use. |
| [Frame skipping test](https://devicetests.com/frame-skipping-test) | One box lights per picture. You take a camera photo with a long exposure, 1/10 second or longer. | Your photo: an unbroken row passes, gaps get a warning. A screenshot can’t show it. |
| [Monitor gamma test](https://devicetests.com/monitor-gamma-test), [color banding test](https://devicetests.com/color-banding-test), [chroma subsampling test](https://devicetests.com/chroma-subsampling-test) | Patterns drawn pixel by pixel, not with browser fades, because browsers smooth those. If zoom changes the pixels, the answer counts as not tested. | Your answer. Gamma 2.0 to 2.4 passes. That is our own range around the two usual targets: about 2.2 for computers (sRGB) and 2.4 for TVs (BT.1886). These tests never fail a screen. |
| [HDR test](https://devicetests.com/hdr-test) | Reads what your browser reports about HDR, and plays our own HDR clips with known brightness, made to the HDR standard (SMPTE ST 2084). | Brightness in nits (a measure of light) is what the clip asks for, not what your screen makes. Your eyes judge the rest. |
| [Screen resolution](https://devicetests.com/screen-resolution) | Reads the size and pixel ratio your browser reports, and works out the zoom. | Labelled by how sure it is: measured, exact or close. |

**Flashing.** No test flashes a large part of the screen more than 3 times a second, the WCAG 2.3.1 limit. Moving patterns, such as the ghosting test, are motion, not flashes. The [stuck pixel fixer](https://devicetests.com/stuck-pixel-fixer) flashes faster only in a small square, inside WCAG’s allowance for a small area, and only after you start it.

**What screen tests can’t measure:** brightness, colour accuracy, response time, input lag, bit depth, or the highest refresh rate your screen supports. Those need lab tools. Safari and iPhones may also hold web pages at 60 pictures a second.

## Keyboards, mice and controllers

These tests read the events your browser gets from the keyboard, mouse, pen or controller. Your system, drivers and apps can change those events first, so a test can’t see a key that never reaches the browser.

### Keyboards

| Test | How it works | How we judge it |
| - | - | - |
| [Keyboard tester](https://devicetests.com/keyboard-tester) | Reads each key by its place on the keyboard, the same on every layout. Keys held down to repeat are counted once. | A key you didn’t press is “not tested”, never “broken”. A key held 2 seconds and still repeating, or held 5 seconds, may be stuck. Our own limits. |
| [Keyboard ghosting test](https://devicetests.com/keyboard-ghosting-test) | Groups of keys that games use together. Records the most keys held at once. | A group that stops at exactly 6 keys points to a 6-key limit: a basic USB keyboard report has room for 6 keys, plus Shift, Ctrl and Alt. A missing key is a warning, never a fail. |
| [Keyboard chatter test](https://devicetests.com/keyboard-chatter-test) | Times the gap between letting go of a key and its next press. | A gap under 25 ms is a double. A fail needs 2 doubles, one of them very short: a gap under 15 ms, or a press shorter than 10 ms. These limits are our own. Switch makers (Cherry, Gateron) say a switch bounces for under 5 ms. In a study of 136 million key presses, even the fastest typists left about 60 ms between keys (Dhakal and others, 2018). |
| [Keyboard polling rate test](https://devicetests.com/keyboard-polling-rate-test) | Finds the beat in your key times: steps of 8 ms mean 125 Hz, steps of 1 ms mean 1,000 Hz. | Up to 2,000 Hz where the browser’s key times are fine enough (Chrome and Edge on Windows and Mac), and up to 500 Hz elsewhere. It sees the rate that reaches the browser. Below 100 Hz is a warning, our own limit. |
| [Rapid trigger test](https://devicetests.com/rapid-trigger-test), [keycode finder](https://devicetests.com/keycode) | Times how fast a key resets, and what Snap Tap does when two keys overlap. The keycode finder shows the raw values. | Snap Tap (one key taking over from another) results are worded “probably”, because a keyboard setting and an app look the same to a browser. The keycode finder judges nothing. |

