Keyboard Latency Glossary

Forty-four terms used across this site, defined in one to three sentences each — with the pairs people routinely confuse set side by side.

Run the keyboard latency test

Published 15 August 2026 · Last updated 15 August 2026

Every entry has its own anchor, so you can link straight to a definition — for example the jitter entry lives at /guides/glossary/#jitter. Where a term is measured or explained in depth elsewhere on the site, the definition links to that tool or guide.

One note on vocabulary before you start. This site uses latency in a specific, narrow sense: the browser-observed delay between the browser creating a key event and the page's JavaScript handling it. That is only one segment of the full input chain, and the glossary marks clearly which terms describe segments a browser can see and which describe hardware behaviour it cannot. The measurement methodology page sets out the exact formulas.

Terms people mix up

Six pairs account for most of the confusion in keyboard discussions. If you read nothing else here, read these.

Pairs of terms that sound interchangeable but are not
Pair	The actual difference
Ghosting vs blocking	Ghosting reports a key you never pressed. Blocking drops a key you did press. Both come from the same ambiguity in a switch matrix; blocking is the deliberate, safer response to it.
Latency vs response time	Latency is a delay between a named start point and a named end point. "Response time" is a display specification — how fast a pixel changes colour — that keyboard marketing borrows without saying which segment it means.
Jitter vs standard deviation	Standard deviation measures how wide the spread is. Jitter, as defined here, measures how much each sample differs from the one immediately before it. Shuffle your samples and the standard deviation is unchanged while the jitter changes.
Polling rate vs scan rate	Polling rate is how often the host computer asks the keyboard for its state. Scan rate is how often the keyboard's own controller reads its switch matrix. They are set by different parties, and neither is visible to a web page.
Repeat rate vs polling rate	Repeat rate is an operating system typing setting — how fast a held key repeats, usually a handful to a few dozen characters per second. Polling rate is a USB transfer setting in the hundreds or thousands of hertz. They are unrelated.
Mean vs median	The mean is the average and shifts noticeably when one press is very slow. The median is the middle value and barely moves. A large gap between the two is itself a signal that something intermittent is interfering.

A to Z

#

2.4 GHz wireless: A dedicated radio link between a keyboard and its own small USB receiver, using the vendor's protocol rather than Bluetooth. Because one vendor controls both ends, these links generally add less delay than Bluetooth, and some support high report rates — but the actual figure is vendor-specific and rarely published, so treat marketing numbers as claims rather than measurements.
6KRO (six-key rollover): A keyboard that reports at most six ordinary keys held simultaneously, plus the modifier keys, which are carried separately. The limit comes from the fixed-size keyboard report used by the USB HID boot protocol. See what N-key rollover means, or check what your browser sees with the N-key rollover test.

A

Actuation point: The distance a keycap travels downward before the switch registers a press. Mechanical switches typically actuate somewhere in the first millimetre or two of travel depending on the design, and Hall-effect and optical analogue switches often let you set it yourself. A shorter actuation point shortens the physical part of the press, not the electronics that follow it.
Anti-ghosting: Design measures that stop a keyboard reporting keys nobody pressed — most reliably a diode behind every switch, so current cannot take an unintended path through the matrix. Cheaper implementations instead block the extra key. The ghosting test shows which combinations reach your browser intact.

B

Blocking (key jamming): When a keyboard refuses to register an additional key because reporting it would be ambiguous, so a key you genuinely pressed is dropped. Blocking is the safe alternative to ghosting: you lose a keypress rather than gaining a phantom one. Ghosting vs key rollover works through both failure modes.
Bluetooth latency: The delay a Bluetooth link adds. Bluetooth Low Energy devices exchange data on a negotiated connection interval, and a keypress generally waits for the next one, so the interval length sets the added delay. Battery-saving behaviour commonly pushes that interval well beyond the shortest the devices could agree on, which is why the same keyboard can feel different on two hosts.
Bounce: The brief electrical flicker as a mechanical contact makes or breaks, during which the circuit reads as alternately closed and open. It lasts a fraction of a millisecond to a few milliseconds depending on the switch and its age, and firmware is expected to filter it out — see debounce. Unfiltered bounce is what people experience as chatter.

C

Chatter: One physical press producing two or more registered keystrokes. It usually means contact bounce is outlasting the firmware's debounce window, often because a switch has worn or picked up contamination. The chatter test reports it as a pattern — "possible chatter detected" — never as a verdict about your hardware; what keyboard chatter is explains why that distinction matters.
Coefficient of variation (CV): Standard deviation divided by the mean, shown as a percentage. It expresses spread relative to size, so a 2 ms spread around a 5 ms median and a 2 ms spread around a 40 ms median are correctly treated as very different situations. The consistency label on this site is derived from CV and is always printed next to the number that produced it.

