Scoring version 1
How scoring works
Every number AbackType shows, defined exactly. Each saved result records the scoring version it used; if a definition ever changes, it gets a new version and old results keep their meaning.
Speed
WPM counts characters that are correct in the text you end with, including correct spaces and line breaks, divided by five (the conventional word length) and by the minutes taken.
wpm = correct characters in final text / 5 / minutes
Raw speed counts every character keystroke, right or wrong, including ones you later deleted. Backspace is not counted. Raw speed shows how fast your fingers moved; WPM shows how much correct text you produced.
raw = all character keystrokes / 5 / minutes
Accuracy and errors
Accuracy is measured per keystroke, at the moment you type it: was it the character expected at the caret? A mistake you fix still counts, which is why accuracy can be below 100% even when the final text is perfect.
accuracy = correct keystrokes / all character keystrokes x 100
- Corrected errors: wrong characters you removed with Backspace or word delete.
- Uncorrected errors: problems left in the final text. Wrong characters, extra characters typed past the end of a word, characters skipped when you pressed space early, and a space where a line break belongs (or the reverse).
Consistency
Keystrokes are counted in each whole second of the test. Consistency is 100 minus the coefficient of variation of those counts, clamped to 0 to 100. Perfectly even typing scores 100. It needs at least 5 whole seconds; shorter tests show n/a.
Timing
- The clock is the browser's monotonic high-resolution timer, unaffected by system clock changes.
- It starts on your first character, not when the page loads.
- Timed tests last exactly their duration; keys after the end are ignored.
- Other tests run from the first keystroke to the last.
- The timer does not pause. If you switch tabs during a test, the test stops and is not saved, because its timing would no longer be accurate.
- A pause longer than 10 seconds keeps a result out of best scores and comparisons.
- Results with fewer than 10 keystrokes are not saved.
- No network request is made while you type, so network speed never affects timing.
How input is read
- Text is split into words. Space (or Enter in code) moves to the next word. Space at the very start of a word is ignored, so a double space cannot skip a word.
- Errors stay inside their word, so one missed letter does not shift every following character out of line.
- Backspace can step back into the previous word only if it contains a mistake. Ctrl+Backspace (Option+Backspace on macOS) clears a word.
- Up to 12 extra characters are accepted per word; beyond that, keys are ignored.
- Paste is turned off in tests. Characters produced by holding a key down (auto-repeat) are ignored; holding Backspace still deletes.
- Input methods and phone keyboards are supported: composed text is scored when you commit it. Because timing is then less precise, those results are marked and kept out of comparisons.
- Stop-on-error mode records a wrong key as a mistake but keeps the caret in place until you press the right key.
Code and multi-line text
- Line breaks are typed with Enter. A faint return mark shows the end of each line.
- Leading indentation is filled in automatically. Tabs count as two spaces.
- Runs of spaces inside a line become one space; trailing spaces and blank lines are removed.
- Only the characters are compared. Nothing is run or checked for meaning, so a perfect score says nothing about whether you know the language.
Per-key statistics and suggestions
For each expected character, AbackType counts attempts and mistakes. A keystroke is timed only when the previous keystroke was correct, this one is correct, and the gap is at most 2 seconds, so pauses and corrections do not distort key speed. Letter pairs are counted the same way.
A key or pair is suggested for practice only when all of these hold:
- At least 600 keystrokes saved in the last 60 days (up to 50 results, keyboard input only).
- At least 40 attempts for a key, or 25 for a pair.
- For accuracy: the lower bound of the 95% Wilson interval of its error rate is at least 25% above, and 1 point above, your error rate across all keys.
- For speed: with at least 30 timed samples, the lower 95% bound of its average time is at least 15% above your typical key time.
Suggestions describe characters and pairs you typed; they never guess which finger or hand is at fault. They are labelled tentative until the sample is large, and you can turn them off in Settings.
What these numbers do not tell you
- Speed on one text does not predict speed on very different text; compare like with like.
- A drill score mostly reflects that drill. Retests use words the drills never show, to check that a gain carries over.
- Prompt Construction checks the structure of an instruction. It is not a quality grade and is never mixed with typing scores.
- None of this measures programming ability or productivity.