### Mice, touchpads and pens

| Test | How it works | How we judge it |
| - | - | - |
| [Mouse test](https://devicetests.com/mouse-test), [double-click test](https://devicetests.com/double-click-test) | Times every press and release of each button. | An extra click is a second press of the same button within 80 ms, or a press again within 20 ms of letting go. These are our screening limits: the fastest single-finger taps in a study took about 110 ms (Aoki and others), and other testers use 80 ms too. Very fast clicking on purpose can still trip them. In the double-click test, 2 or more extra clicks are a fail, 1 is a warning. |
| [Mouse polling rate test](https://devicetests.com/mouse-rate-test) | Counts the position reports your browser gets in each tenth of a second while you move. The rate is the middle value of those counts. | Named as a standard rate (125, 500, 1,000 Hz…) when within 12%. Reliable up to about 1,000 Hz. When the browser ties reports to the screen, we say so instead of showing a number. |
| [Scroll wheel test](https://devicetests.com/scroll-wheel-test) | Learns how far one click of your wheel goes, then follows every click. | A jump is one or two clicks the wrong way, with at least 2 clicks the right way before it and 1 after, all within 1.5 seconds. 2 jumps in one direction are a fail. A direction counts as checked after 10 clicks. Our own rule. Smooth wheels and touchpads aren’t judged. |
| [Trackpad test](https://devicetests.com/touchpad-test), [pen pressure test](https://devicetests.com/pen-pressure-test) | Guided steps for moving, clicking, scrolling and zooming. The pen test reads pressure, tilt and buttons. | Pen pressure works when we see at least 16 different levels. One fixed value, usually 0.5, is what a browser reports for a pen without pressure. |

### Game controllers

The [controller tester](https://devicetests.com/controller-tester) and the [stick drift test](https://devicetests.com/stick-drift-test) read the controller once for every picture your screen shows, often 60 times a second. They use the stick values exactly as the browser gives them, with no dead zone of our own. For drift, you leave the sticks alone for 5 seconds. The offset is how far the stick rests from the centre. We leave out the 5% of readings furthest out, so a single bump doesn’t count. A dead zone is a small area around the centre that a game ignores.

| How far the stick rests from the centre | What it means |
| - | - |
| 5% or less | Normal. New controllers rest up to about 5% off centre (Gamepadla measurements). |
| More than 5%, but inside the dead zone (10% unless you change it) | Most games ignore this much. |
| Up to 15%, outside your dead zone | Likely mild drift. |
| More than 15% | Likely moderate drift, past most games’ default dead zones (10–15%). |
| More than 25% | Likely severe drift, past the dead zones Microsoft suggests for Xbox controllers (about 24% and 27%). |

The 15% and 25% lines are our judgement, based on game and Microsoft dead zones. They are not failure limits from a maker: there is no published failure data. So the result says “likely”. Triggers pass when they reach 90% of their range.

### Click, spacebar and reaction games

[CPS test](https://devicetests.com/cps-test) and [spacebar counter](https://devicetests.com/spacebar-counter) scores are presses divided by the seconds, the usual way. A round starts at your first press. A press counts if it happened before the end, even when it arrives up to 50 ms late. Rounds with presses under 15 ms apart, or with gaps too even for a hand, are marked unusual, and don’t count as a personal best.

The [reaction time test](https://devicetests.com/reaction-time-test) uses the middle value of 5 tries. A press under 100 ms after the signal is a false start, as in athletics starts. The comparison by age comes from a study of 1,469 adults (Woods and others, 2015). Your device’s own delay is part of every time, so we say how big it can be instead of hiding it.

## Sound, microphones and cameras

Your browser makes every test sound itself, so there is no sound file to download. Camera pictures and microphone sound stay in your browser tab.

