What these tests can tell you

Different states mean different things

Detected

The page received that position or button at least once.

Held down

The last received state indicates a press is still held. Losing context ends that observation.

Not observed

No matching input reached this run. The key may be untested, absent or intercepted.

Excluded by you

You marked a position absent from your hardware. It does not become a tested key.

Interrupted or inconclusive

The context or signals do not allow the check to be interpreted fully. Restart the protocol deliberately.

Keyboard: positions, cycles and combinations

The drawing follows reference physical positions. Choose a layout and its system from the international catalogue. Windows, Mac and ChromeOS references have their own character mappings: choosing Mac also changes the symbols, including the Option layer. Received characters still depend on your active software layout, modifiers and remapping.

Mac references come from Apple’s built-in layouts in macOS 26.6.2, with normal, Shift, Option and Shift+Option symbols. Windows and ChromeOS references use Unicode CLDR 43 keyboard data, with historical snapshots from Windows 10 and ChromeOS 33. These are software references; hardware variants and other system versions may differ. The Apple keyboard identification guide helps compare country and physical variants.

Dead-key composition, Caps Lock and input methods such as Japanese or Chinese IMEs are not simulated. Where a reference uses Latin keys with an IME, its Latin base is shown. The catalogue is included with the page; choosing a reference contacts no external layout service. No typing history is sent.

Free testing shows the positions received. A single-key check separates press, release, hold and new cycles. Native repeat during a hold is distinct from new closely spaced presses. A repetition threshold is exploratory; the browser cannot see electrical contacts or firmware filtering.

Combination testing covers only the groups actually tried. A maximum observed count never certifies universal NKRO. Fn, media keys and reserved shortcuts may not reach the page. Tab and Escape leave the test. Clicking the drawing never validates a physical key.

Mouse: logical buttons, holding and scrolling

Primary and secondary describe logical assignments. Left-handed settings can swap them. Mouse or trackpad is your chosen setup: the browser cannot reliably distinguish every device.

For a guided hold, choose the primary, secondary or middle button. Hold that button for at least two seconds, move inside the area, then release when invited. Another button cannot complete this check. A short hold, no movement, unusable timing and interruption are separate outcomes. Leaving the area or switching applications interrupts the observation; it does not establish contact failure.

The wheel test includes a check of both directions and a five-second observation in one direction. Choose the vertical or horizontal axis and describe your setup: notched wheel, free-spinning wheel, unspecified mouse or trackpad. During the timed observation, keep scrolling in the requested direction; the page reports brief opposite inputs followed by a return, sustained direction changes, unconfirmed changes and pauses. Axes and browser units remain separate. Trackpad inertia and browser units are not universal physical wheel notches.

Method 3 uses explicit reading windows: a brief return is at most two opposite events followed by a return within 350 ms; a sustained change contains at least three opposite events spanning 350 ms. Other changes remain unconfirmed. A gap of at least 800 ms is recorded as a pause. Event timing that cannot be interpreted, delivery delayed by more than 250 ms or a timer gap over 500 ms interrupts the timed check. These thresholds organise the observations; they are not hardware-calibrated fault thresholds and cannot identify whether a change was intentional.

Repeated clicks: read a series in context

Make intentional single presses and release fully between them. The count describes received presses, not the number of gestures you intended. Your threshold identifies close intervals; it is not a universal boundary between a healthy and faulty mouse.

Intervals use event timestamps, whose precision depends on the browser. They do not measure electrical latency. Repeat a series interrupted or affected by uninterpretable timing. Statistics describe only comparable intervals and the stated sample.

The browser’s own double-click signal is separate from presses and never increases the press total. An intentional double-click demonstrates the display, not detection of an intermittent fault. To compare two series, keep the button and threshold the same. Choose one changed condition, or explicitly repeat the same setup. Each saved series retains its optional context label locally; that label never goes to Analytics.

CPS: a fixed observation window

The first primary press or single-finger touch starts the round and counts. Each new press requires a release; the total is divided by the selected full duration. Mouse-like input and touch use separate modes. The page cannot reliably distinguish a mouse from a trackpad or certify human performance.

