Typing Analytics: per-key heatmaps, rhythm fingerprints and session replay

The measurements, and what each is for

Per-key heatmap

Which letters are slow, which are error-prone, and which have quietly gone backwards since last month. The single most useful view for deciding what to practise.

Transition timing

Every letter pair, ranked by how long your hands take to move between them. Slow pairs are where an intermediate typist's missing speed usually lives.

Rhythm fingerprint

Your inter-key intervals drawn as a ring. A round ring means metronome-steady hands. Spikes are hesitations, and they cost more than any single mistake.

Consistency score

How repeatable your speed is, run to run. Rising consistency at flat speed is real progress, and it is the thing that always improves before speed does.

Finger and hand balance

How the workload is distributed. A pinky doing an index finger's job is a technique problem no amount of practice volume will fix.

Session echo

A keystroke-level replay of any run. Hesitations glow, misses ring. It is the fastest way to see a habit you did not know you had.

Records and calendar

Personal bests per passage and per mode, and a practice calendar that shows frequency without turning a missed day into a punishment.

Raw speed against net speed

Both numbers are always shown. Raw counts every keystroke, net counts only the ones that survived. The gap between them is the price of your mistakes in words per minute, and watching that gap close is a far better measure of progress than watching either number on its own.

Where the data lives

Keystroke timings are computed and stored in your browser, so nothing about your analytics depends on a connection. What syncs to your account is the finished result of a session and the mastery summary built from it, never the text you typed. The privacy policy sets out the whole chain.

What each number actually measures

A word in typing measurement is five characters including spaces, not an actual word, which is the convention that lets scores from different passages be compared at all. Gross WPM counts everything you typed. Net WPM subtracts the cost of errors, and it is the figure worth watching, because raw speed with poor accuracy is mostly a measurement of how fast you can create work for yourself.

• Accuracy: the share of keystrokes correct on the first attempt, before any correction.
• Consistency: how even the intervals between your keystrokes are, which is the earliest signal that practice is working.
• Per-key response time: how long each individual key takes you, measured separately from the keys around it.
• Transition timing: how long each letter pair takes, which is where most lost time actually sits.

What the population numbers actually are

The largest study of modern typing, Dhakal and colleagues at CHI 2018, recorded 136 million keystrokes from 168,960 people and reports a single population mean of 51.6 WPM with a standard deviation of 20.2. That spread is the important part: a standard deviation of 20 on a mean of 52 means the range of ordinary is very wide, and a number that sounds slow in isolation is often unremarkable.

Two caveats travel with that figure, and both matter when you compare yourself to it. The participants were self-selected visitors to a typing-practice site, three quarters of them aged between eleven and thirty, so it describes people who chose to take a typing test rather than people in general. And the study publishes no breakdown by age, which is why the age tables circulating online are not traceable to it or to anything else comparable.

Why transitions matter more than keys

At any reasonable speed the individual reach is rarely what is slow. The join is. Most people are surprised by how concentrated their slowness turns out to be, with a handful of letter pairs accounting for a disproportionate share of the loss, and the worst offenders are usually sequences that use the same finger twice in a row, where the finger has to travel, arrive, press and travel again with no chance to overlap the movement.

This is also why an overall accuracy figure eventually stops being useful. A 97% average can comfortably hide 88% on four particular transitions, and those four are the entire reason your speed has stopped moving.

The four-sighting minimum

No key is called weak until it has been seen at least four times. Typing data is noisy at small samples, and a single fumbled Z on a bad afternoon is not evidence of anything. Without a minimum, an analytics page confidently reports a different set of weak keys every session and sends you chasing noise.

The same caution applies to reading a trend. Distinguishing a genuine weakness from a bad session realistically takes a few thousand keystrokes, which is a handful of sessions rather than one. The tracker compares one half of your history against the other rather than the last result against the previous one, for the same reason.

Confusion pairs, and reading a plateau

Some errors are not random. When two keys are repeatedly typed for each other, that is a confusion pair, and it is a different fault from being slow: the movement is wrong rather than merely unpractised. Naming the pair matters, because the fix is to slow that specific pair down until the correct finger reliably wins, not to do more general practice.

A plateau in the headline number is usually not a plateau in the underlying skill. Consistency generally improves before speed does, so during the weeks when WPM refuses to move, the variation between your keystrokes is often already tightening. That is the number to watch when the obvious one is flat.