Device Bench

Device test · Input

Typing Test

Type the stream of words and read four figures you can work out yourself: words per minute over the characters that matched, raw words per minute over every character you typed, accuracy as the ratio between them, and consistency as the spread of your per-second speed. The formulas are printed below and nothing on this page is adjusted by anything they do not mention. Pasted input is refused by name, a run under five seconds prints no WPM at all, and nothing you type leaves this device.

WPM— Raw WPM— Accuracy— Consistency— Time60.0 s left

Click the box and start typing. The clock starts on your first character and stops itself at the end of the mode; a space submits the word and moves the stream on. Under 5 seconds nothing is scored.

The stream is drawn from a first-party list of 300 common English words: the first 120 are the ones ordinary English is mostly made of and are drawn three times as often as the other 180, and the generator never emits the same word twice in a row. The seed beside the buttons is the whole of the randomness — the same seed and the same mode give the same words, forever — so a run can be repeated exactly rather than approximately.

On a phone or tablet this is a different instrument: autocorrect, predictive text and swipe entry all produce characters you did not press, so a touch-keyboard result is not comparable with a hardware keyboard — the test does not block it and does not correct for it.

No result yet. Finish a run — or stop one after at least 5 seconds — and the four figures are repeated here beside the character counts they came from.

WPM (correct characters ÷ 5 ÷ minutes)—
Raw WPM (all typed characters ÷ 5 ÷ minutes)—
Accuracy (correct ÷ all counted)—
Consistency (CV of the per-second speeds)—
Words submitted—
Characters counted—

How this test works

Five figures are on the strip and every one of them is arithmetic you can repeat by hand. WPM is correct characters ÷ 5 ÷ minutes; raw WPM is every counted character ÷ 5 ÷ minutes; accuracy is correct characters ÷ every counted character, as a percentage; consistency is the coefficient of variation of the per-second speeds, also as a percentage, where a lower figure is a steadier run; and the last field is the time remaining while you type and the time typed once the run is over. Five characters is the standard word, so the two speed figures are character counts wearing a friendlier unit. A character of a word is correct when it matches the letter the stream asked for in that position, and the single space that submits a word is correct only when the word before it was complete and correct — which is why one wrong letter in a four-letter word costs two characters and not one. Characters typed past the end of a word have no letter to match, so they are always wrong.

The clock is performance.now(), which is monotonic, and it starts on your first counted character rather than on a button, so no reaction time is folded into the measurement. A space with nothing to submit is not a counted character and does not start it: typed before any letter of a word it is ignored, and so is a second space straight after one, rather than either being scored as a wrong character. The counted window is exactly the mode's length from that first character: a character arriving after the boundary ends the run instead of being counted — a space included, since the space that submits a word is a counted character like any other — so the elapsed time the figures divide by is 15, 30, 60, 180 or 300 seconds exactly, and stopping early can only ever print less than that. Every character is counted at the moment it arrives, and backspace is never counted and never un-counts — a character you typed happened, so a deletion tidies the word on screen without editing the ledger, which is the entire difference between the raw figure and the adjusted one, and a correction made inside a word cannot add to it either, because the letters such an edit slides along are letters nobody typed again. Each completed second contributes one consistency sample, taken as the correct characters inside that second ÷ 5 × 60, each character counted in the second it arrived in rather than in the second a frame noticed it; the figure is the population standard deviation of those samples divided by their mean, it needs two completed seconds to exist, it prints nothing when no character was correct, and the final part-second is never a sample. The words themselves come from a 300-word first-party list drawn against a visible integer seed, so the same seed and mode reproduce the same stream exactly.