The window includes its start and excludes its end. Event timestamps determine which presses belong to it. Some WebKit mouse events use a separate platform clock: their intervals stay in that clock, while the visible timer starts on receipt. The initial delivery delay is then unavailable. Finalisation takes 250 ms to include briefly queued inputs from before the deadline, without extending the scored window.

Leaving focus, hiding the page, cancelling touch or an unusable timing interruption invalidates the round. A timer callback gap greater than 500 ms also invalidates it. No interrupted score is exported. Results are local totals and one count per second; they do not certify physical latency, a switch or an anti-cheat result.

Measure a CPS round

Verified configurations

8 September 2026 baseline · software verification. Tests run with Playwright on macOS 26.6.2 (Apple silicon), using automated input. These are browser-engine checks, not tests of physical keyboards or mice. The checks cover keyboard protocols, mouse cycles, click series, interruptions and text reports.

9 September 2026 follow-up. The re-audit repeated 251 scenarios in Chromium, Firefox and WebKit: keyboard modes, cycles, interruptions, CPS and preferences. The 187 preference scenarios simulate browser and storage signals; they do not represent 187 devices. The checked catalogue contains 501 software references: 202 Windows, 248 Mac and 51 ChromeOS. Separate checks of the eight English and French tool pages cover automatic reports, actual PDF downloads, resets and simulated generation failures. Long and international report cases were also checked; this does not verify their reading in every PDF application.

WebKit here is an automated engine, not a physical iPad Safari test. English and French pages were also checked at simulated widths of 320, 390, 768 and 1440 pixels. Printable reports and the equipment checklist were rendered and visually checked as Chromium PDFs. Browser automation does not exercise every operating-system shortcut, device driver or physical input path.

Not yet tested on physical equipment: Windows keyboards and mice, Apple keyboards and trackpads, iPad Safari with external input, and Android Chrome with external input. No healthy or faulty switch has been certified. A useful next validation compares the same deliberate gesture on known working and suspect hardware, with the device, connection, system and browser recorded.

Privacy and reports

Positions, counters and test statistics stay in bounded page memory. They disappear when you leave or reload; manually chosen keyboard system, layout and format preferences can remain on your device. The report appears below each test as observations become available; CPS requires a completed round. “Download PDF” creates a file on your device, without a print dialog or an external conversion service. It contains your selected context, aggregate observations and method limits, not typed characters, a hardware identifier or text history. A downloaded file remains wherever you save it.

The PDF combines rendered page images with Unicode replacement text (“ActualText”). Selection, search and assisted reading depend on the PDF reader; compatibility in Preview, Acrobat and screen readers still requires checks in those applications. The format is not PDF/UA certified. The on-page report remains available before downloading.

The site retains its existing Google Analytics for visits and a few broad actions: first use, visible report, export and related navigation. Visibility of the automatic report does not prove it was read. No key, coordinate, input sequence or raw interval is attached to those events. Disabling or blocking measurement does not block the tools, and past interactions are not replayed when measurement becomes enabled.

Method changes

Mouse method 3 · 9 September 2026. Added guided holds for a chosen button and a timed, single-direction wheel observation. Reports retain the setup, reading windows and incomplete outcomes. These software rules still require validation on physical equipment. Double-click series retain method version 2.

PDF reports · 9 September 2026. Automatic reports and local PDF download are shared by the four tools in English and French. Checks cover creation, download, reset and simulated failures; they do not certify every PDF reader or universal accessibility.

Keyboard catalogue · 9 September 2026. Added international references and system-specific character mappings, including Mac Option symbols. This extends the reference drawing; the method for observing presses and releases is unchanged. Source checks and software checks do not certify physical equipment.

Tools v3 · 8 September 2026. Added CPS (protocol cps-v1), direct keyboard modes, guided mouse checks and local context for comparison series. The keyboard and click observation engines retain method version 2. Browser checks cover the new paths; physical equipment validation remains pending.

Version 2 · 8 September 2026. Event timestamps, keyboard protocols, comparable click series and reports that distinguish interruptions. Software checks and physical hardware tests remain separate evidence categories. Version 1 introduced keyboard coverage, logical buttons and the initial exploratory threshold.