The Page Has a Price, Not a Verdict

{{first_name | みんな}}さん、こんにちは ☕

The most quoted number in reading research is 98%. Know 98 words out of every 100 on a page, the story goes, and you can read that page on your own. 📌

In the study that made it famous, 98% was never tested.

The researchers ran 80%, 90%, 95% and 100%, drew a line through the results, and the line happened to pass through 98. Sixty people in total. Fifteen per condition. 👀

Then a much larger study, 661 readers across eight countries, went looking for the cliff everyone had been describing, and did not find one anywhere.

There is a slope instead. Which is better news, because you can stand anywhere on a slope on purpose.

Know someone who keeps abandoning books on page four? Send them this. Share this issue →

Most of us pick what to read by feel. It feels hard, you cannot tell whether that is "too hard" or "hard in the way that is good for you," so you stop at page four and feel bad. 🚪

見当（けんとう）がつかない. No idea. That is the honest name for it, and the feeling is not an instrument.

What is solid underneath: the share of the words you know on a page tracks how much of it you understand. That part is real. ⚡

What is not real is the gate. In that larger study, comprehension never jumped at any percentage. It just kept climbing.

So the number was never permission. It was always a price. 💡

The reference points below are real research. The counting is ours. 🤝

Open the book in the middle. Openings are unrepresentative, usually easier or harder than the rest.

Take three short stretches, not one long one. Two or three lines each, from different parts. One passage lies badly: a repeated name swings the sample.

Count about 100 running words across the three. Every word as it appears, repeats included. In a 文庫本（ぶんこぼん）that is very roughly four or five lines. If counting gets fiddly, count 200 characters instead. 📝

Tally in two columns. Column one, words you do not know. Column two, words you would know instantly if you heard them, where the kanji stopped you.

Add them and read the dial. Zero to 2 per hundred: unassisted reading goes comfortably and you will finish this. Three to 5: the contested zone, where plenty manage and plenty do not, depending on what you already know about the subject. Expect work. Six and above: the honest name for it is study, not reading, so pick your tools accordingly. 🎯

Two things the count misses. Unknowns that repeat get easier, and where they fall matters as much as how many. 🎴

Every study behind the coverage question was run in English, where there is nothing to split. In Japanese a known word comes apart. You can hold the meaning, know the sound, and still not recognise it on the page in kanji. 👀

Nobody seems to have measured that gap. The research has the pieces and nobody has assembled this one, so the figure a vocabulary test gives you is the optimistic one. That reading is ours, not the field's.

Which is why you measure it on yourself. Column one bigger, you need more words. Column two bigger, you already know more Japanese than your reading shows, and the thing in your way is the script. 💪

And no figure is your level. What a text demands runs from around six thousand word families for ordinary conversation to about nine thousand for a novel, and higher again in some academic writing. The number is yours and this book's, never yours alone.

「二列（にれつ）で数（かぞ）える」. Count in two columns. About fifteen minutes.

You will know you did it when: you can say out loud which of your columns is bigger, and name what that changes.

The material was always a variable and almost nobody measures it. A minute of counting turns "this feels hard" into a price, and a price is not a verdict. It is something you decide about. 🍵

Hit reply with your two numbers from one book. Column one, column two. We want to know how often the second is bigger, because nobody has counted it. 👀