Almost everything people believe about their typing speed comes from a number produced under conditions nobody wrote down. A word is five characters rather than a dictionary word; net speed counts only the characters you got right while raw speed counts everything; a one-minute run reads two to five words per minute above a five-minute average. None of that is secret, but all of it is left out of the sentence "I type 80 words per minute", which is why two people saying it can mean two different things. This guide covers the measurement, the practice that actually moves it, and the hardware underneath — including the parts a web page genuinely cannot see. Everything here runs in your browser, and nothing you type is stored or uploaded.
A word is five characters, and that is not a quirk
The five-keystroke word exists because a typing test cannot know where your word boundaries were — it can only count characters. Standardising a word at five characters makes every test comparable regardless of the text, which is why 300 characters is 60 words whether they are “the cat sat on the mat” or “extraordinary circumstances warranted reconsideration”. The practical consequence is that written words and test words differ: English averages nearer six characters per word, so the five-keystroke figure usually runs slightly higher than the words divided by minutes you would compute from a document. The WPM Calculator on this site reports both, plus a strict net figure with uncorrected errors subtracted, because the interesting thing is how far apart the three are for the same session.
Net, raw and why the gap between them is the real score
Net words per minute counts only the characters you got right. Raw counts every character you typed, wrong ones included. The difference is your accuracy expressed as speed, and it is the most actionable number on the page: two typists at 80 raw WPM are having very different days if one finishes at 98% accuracy and the other at 88%. The reason is mechanical rather than moral. A correction costs a Backspace press and a decision, and the decision is the expensive part — you have to notice the error, locate it and resume. That is why the accuracy test on this site has no clock at all: a timer pushes people into exactly the error rate they are trying to reduce, so the one measurement that predicts whether speed will stick is the one that cannot be taken under time pressure.
The three layers between your fingers and the page
A browser cannot see your keyboard. A keystroke reaches this site only after three layers have had their say: the keyboard's own matrix, where ghosting and anti-ghosting decide what the board reports; the operating system, which intercepts combinations it owns and swallows several Win+ and ⌃ combinations before any application sees them; and the browser, which reserves some chords for its own interface. That is why the Key Rollover Tester reports what arrived rather than what you pressed, and why an absent combination is not proof that your keyboard cannot do it. What the test is genuinely good at is showing a drop — hold six keys, watch three arrive, and you have found the limit of the board rather than of the measurement.
Input latency is the whole chain, not the keyboard
"Keyboard latency" usually names the first stage of five: switch scan, USB polling (125 Hz to 8 kHz, so 0.125–8 ms), the operating system's input queue, the browser's event loop and render, and the display's own refresh. The display alone is a floor no hardware can beat — a 60 Hz panel presents a frame every 16.7 ms, so a press lands somewhere within one of those frames. That is why this site's latency test times the key press to the second animation frame and reports tens of milliseconds rather than the sub-millisecond figure a keyboard's marketing quotes, and why the page states the comparison is only valid between samples taken under the same conditions: a throttled background tab or a laptop in a power-saving state moves this number without the keyboard having changed at all.
Keyboard layouts: the geometry is real and the conclusion is not
Dvorak was designed in 1936 to put the commonest letters under the strongest fingers and to alternate hands; Colemak, in 2006, to get most of that benefit while leaving the ten most common keys where QWERTY already puts them, so muscle memory and every shortcut survive. The Layout Comparison tool computes the difference rather than quoting it: each key's distance from its finger's home position, weighted by a published English letter-frequency table, on a stagger-matched grid at the standard 19.05 mm key pitch. On ordinary English text that lands at a bit over 45% less finger travel for Dvorak and slightly more than that for Colemak — figures that depend on the frequency table used and on the fact that travel ignores same-finger bigrams, which is why the page prints its method rather than a borrowed number. What the geometry does not price is everything else — every keyboard, laptop and shortcut assumes QWERTY, and no study has measured a convincing trained-typist speed advantage for either alternative. The honest reading is a comfort and effort argument, not a speed one.
How to actually get faster
Accuracy first, in short sessions: ten focused minutes a day beats an hour on Sunday, and the gate should be an accuracy rather than a speed. That is the whole design of the practice lessons here — each one passes on accuracy, nothing is timed, and the sequence runs home row, full alphabet, numbers, punctuation, common words, then real sentences. Two habits do most of the work. Stop looking at your hands, which is what the home-row lessons exist to enforce. And slow down when the error rate climbs rather than pushing through it: an error rate that doubles while your speed rises by a tenth is a net loss, and the Backspace Error Rate Calculator exists to make that visible — one correction per twenty keys against one per hundred is the difference between a burst and a working pace.