Ebbinghaus memorized lists of nonsense syllables and measured how much effort relearning them saved after delays from twenty minutes to a month. In his data, savings dropped steeply at first and then flattened, with roughly a third of the original learning still effective after a day and about a fifth after a month. Murre and Dros repeated the entire procedure in 2015 and reproduced the curve's shape.
The curve is not a fixed law of loss. Its steepness depends on how well the material was learned, how meaningful it is, and what happens afterward. Retrieval practice and spaced review flatten it: each successful recall slows subsequent forgetting, which is the logic spaced repetition systems are built on.
The practical reading of the curve is not that memory is hopeless but that a single exposure is a rental. Most of what you read this week will be inaccessible within days unless something, a note, a highlight resurfaced later, a self-test, gives the memory another event to consolidate around.