Why Typing Speed Plateaus

Two things are commonly mistaken for one. A plateau is a period during which real improvement has stopped. A measurement plateau is a period during which real improvement continues and the instrument cannot see it.

Words per minute is a coarse instrument. Your consistency between attempts, your error rate on individual keys, and the time you lose recovering from a mistake can all improve substantially while the headline figure moves by one or two — well inside the noise of any single attempt.

So before diagnosing anything, take a different measurement. Per-key error rates from the weak-key diagnostic move detectably over a fortnight when overall speed does not. So does the gap between your best and worst minute on Marathon. If those are moving, nothing is stuck and the only problem is what you chose to watch.

The first big gain in typing comes from knowing where the keys are. The second comes from stopping treating words as sequences of keys and starting to treat common fragments as single motions.

If you have never made that transition, your speed is bounded by how fast you can make individual decisions, and no amount of additional practice on whole passages will change it, because passages give you each fragment too rarely to consolidate.

The symptom is a speed that is roughly the same on every kind of text. A typist who has chunked is markedly faster on common words than on unusual ones; a typist who has not is uniform, because every word costs the same per letter. Compare a common words test against a prose test: a small gap means the chunking gain is still available to you. The response is concentrated repetition of fragments rather than passages — common bigrams, word endings, the top hundred words.

A common and invisible plateau comes from operating permanently at the fastest pace you can manage, which is usually slightly above the fastest pace at which you are accurate.

What that produces is a stable equilibrium: pushing harder generates more errors, the corrections cost back exactly what the extra speed gained, and the net figure does not move no matter how hard you try. From the inside it feels like a hard limit. It is actually a balance point, and it can be moved.

The symptom is a large gap between gross and net speed, or a net figure that stays flat while gross climbs. The response is counter-intuitive: type slower on purpose until the error rate approaches zero, then let speed return. The zero-error accuracy drill exists for this, and the arithmetic explaining why it works is in accuracy versus speed.

Two typists can average the same speed with completely different distributions. One produces roughly even intervals between keystrokes. The other produces fast bursts separated by stalls — usually at the same places, on the awkward letter pairs and rare keys.

The burst-and-stall typist has a lower ceiling for a structural reason: the bursts are already near the maximum rate the hands can produce, so there is no headroom there, and all the remaining time is in the stalls. Practising general speed makes the fast parts marginally faster and leaves the stalls untouched.

The symptom is a much bigger gap than expected between your best short result and your sustained one, since a short window can be entirely burst. The response is to work on the intervals directly with the steady rhythm drill, and to find out where the stalls actually are — they are usually same-finger pairs or row jumps, both of which have their own drills for exactly that reason.

Hand geometry that was adequate at forty words a minute stops being adequate higher up. A flattened arch, wrists carrying weight, fingers that do not quite return home — none of these prevent moderate speed, and all of them cost a fraction of a second on every long reach.

At a moderate pace those fractions disappear into the gaps. At a faster one there are no gaps left, and the same fractions become the ceiling.

The symptom is errors that are consistently one column off in the same direction, or a specific reach — P, Z, the number row — that feels longer than it used to. That is a hand that has moved rather than a memory that has failed. The response is not more practice but a check of hand position, and a habit of resetting on the F and J bumps rather than correcting gradually.

Sometimes the honest answer is that your ordinary typing does not contain the thing that is slowing you down.

If your speed on prose is fine and collapses on anything with punctuation, numbers or symbols, then your prose figure is not plateaued at all — it is fine, and your real work is dominated by material the figure never measured. More prose practice cannot fix that. Comparing an ordinary result against a punctuation-heavy test or a numbers test shows the size of the gap directly.

And sometimes the plateau is a scheduling artefact. Practice that is irregular, or long and infrequent, produces much less than the same total time spread across more days, because consolidation happens between sessions. Before concluding anything about technique, it is worth checking against how to structure typing practice that the practice itself is shaped to produce a change.

Not every plateau needs solving. Somewhere above the point where typing has stopped being the slow part of your work, further speed buys progressively less, while costing progressively more to obtain — a trade examined directly in is 120 WPM worth chasing.

It is worth being explicit about this because the framing of typing practice tends to assume that more is always better, and for most people it stops being true somewhere in the eighties. If your speed comfortably exceeds what your work asks of you, a plateau is not a problem to be solved — it is the point at which the effort is better spent elsewhere.

Where it is worth continuing is when a specific bottleneck is costing you daily: symbols for programmers, numbers for anyone in data work, sustained accuracy for transcription. Those are targeted problems with targeted drills, and they are worth work in a way that a general pursuit of a higher number often is not.