He measured it on himself.

The nonsense was the point

Ebbinghaus needed material with no meaning attached, because meaning is a shortcut. Remembering a poem tells you about the poem as much as about memory. So he built lists of syllables like ZOF, KEL and TID, memorised them to perfect recall, then waited, then measured how much survived.

What he found was not a straight line down. Forgetting is steepest immediately: most of what goes, goes in the first day. After that the curve flattens into a long tail, and whatever is left is oddly durable.

Why review beats rereading

The second curve is the useful one. Each review does not simply put retention back to full. It changes the slope of what follows, so the next fall is gentler than the last. That is why the reviews on the chart get further apart and still hold the line: you are not topping up a leaking bucket, you are making a slower bucket.

This is the entire idea behind spaced repetition, and every scheduling algorithm from SM-2 through to FSRS is essentially an argument about one question: how late can the next review be left without losing it? Too early wastes the effort. Too late and you are relearning from scratch. The sweet spot sits just before the drop.

The counterintuitive part is that the difficulty is doing the work. Rereading feels productive and is nearly useless, because recognising something is not the same as retrieving it. The effort of dragging something back out of your head unaided is what strengthens it. If recall feels easy, you probably left too little time.

The connection to everything else here

Ebbinghaus found a system that forgets unless something intervenes on a schedule. Hold that thought going into the next one, because your computer solves precisely the same problem, in hardware, about fifteen times a second, and it has no choice about it either.