D

Dead key: Two unrelated meanings, which is why the term causes trouble. In typography a dead key produces no character by itself and instead modifies the next key pressed — the usual way to type accented characters on many layouts. Colloquially, people also call a key "dead" when it never registers at all; for that sense, the keyboard tester shows every key you have and have not yet hit.
Debounce: The firmware technique of ignoring further state changes on a switch for a short window after it changes, so one press yields one keystroke despite contact bounce. Essentially every keyboard does this; the trade-off is that the window is added to the time between pressing a key and the keyboard reporting it.
Debounce time: The length of that ignore window. Longer windows suppress chatter more reliably but add delay; shorter windows are quicker but let a worn switch double-fire. Values are firmware-specific, commonly a few milliseconds, and only some keyboards expose the setting to the user.
DOMHighResTimeStamp: The floating-point millisecond timestamp used by performance.now() and by KeyboardEvent.timeStamp in modern browsers. It is measured from the page's time origin and can express fractions of a millisecond — subject to the browser's timer resolution. Because both values sit on the same clock, subtracting one from the other gives a meaningful interval.

E

Event cadence (browser-observed): The interval between consecutive keyboard events as the browser delivers them. It is shaped by how fast you type, by the operating system's auto-repeat when a key is held, and by browser scheduling — it is not a readout of a polling rate. Measured on the event cadence test and explained in polling rate vs browser event cadence.
event.code: The event property identifying the physical key position, independent of layout and modifiers: the key to the left of "1" reports Backquote whatever character it actually prints. Use it for game controls and shortcuts that should stay in the same place on any layout. Inspect it live in the event tester.
event.key: The event property giving the character or named value the key produces right now, after layout and modifiers are applied — "a", "A", "Enter", "Shift". Use it when you care what the user typed rather than which physical key they hit.
event.location: A numeric hint distinguishing keys that appear more than once: standard, left, right, or numeric keypad. It is how you tell left Shift from right Shift, or the numpad Enter from the main one. The key test shows it for any key you press.

F

Frame time: The interval between one drawn frame and the next — the reciprocal of the refresh rate. At 60 Hz a frame lasts about 16.7 ms, at 144 Hz about 6.9 ms, at 240 Hz about 4.2 ms. A keypress arriving just after a frame has been composed waits for the next one, so on average it waits about half a frame.

G

Ghosting: When a keyboard reports a key that was never pressed, because current found an unintended path through a diodeless switch matrix. It classically requires three held keys that form three corners of a rectangle in the matrix; the fourth corner appears from nowhere. Modern keyboards prevent it with per-switch diodes or by blocking the extra key — see what keyboard ghosting is.

H

HID (Human Interface Device): The USB device class covering keyboards, mice, gamepads and similar input devices. It defines a standard way for a device to describe its own report format, so an operating system can use it without a vendor-supplied driver. Bluetooth input devices use the same descriptor model over a different transport.

I

Input lag: The end-to-end delay between doing something and seeing the result — for a keyboard, from your finger moving to a character appearing on screen. It includes the switch, the firmware, the transmission link, the operating system, the application and the display, which is why no single component's specification equals it. How to reduce keyboard input lag takes the chain segment by segment.
Interquartile range (IQR): The distance between the 25th and 75th percentiles — the width of the middle half of your samples. It describes spread without being dragged around by a single very slow press, and it is the basis of the outlier rule used on this site.

J

Jitter: On this site, the mean absolute difference between consecutive samples: how much each press differs from the one immediately before it. That makes it a measure of instability over time rather than of overall spread, which is why it is reported alongside standard deviation rather than instead of it — see the comparison table above. Why latency results vary covers the usual causes of a high figure.

K

Key rollover: How many keys can be held down at once while the keyboard still reports each of them correctly. "2KRO", "6KRO" and "NKRO" describe increasing capability. Rollover is about simultaneous keys; ghosting is about phantom keys, and the two are frequently conflated.
keydown: The event fired when a key is pressed, and fired again repeatedly while it is held if the operating system's auto-repeat is active. Repeated events carry repeat: true, and this site excludes them from press-latency statistics because they measure the OS repeat timer rather than a fresh press.
keypress: A legacy event that fired only for keys producing character input. It is deprecated and was never implemented consistently across browsers; current code should use keydown for key identity and the beforeinput or input events for text content.
keyup: The event fired when a key is released. Pairing each keyup with its matching keydown is how hold duration is calculated, and a keydown with no keyup is exactly the pattern the stuck key test looks for.

L

Latency: The delay between a cause and its observable effect. Used without qualification the word is ambiguous, so this site always names the segment: the flagship figure is browser input event latency, the time from the browser creating a key event to the page's JavaScript handler running. Start with what keyboard latency is, then measure your own with the browser-based keyboard latency test.