**dBFS is not room loudness.** dBFS means decibels below the loudest level a recording can hold. Your microphone volume setting changes it. A web page can’t measure the room in real decibels (dB SPL), because it doesn’t know your microphone. Most browsers also clean up the sound, as in calls, so levels are estimates.

| Test | How it works | How we judge it |
| - | - | - |
| [Sound test](https://devicetests.com/sound-test), [headphone test](https://devicetests.com/headphone-test) | Each sound goes only to the side it names. An automatic test records the sound inside the browser to check this. At the default test volume, the chime plays at about −18 dBFS. Your device’s own volume comes on top. | A browser can’t hear your speakers, so the result comes from your answers. |
| [Audio latency test](https://devicetests.com/audio-latency-test) | You tap along to clicks, then to flashes. Or your microphone hears short chirps. We use the middle value of at least 8 taps, or 4 chirps. | Tapping: sound is called late from 80 ms, and very late from 185 ms. The 185 ms line is from ITU-R BT.1359, on when sound out of step with the picture becomes unacceptable. The 80 ms line is our own, stricter than the 125 ms at which most people notice. The microphone method allows 120 and 250 ms, because the sound travels out and back. |
| [Tone generator](https://devicetests.com/tone-generator) | You set the pitch in steps of a hundredth of a hertz. | Not calibrated, and not a hearing test: your device sets how loud it is. |
| [Microphone test](https://devicetests.com/mic-test) | Measures the sound level many times a second, with the browser’s usual clean-up on, as in calls. | Works when a sound is at least 10 dB louder than the background, and louder than −50 dBFS, for at least 0.4 seconds. Our own limits. |
| [Microphone quality test](https://devicetests.com/mic-quality-test) | Asks the browser to turn its clean-up off, and shows whether it did. Then measures the background for 3 seconds and your voice while you read. | Background below −60 dBFS counts as low, like the −60 dB limit of the ACX audiobook rules. Voice 40 dB or more above the background is excellent, 30 good, 20 fair. Clipping means 100 ms or more at the recording’s limit. These grades are our own. |
| [Webcam test](https://devicetests.com/webcam-test), [camera resolution test](https://devicetests.com/camera-resolution-test) | Counts the pictures that arrive, and checks how bright they are. The resolution test asks for each common size in turn. | A picture that stays almost black for 2 seconds is a black picture. In Chrome and Edge a size counts as the camera’s own only when the browser didn’t resize it. |
| [Video call test](https://devicetests.com/video-call-test), [voice recorder](https://devicetests.com/voice-recorder) | The camera, microphone, speaker and internet checks together. The recorder saves the sound as it came in. | Each part uses the rules of its own test, with a short internet check. |

## Internet

Internet tests must send data to our test server and back. The result names the city of the server. They measure the path from your device to that server, at that moment.

| Test | How it works | How we judge it |
| - | - | - |
| [Internet speed test](https://devicetests.com/internet-speed-test) | Downloads random data over 4 connections for 10 seconds, then uploads for 10 seconds. The first 1.5 seconds down and 2 seconds up are left out while the line speeds up. Random data can’t be squeezed smaller. | Your average speed, not your fastest moment. Compared with what common uses need, based on advice from Netflix, YouTube, Zoom and Microsoft, with room to spare: 5 Mbps (megabits a second) for HD video, 25 Mbps for 4K, 3 Mbps each way for an HD call. |
| [Ping test](https://devicetests.com/ping-test) | Sends a tiny message twice a second over a WebSocket, a web connection that stays open, and times each answer. The first one is left out. | Ping up to 30 ms is great, 60 good, 100 fair, 200 poor. Jitter (how much ping changes) up to 5 ms is great, 15 good, 30 fair, 60 poor. The rating is the worst of these. Our own bands. Microsoft’s targets for Teams calls (ping under 100 ms, jitter under 30 ms) are a rough guide. Our server and method differ, so this is not a Teams check. |
| [Packet loss test](https://devicetests.com/packet-loss-test) | Sends numbered packets over a WebRTC data channel, the kind of connection calls and games use. Lost packets are never sent again, so they stay lost. | Loss up to 0.5% is good, 2% fair, 5% poor. Microsoft asks for under 1% for Teams. For calls, a packet back after 300 ms counts as late: our choice, based on ITU-T G.114’s advice to keep the delay each way under 150 ms. At least 100 packets are needed for a rating. |
| [Bufferbloat test](https://devicetests.com/bufferbloat-test) | Measures your ping at rest, then while downloading and uploading. | The grade uses how much the ping rises, in the worse direction, on Waveform’s scale, which grew from the DSLReports one: A+ under 5 ms, A under 30, B under 60, C under 200, D under 400, F at 400 ms or more. |

**What internet tests can’t measure:** your speed or ping to other servers, such as a game’s; where on the way data gets lost; and your connection at other times. Browsers can’t send the “ping” that the command on your computer sends (ICMP), so our ping can be a little higher.

