What net WPM and raw WPM actually count
Keyboardo reports two speeds after every test, and they are the same formula applied to two different counts. Net WPM takes the characters that are correct in your final text, divides by five, and divides again by the elapsed time in minutes. Raw WPM takes every printable key you pressed — right, wrong, or later deleted — and puts it through identical arithmetic. Backspace appears in neither count. The time you spent pressing it appears in the denominator of both.
That last sentence is where the entire accuracy question lives. There is no penalty term in either formula and no points are deducted for mistakes. Errors do not subtract from your score; they fail to add to it, and repairs quietly spend the one resource the formula divides by. Once you see it that way, "should I go faster or be more careful" stops being a matter of temperament and becomes a question with an arithmetic answer.
What one uncorrected mistake costs
Net WPM is built from a character count with one extra rule: a word contributes its matching characters plus exactly one space, and it earns that space only if every character in it is correct. Characters you got right inside a broken word still count individually — but the word-completion space does not, so a single wrong letter usually costs you two characters rather than one.
Here is the same short word typed five ways. The target is "the" followed by a space, which is four characters of credit when it goes perfectly.
| What you typed | Printable keystrokes | Characters credited to net WPM |
|---|---|---|
| the + space | 4 | 4 — three letters plus the word-completion space |
| thw + space | 4 | 2 — t and h; the wrong letter earns nothing, and the word never completes, so the space is lost as well |
| thee + space | 5 | 3 — the three correct letters; the overflow character earns nothing and blocks the space |
| th + space, pressed early | 3 | 2 — t and h; the skipped letter was never typed and the word is incomplete |
| thw, backspace, e + space | 5 | 4 — the text is repaired, so the credit is full |
On a 60-second test the elapsed time is exactly one minute, so net WPM is simply your credited characters divided by five. Losing two characters therefore costs 0.4 WPM. That sounds negligible until you count how often it happens: twenty broken words in a minute is forty lost characters, which is eight word-equivalents, which is eight WPM you typed for and were not paid for.
The cost also depends on the clock, because minutes are the denominator. The same two lost characters are worth 0.8 WPM on a 30-second test, 0.4 WPM on a 60-second test and 0.2 WPM on a 120-second test. Short tests magnify every mistake, which makes them the better tool when you are hunting for error habits and the worse one when you want a stable estimate of your speed.
What a correction costs, keystroke by keystroke
Fixing a mistake restores the text but not the record of it. Take the repair from the bottom row of the table — you typed "thw", pressed backspace, typed "e", then the space. That is six key presses: five printable ones and a backspace. Each of the four things Keyboardo measures sees a different part of it.
- Net WPM sees four credited characters, exactly as if the word had gone in cleanly. The correction did its job.
- Raw WPM sees five printable keystrokes for a four-character word, because the wrong letter still counts as output.
- Accuracy sees four correct keystrokes out of five printable ones, or 80% across that word. The backspace is invisible to it, and the mistake is not erased by having been fixed.
- The clock sees all six presses. It is the only one of the four that charges you for the backspace at all.
The time cost is worth making explicit, because it is the part people underestimate. Typing at 60 WPM means placing 300 characters a minute, which is one key press roughly every 200 milliseconds. If your presses are evenly spaced — a simplification, but a fair one for a rough figure — the clean version of that word takes about 800 ms and the repaired version about 1,200 ms. The extra 400 ms is two key presses you never got to make, which on a 60-second test is 0.4 WPM gone.
So a corrected mistake and an uncorrected one cost about the same on a one-minute test, roughly 0.4 WPM each, by two completely different routes: the uncorrected one costs you credit, the corrected one costs you time. There is no version of the mistake that is free, and the only cheap error is the one you do not make.
Reading the gap between your numbers
Because net and raw run identical arithmetic over different counts, the distance between them is a direct measurement of wasted keystrokes. Suppose a 60-second test ends at 80 raw and 60 net. Raw 80 means 400 printable keystrokes went in; net 60 means 300 characters came out. A hundred keystrokes — a quarter of everything your fingers did — never reached the final text.
The characters figure on the results screen breaks that quarter into four numbers, shown as correct / incorrect / extra / missed:
- correct — characters matching the target, plus one space for every word you completed perfectly. This is the count net WPM is computed from.
- incorrect — wrong characters typed within the length of a word.
- extra — characters typed past the end of a word. They cost keystrokes, earn nothing, and prevent the word from completing.
- missed — characters of a word you skipped by hitting space early. They cost no keystrokes, but every one is a character of credit you did not earn, and the incomplete word forfeits its space too.
Those four point at different habits and do not respond to the same practice. A large extra count means you are overrunning the ends of words. A large missed count means you are reaching for the space bar early. A large incorrect count with the other two low means the letters themselves are going wrong, which is the only one of the three that is really a finger-accuracy problem.
Where the consistency score fits
Consistency is the fourth number on the results screen, and it is the one most closely tied to error habits even though it never looks at correctness at all. Keyboardo samples your raw speed once per completed second — the characters typed in that second multiplied by twelve, because a minute holds sixty seconds and a word holds five characters — and then scores how much those samples varied around their own average.
The score is 100 × (1 − CV), where CV is the standard deviation of the per-second samples divided by their mean, capped at 1 so the result never falls below zero. Two worked examples fix the scale:
- Four seconds at 60, 60, 60, 60 WPM: mean 60, standard deviation 0, so the score is 100.
- Four seconds at 90, 30, 90, 30 WPM: mean still 60, standard deviation 30, so CV is 0.5 and the score is 50 — identical average speed, half the consistency.
Because it is a ratio rather than a raw spread, the measure is scale-free: a 40 WPM typist and a 140 WPM typist with equally even rhythm score the same. And because a repair is arithmetically nothing more than a second in which fewer characters landed, bursts of errors and a low consistency score come out of the same seconds of the test.
How to train accuracy without just typing slowly
"Slow down" is true and useless, because the useful version has a target attached. What follows is practice design rather than evidence: no published study establishes an optimal error rate or an optimal session length, and any site handing you one has invented it.
- Find the speed at which the errors stop, and stay there. Not the speed that feels careful — the speed at which the incorrect count on your results screen actually reaches zero or near it. Clean typing tends to speed up on its own; sloppy typing has to be unlearned first.
- Use short tests to find errors and long ones to measure speed. This follows directly from the arithmetic above: the same mistake costs four times as much WPM on a 15-second test as on a 60-second one, so short runs make error habits impossible to ignore.
- Turn punctuation and numbers on when accuracy is the goal. They are independent toggles on word and time tests, and they add capitals, sentence punctuation and digits — the reaches where accuracy usually breaks first. Either toggle makes the result ineligible for the leaderboards, which is the correct trade when you are practising rather than competing.
- Watch the gap, not only the percentage. Accuracy is a keystroke ratio; the raw-minus-net gap is the output you actually lost. The two can move independently, and the gap is the one that converts into WPM.
- Use zen mode when the clock is the problem. It has no timer and no target text, so nothing is pushing you to guess at the next word. Zen results report net equal to raw and accuracy fixed at 100, precisely because there is nothing to be wrong against.
None of this trades speed away. On a test that reports net WPM, accuracy is the speed number — arriving by a different road. You can watch the two move together on any timed test.