M

Matrix scan: The keyboard controller's method of reading its keys. Switches sit at the crossings of a grid of rows and columns; the controller energises each row in turn and reads which columns respond, which needs far fewer connections than one wire per key. It is also why a matrix without diodes can be tricked into reporting a phantom key.
Mean: The arithmetic average of your samples. A single very slow press — a garbage-collection pause, a background process waking up — pulls it noticeably upward, which is why it belongs next to the median rather than in place of it.
Median: The middle value once the samples are sorted, also written P50: half your presses were faster and half were slower. It is the headline figure on this site because a handful of outliers barely move it.

N

N-key rollover (NKRO): A keyboard that can report every key held at once, with no fixed ceiling. Over USB this requires a report format other than the six-slot boot-protocol keyboard report. A browser test can only tell you how many simultaneous keys actually reached the page, which is a lower bound on what the hardware can do — see what N-key rollover is and the rollover test.

O

Outlier: A sample far enough from the middle of the distribution to be treated separately. This site uses the conventional rule: anything below P25 − 1.5 × IQR or above P75 + 1.5 × IQR. Outliers are counted and displayed rather than silently deleted, because a run full of them is itself the finding.

P

Percentile (P95, P99): The value below which a given share of your samples falls: P95 is the level that 95% of your presses came in under. Percentiles describe the bad cases the median hides, which is usually what "it feels laggy sometimes" actually means. P99 needs a substantial sample to mean anything — below 20 valid presses this site shows a dash rather than inventing a figure.
Polling rate: How often the host computer asks the keyboard for its current state, in hertz. 1000 Hz means one request every 1 ms; a keypress that becomes ready at a random moment waits on average half that interval, so roughly 0.5 ms typical and up to 1 ms worst case. Convert rates with the polling rate calculator, and read latency vs polling rate for why a higher rate shortens a wait rather than removing a delay. A web page cannot measure your polling rate.

R

Refresh rate: How many times per second the display draws a new image. It sets the granularity at which any input can become visible: 60 Hz gives a new opportunity every 16.7 ms, 144 Hz every 6.9 ms, 240 Hz every 4.2 ms. See frame time.
Repeat delay: The pause between holding a key down and the first automatic repeat. It is an operating system setting, typically adjustable somewhere between roughly a quarter of a second and a second. Measure what yours actually does with the repeat rate test.
Repeat rate: How quickly a held key repeats once the repeat delay has elapsed, expressed in characters per second. It is an operating system setting and has nothing to do with polling rate, despite both being quoted "per second".
Response time: Used two very different ways. For displays it is a manufacturing specification: how long a pixel takes to change colour. For keyboards it is loose marketing language, usually meaning some unspecified subset of the input delay — always ask which segment before comparing two figures. Latency vs response time pulls the two apart.

S

Scan rate: How often the keyboard's own controller reads its matrix, independent of how often the host polls the device. Fast scanning with slow polling still waits for the host; slow scanning with fast polling just reports stale state more often. Both matter, and neither is observable from a browser.
Standard deviation: The typical distance of a sample from the mean, computed here with the n − 1 denominator used for a sample rather than a whole population. It describes how wide the distribution is, but says nothing about the order the samples arrived in — which is exactly what jitter adds.
Switch actuation: The moment a keyswitch registers a press: contacts closing on a mechanical switch, a beam breaking on an optical switch, or a magnetic field crossing a threshold on a Hall-effect switch. It is the first event in the input chain and the one a browser has no visibility of whatsoever — see why browser tests can't measure hardware latency.

T

Timer resolution: The smallest time difference a clock can actually express. Browsers deliberately coarsen their high-resolution clocks as a security measure, and how much depends on the browser and on whether the page is cross-origin isolated. If your clock rounds to whole milliseconds, sub-millisecond jitter cannot be resolved no matter how good the keyboard is — so this site probes the resolution and reports it alongside every result. The methodology page shows how the probe works.

U

USB HID boot protocol: A simplified, fixed report format defined so a keyboard works in a system BIOS or bootloader before any full driver has loaded. Its keyboard report carries a modifier byte and six key slots, which is where the familiar six-key limit comes from. Keyboards offering NKRO over USB use a different report format once the operating system is running.

Keyboard Latency Test Measure browser-observed input timing, with median, percentiles, jitter and a consistency label. Measurement Methodology The formulas behind every statistic here, plus the validation rules and timer-resolution limits. All Guides Nineteen guides grouped into core, diagnostic and technical, with a start-here path. What Is Keyboard Latency? The long-form version of the latency entry, with the full input chain laid out. Keyboard Event Tester See key, code, location, modifiers and timestamps for any key you press.

Keyboard Latency Glossary

Terms people mix up

A to Z

#

A

B

C

D

E

F

G

H

I

J

K

L

M

N

O

P

R

S

T

U

Related pages