About · Reviewed August 18, 2026

About Device Bench

Device Bench is a collection of small hardware instruments that run in a web browser. It exists because the first useful question about a troublesome key, microphone, camera, speaker, controller, or screen is often simple: what does this device report right now? Answering that question should not require installing an application, creating an account, or sending a recording to a server.

The product is the evidence

A test page is useful only when it distinguishes an observation from a conclusion. Device Bench therefore publishes the evidence behind each summary. The keyboard tester shows individual events before it reports rollover. The microphone test shows its level windows and distinguishes digital full scale from physical loudness. The gamepad tester exposes raw axes before it describes drift. The video-call test labels rows as reported by the browser, measured by the page, or confirmed by the person using it.

When the evidence is not strong enough, the tool should refuse the result. Silence is not a pitch. An unavailable browser capability is not a pass. A camera preview cannot tell us what a remote caller sees, and a web page cannot see the light leaving a display. The limits section on each tool is part of the instrument, not legal furniture added after it.

How tools are checked

Every new instrument is exercised on its central success path and on at least one failure or refusal path. Deterministic inputs are used where the browser permits them: known event sequences, known audio samples, known elapsed times, and known geometry. Structural checks keep the numbers in explanatory text tied to the thresholds and formulas the engine actually uses. Browser checks cover security policy and visual behavior that a simulated document cannot prove.

The full process—including the difference between reported, measured, and confirmed values—is published on How Device Bench tests its browser instruments. That page also describes the privacy boundary, browser limitations, source policy, and correction process.

Three commitments

Local execution. Tests use capabilities already present in the browser: keyboard and pointer events, the Gamepad API, Web Audio, camera and microphone tracks, animation timing, and fullscreen rendering. Tool inputs and results stay in the page. Cloudflare serves the site files and necessarily receives ordinary page requests, but there is no upload endpoint, analytics collector, account system, or database behind a test.

Published methodology. Every tool explains what is measured, how the result is derived, and where the method stops being reliable. Guides show a Device Bench byline, publication and review dates, and the standards or platform documentation supporting their factual claims.

No accounts or dark patterns. There are no sign-ups, email capture forms, artificial run limits, or locked results. Advertising is not active. If advertising is introduced after approval, privacy, consent, security policy, and placement review are separate changes rather than a script switched on silently.

Corrections

Device Bench is responsible for the claims it publishes. A useful correction report names the page, browser and operating system, the value shown, what was expected, and the shortest steps that reproduce the disagreement. Do not send recordings, screenshots containing personal information, or other private test data. Report a correction or measurement problem.

General questions and requests can also go through the contact page. Requests are considered against the same rule as existing tools: the browser must be able to observe the job honestly, and the result must add something more useful than a keyword variation.