The limits are worth stating plainly, and the first one is that WPM depends on the difficulty of the text. This stream is common words — 300 of them, all three to five letters, no capitals, no punctuation, no numbers — which reads high against punctuated prose, and a figure from here is not comparable with a figure from a test that made you type quotation marks and proper nouns. The second is the keyboard you are using. On a phone or tablet this is a different instrument: autocorrect, predictive text and swipe entry all produce characters you did not press, so a touch-keyboard result is not comparable with a hardware keyboard — the test does not block it and does not correct for it, because the page cannot tell a corrected word from a typed one and would be inventing a distinction it cannot make. The third is the sample. A run under five seconds prints no WPM at all, because the divisor is small enough for one keystroke to move the answer by tens of words a minute, and consistency on a 15-second run rests on 15 samples, which is a coarse picture of a rhythm. The fourth is that this instrument is a visual one: the word stream is drawn for the eye and is deliberately hidden from screen readers, because a region that re-announced the whole line on every keystroke would be noise rather than access — so this test cannot be taken without sight of the stream, and there is no spoken or audible alternative to it here. Nothing here is a verdict about you: it is a count of characters over a stretch of time, and it says nothing about your accuracy on your own writing, your keyboard, or anybody else's numbers.

Frequently asked questions

How is the WPM figure worked out?
Correct characters divided by five, divided by the minutes the run lasted. Five characters is the standard word that every typing test divides by, so 300 correct characters in one minute is 60 WPM. Two details decide what counts as a correct character. The letters of a word are correct when they match the letter the stream asked for in that position, and the single space that submits a word is correct only when the word before it was complete and correct — so one wrong letter in a four-letter word costs two characters rather than one. Backspace is never counted and never un-counts: a character you typed happened, and deleting it does not make it not have happened. A correction cannot add to the ledger either: an edit made inside a word slides the letters after it along without anyone pressing them again, and letters that only moved are not counted a second time — what counts is what arrived. Raw WPM is the same arithmetic over every character that was counted, right or wrong, which is why the two figures differ and why both are printed.
Why is my raw speed higher than my WPM?
Because raw counts every character that entered the ledger, right or wrong, and WPM counts only the ones that matched. A keystroke that submits nothing enters neither: a space typed before any letter of a word, or a second space straight after one, is ignored rather than counted as a wrong character. Both divide by five and by the same elapsed minutes, so the gap between them is made entirely of wrong characters — including the submitting space after a word that was not right. If the two figures sit close together, almost everything you typed landed; if raw runs far ahead, the fingers were fast and the accuracy was not, and the accuracy percentage is that same fact stated as a fraction: correct characters divided by every character counted. There is no hidden penalty in either number and no adjustment beyond the ones this answer describes.
Why does pasting text into the box do nothing?
Because pasted input is not typing, and a test that accepted it would be measuring your clipboard. The page refuses a paste or a drop by name on the line under the box rather than swallowing it silently, because a paste that vanishes without a word looks like a bug. Nothing is inserted, no character is counted and the clock is not touched. Autocorrect is a different case and is deliberately allowed through: on a touch keyboard the page cannot tell a corrected word from a typed one, so it counts what appeared and tells you plainly that a phone result is a different measurement.
What does the consistency figure mean?
It is the coefficient of variation of your per-second speed, printed as a percentage, so a lower figure is a steadier run. Every completed second contributes one sample — the correct characters typed inside that second, divided by five, multiplied by sixty — and the figure is the population standard deviation of those samples divided by their mean. A run typed at one steady rate the whole way through reads near zero; a run of bursts and pauses reads high. Each character is attributed to the second it arrived in rather than to the second a frame noticed it, so a browser that drops frames prints the same figure as one that does not. It needs at least two completed seconds to exist at all, the final part-second is never a sample, and it prints nothing at all when no character was correct, because a mean of zero has no ratio to divide by.
Will my result be lower on a phone, and is it comparable?
It is not comparable, and this page says so rather than adjusting for it. On a phone or tablet this is a different instrument: autocorrect, predictive text and swipe entry all produce characters you did not press, so a touch-keyboard result is not comparable with a hardware keyboard — the test does not block it and does not correct for it. There is a second reason not to carry a figure from here to anywhere else: WPM depends on the difficulty of the text, and this stream is common words, which reads high against punctuated prose with capitals, numbers and unfamiliar vocabulary. Compare a run from this page against another run from this page, on the same mode, and the comparison means something.

Related tests and guides