## Phones and computers

| Test | How it works | How we judge it |
| - | - | - |
| [Touch screen test](https://devicetests.com/touch-screen-test) | Paints every square about 48 pixels wide that your finger crosses, even between two reports. | Every touch your browser reports. The browser or system may limit how many fingers a page sees: iPhones, iPads and many Android phones say 5, whatever the screen can do. |
| [GPS test](https://devicetests.com/gps-test) | Asks your browser for a fresh position at the best accuracy it offers, then watches it for 1 minute while you stand still. | Uses the accuracy your device reports, not a check of where you really are. On a phone, within 30 m passes, within 150 m is a warning, more fails. Computers and iPads are judged by what Wi-Fi can do, because most have no GPS. Our own bands. |
| [Gyroscope test](https://devicetests.com/gyroscope-test) | Reads the motion sensors about 60 times a second. | A sensor works when its reading changes as you move: a turn faster than 30° a second, or a small push (a change of 2 m/s²). Our own limits. |
| [GPU test](https://devicetests.com/gpu-test), [My device](https://devicetests.com/my-device) | Reads what your browser reports about the graphics chip, the system and the screen. The GPU load test draws the same small picture on every device. | The GPU score is the most work done in a second. It is not a game benchmark. |
| [Printer test page](https://devicetests.com/printer-test-page) | Exact colour blocks, greys, fine lines and rulers of 100 mm and 4 inches. | You judge the print. Only your printer’s own nozzle check shows each nozzle. |

## File tools

The [HEIC to JPG converter](https://devicetests.com/heic-to-jpg), [image compressor](https://devicetests.com/image-compressor) and [EXIF viewer](https://devicetests.com/exif-viewer) work inside your browser. Your files aren’t uploaded. They use free, open-source programs that we run unchanged: libheif for HEIC photos, MozJPEG for JPG, libwebp for WebP and UPNG for PNG.

- Photos are turned the right way up. Where the format has a colour profile, it is kept, so colours stay true. Some formats are changed to standard colours (sRGB).
- The image compressor tries a few quality levels and keeps the best one that fits your size.
- When you remove details such as the location, the EXIF viewer checks the cleaned copy twice before you can save it: it reads the copy again, and searches the whole file for known kinds of location data. It can’t remove a place you can see in the picture itself.

## Device finders

Our [device finders](https://devicetests.com/finders) suggest products, such as a mouse or a monitor, after a few short questions. They work differently from our tests:

- **Your answers come first.** Some answers rule products out: a left hand rules out mice made for the right hand, and a PlayStation rules out headsets that can’t connect to one. Other answers move a product up or down, such as its weight, its size or its price.
- **Then reviews and price.** When products fit your answers equally well, the one reviewers rate higher comes first, then the cheaper one.
- **Every reason comes from a rule.** A “Why it fits you” line only appears when the product really meets that answer. Every product also gets a “Keep in mind” line with a real trade-off. No AI writes these lines while you use a finder.
- **We show fewer when few fit.** If only 3 products fit your answers, you see 3. A product that misses one of your answers by a little is marked as a close match, with what it misses.
- **At most 2 from one maker** in a top 5, as long as other products fit.
- **Prices are rounded and dated.** Prices change often, so check the shop before you buy.
- **Nobody pays us to be listed.** No maker or shop has paid to be included.

## Your privacy during a test

- **Camera and microphone:** a test can only use them after you allow it in your browser. The picture and sound stay on your device. A recording you make to play back stays on your device too. We don’t upload any of it.
- **Files:** file tools work inside your browser. Your files aren’t uploaded.
- **Internet tests:** these must send test data to our server and back. That is how they measure your connection.

The [privacy policy](https://devicetests.com/privacy-policy) has the full details.

## Found a wrong result?

Tell us which test, which device and which browser you used. It helps us fix the test for everyone. You can reach us through our [contact page](https://devicetests.com/contact).
