how adaptive typing practice works

what is actually being tracked

Not your words per minute, and not the exercise you finished. The unit is the single keystroke: every time you type a character, the result goes into a rolling, recency-weighted record for that character, so a key you missed constantly last week but have since fixed reads as fixed. Transitions are tracked the same way — the pair of keys, not just each one — because the thing that slows most people down is a specific move between two letters rather than either letter alone.

the six intervals

Each key sits in one of six boxes, and the box decides how long before it should come round again: the same session, one hour, one day, three days, seven days, then fourteen days — after which it counts as settled and drops out of the rotation entirely. Fourteen days is a deliberate ceiling rather than a technical limit. Systems built for vocabulary happily push a well-known card months out, but a typing key that vanished for two months would make the whole loop feel dead, and the point of the practice is that it stays close.

what moves a key up or down

A key moves up a box when it is at least 98% accurate across ten attempts or more. Both halves matter: the accuracy bar is high so a short lucky run cannot graduate a key, and the attempt floor stops three clean presses from counting as evidence. It moves back down when accuracy drops below 85% — low enough that a single typo does not demote anything, because a system that panics at one mistake teaches you to fear mistakes. A key that has settled comes back if its accuracy slips under 90%, which is how a letter you learned in March returns in June when it quietly starts failing again.

why typing does not use the flashcard algorithm

The scheduling algorithms behind most flashcard software — SM-2 and its descendants — need you to grade your own recall after each card: again, hard, good, easy. That grade is the input the whole schedule is tuned from. Typing has no honest version of that question. You cannot usefully ask someone how well they think their ring finger is doing, and stopping to ask would break the one thing typing practice has going for it, which is flow. But typing produces the grade for free: whether you hit the key is the answer, measured hundreds of times a minute. So the boxes are driven by measured accuracy instead of self-assessment, which is simpler and, here, more truthful.

you can see all of it

Adaptive is easy to claim and hard to check, so the state is on screen rather than behind the curtain. The stats page draws your keyboard as a heatmap of per-key accuracy — including the shape of each band, so it reads without colour — and any key on it can be clicked to practise that key immediately. The on-screen hints fade per key as it settles rather than all at once, so the board gets quieter in exactly the places you have stopped needing it. And the daily line is chosen, once a day, from the hand-written pool by which item best exercises the keys you currently owe.

it runs on your device

None of this involves a server. The record of your keystrokes is held in your browser, the scheduling is arithmetic done on the page, and the practice text is assembled locally from banks that ship with the site — there is no model call, no account, and no request that carries what you typed anywhere. It works offline for the same reason. The practical consequence is that your practice history lives on one device unless you export it yourself, which is the trade this site makes on purpose.

what it will not fix

Scheduling decides what you see, not how you sit or where you look. A learner who watches their hands will improve slowly whatever the algorithm does, because they are rehearsing looking; the fix for that is covering the board or moving the chart off the desk, not more practice. Adaptive practice also cannot make progress linear — accuracy on a key wobbles for weeks before it settles, and a bad session usually means a tired hand rather than a lost skill.

questions