A keyboard can look fine while typing an email and still fail the moment two or three keys go down together: in a game, in a fast typist's overlapping keystrokes, or in a shortcut like Ctrl+Shift+T. This guide walks through a complete hardware check with our keyboard tester: coverage of every key, rollover, ghosting and blocking, key repeat, stuck keys and chatter. It explains what each result means and which mistakes make a good keyboard look broken.
Why keys interfere with each other
Most keyboards do not wire every key to the controller separately. The keys sit at the crossings of a grid, the matrix: say 8 rows and 18 columns for a full-size board. The controller drives one row at a time and reads which columns answer. That works perfectly while one key is down. Now press three keys at three corners of a rectangle in that grid, for example row 2/column 3, row 2/column 7 and row 5/column 3. While the controller drives row 5, current can flow through the row 5/column 3 key into column 3, back through the row 2/column 3 key into row 2, and through the row 2/column 7 key into column 7, so the controller sees row 5/column 7 as pressed too. That phantom fourth key is ghosting.
Keyboards deal with it in one of two ways:
- Blocking: the controller notices the ambiguous pattern and refuses to report the extra key. Nothing phantom appears, but a key you really pressed may be ignored.
- Diodes on every switch: current can only flow one way, so the phantom path does not exist and every combination is read correctly. That is what most mechanical "anti-ghosting" or "NKRO" boards do.
Rollover: how many keys at once
Rollover is the number of keys the keyboard can report simultaneously. Two limits stack on top of each other:
- The matrix. Without diodes, some combinations of only two or three keys may already be blocked, while others work with six.
- The USB report. A keyboard using the standard USB boot protocol sends at most six ordinary keys plus the eight modifiers (left and right Ctrl, Shift, Alt and the Windows/Command key) in one report. That is where the familiar 6KRO comes from. Boards that claim NKRO use a different report format to go beyond it.
The tester shows both the total keys held and how many of those were ordinary keys, because Shift, Ctrl, Alt and Meta travel in a separate part of the report.
The test, step by step
Press Start test and work through it in this order:
- Coverage. Press every key once, row by row. The counter heading shows, for example, "87 of 104 keys tested", and the list underneath names the 17 you missed. On a tenkeyless board, switch the layout to 87 keys first, or the numpad will always count as untested.
- Rollover on the home row. Hold A, S, D and F, then add J, K, L and ;. On a typical office keyboard the count stops somewhere between 2 and 6; on a 6KRO board it stops at exactly 6 ordinary keys however many you add; an NKRO board climbs to 8.
- Gaming clusters. Hold W, A and D, add Shift and Space, then try Ctrl + W + A + Space. Manufacturers often route these keys so that they work together even when other combinations block.
- Ghosting hunt. Hold Q, W and A, then Q, W and S, and other little triangles around the board. Watch for a key you did not touch lighting up. If one does, that board ghosts; if instead the third key stays dark, it blocks.
- Repeat. Hold J for two seconds. The repeat panel reports the delay before the first repeat and the rate. With a delay of 500 ms and a repeat every 33 ms, it shows 500 ms and 30.3 per second (1000 ÷ 33). Those numbers are set by your operating system's keyboard settings, so they are a settings check rather than a hardware test; they should match what you configured.
- Stuck keys. Take your hands off the keyboard. Held now should be empty. A key that stays listed for more than five seconds is physically stuck, shorted by spilt liquid, or a sticky keycap pushing a switch down.
- Chatter. Tap a few suspicious keys quickly. The tool flags a press that arrives within 30 ms of the same key's release. Human fingers rarely do that; worn mechanical switches that bounce do. The 30 ms threshold is a practical rule of thumb, not a standard, so treat one or two flags during very fast tapping with caution and a key that flags on every slow press as a real fault.
Reading the results
| What you see | Most likely cause | What to try |
|---|---|---|
| A key never lights | Dead switch, broken membrane trace, or a key the browser never receives | Test it in a text editor; if it fails there too, it is hardware |
| A key you did not press lights up | Ghosting (no diodes) | Avoid that combination, or use a board with NKRO |
| Third or fourth key stays dark | Blocking or a rollover limit | Normal for office keyboards; check the combinations you actually need |
| Exactly 6 ordinary keys, never more | USB boot-protocol report | Some boards have an NKRO mode in their software or a key combo |
| A key stays held | Stuck keycap, liquid, short | Unplug before cleaning; compressed air under the cap |
| Double presses on one key | Worn or dirty switch | Clean, raise the debounce time in the keyboard's software, or replace the switch |
Common mistakes
- Testing keys the browser cannot see. Fn is handled inside the keyboard and never reaches the computer as a key. Media keys, the power key and some system shortcuts (Ctrl+Alt+Delete, the Windows key on its own, Alt+Tab) are taken by the operating system first. A "dead" Fn key in a browser test means nothing.
- Judging an ISO or JIS board against the ANSI map. An ISO board has a key next to the left Shift that an ANSI board lacks; it is reported as IntlBackslash and listed below the map, not drawn on it.
- Blaming the keyboard for the language setting. The tester matches keys by physical position (
event.code), so Vietnamese Telex, VNI, AZERTY or Dvorak make no difference to the map. If you want to see what character a key produces, the key code viewer showskeyandcodeside by side. - Switching windows mid-test. When the page loses focus, the browser stops reporting key releases, so the tool clears held keys rather than showing them as stuck.
- Holding Command on a Mac. macOS does not deliver key-up events for other keys while Command is down; the tester releases them when Command goes up, but a combination test that includes Command will look odd.
- Confusing repeat settings with hardware. A slow repeat or a long delay is a setting in the operating system, not a failing keyboard.
Run the test once when a keyboard is new, keep a note of its rollover and the combinations you rely on, and repeat it when something feels wrong. A two-minute check usually tells you whether the problem is the switch, the matrix or the computer.