The Forgetting Curve
Memory runs in three acts — encoding, storage, retrieval — and what we keep decays along Ebbinghaus's steep-then-slow curve while quietly reshaping itself rather than replaying intact. · 12 min
You did not lose your keys because your memory is bad. You lost them because remembering is not one thing but three, and any of the three can fail. A memory has to get in, it has to stay, and it has to come back out on demand. Along the way it fades on a schedule you can predict — and, less comfortably, it edits itself each time you use it. By the end of this folio you will know why cramming fails, why a confident memory can still be wrong, and what actually makes something stick.
Guess before you learn
In the 1880s Hermann Ebbinghaus memorised long lists of nonsense syllables until he knew them perfectly, then tested himself later with no review. Suppose you memorise such a list flawlessly today. With no review at all, about what percentage will you still be able to recall one day later?
Only about a third. Ebbinghaus found that most of what you lose, you lose fast — much of it within the first hours — and then the loss slows to a crawl. If you guessed much higher, that is the natural assumption: memory feels sturdier than it is, right up until you reach for something and it is gone.
9–12
3–5
Remembering has three jobs: put it in (encode), keep it (store), and get it back (retrieve). If any job fails, you forget.
Most forgetting happens fast, in the first day, then slows down. And each time you remember something, you rebuild it — so it can change a little without your knowing.
6–8
Memory works in three stages. Encoding gets information in, storage holds it over time, and retrieval brings it back out. A failure at any stage looks the same from outside: you cannot recall.
What you store does not last evenly. It drops sharply at first, then levels off — the forgetting curve. And retrieval is reconstruction: you rebuild the memory from pieces, so it can shift over time rather than replay unchanged.
9–12
Treat memory as three linked acts. Encoding transforms an experience into a storable trace; storage maintains it; retrieval reactivates it. Because the acts are separable, a memory can be well stored yet inaccessible — the word on the tip of your tongue is a retrieval failure, not a storage one.
Retention over time follows Ebbinghaus's curve: a steep early drop, then a long shallow tail. And retrieval is not playback but reconstruction — the brain rebuilds the memory from fragments and expectations, which is efficient, but leaves every remembered event open to quiet revision.
K–2
A memory has three steps. It goes in. It stays for a while. Then you pull it back out when you need it.
Some memories slip away, mostly soon after you learn them. And when you remember, you build the memory again — you do not play it back like a video.
Undergrad
The three-stage framing maps onto distinct failure modes: encoding failures (the information never formed a durable trace), storage decay and interference (traces degrade or compete), and retrieval failures (intact traces without an effective cue). Encoding specificity predicts that recall is best when retrieval cues reinstate the encoding context.
Reconstruction is not a bug but the default. Bartlett's War of the Ghosts showed recall bending toward the rememberer's schema, and Loftus's misinformation work showed post-event suggestion rewriting the record. Each retrieval can reconsolidate a trace in altered form, so a vividly confident memory carries no guarantee of fidelity.
Postgrad
Systems consolidation and synaptic reconsolidation complicate any static storage metaphor: a reactivated trace becomes labile and must be restabilised, opening a window in which it can be updated or distorted. The forgetting function is better modelled as retention following a power law over log-time than as simple exponential decay, with spacing and retrieval practice reshaping the curve.
On the constructive-memory framework, the same machinery that recombines stored elements to imagine the future also produces predictable memory errors — gist intrusions, misattributions, imagination inflation. Fallibility is thus the cost of a flexible, generative system, not evidence of a broken archive. Memory is for using the past, not for faithfully storing it.
the three stages
Encoding (getting information in), storage (keeping it over time), and retrieval (getting it back out). Forgetting is a failure at one of the three.
Note
The flat tail is the good news. Because loss slows so sharply, a single well-timed review catches the memory before the cliff — the whole principle behind the Fading Ink, the review-what's-fading tool this college is built around.
The curve is only half the story. The other half is stranger: even the memories you keep do not stay fixed. Retrieval rebuilds them, and the rebuild can be edited. In 1974 Elizabeth Loftus and John Palmer showed people a film of a car crash, then asked how fast the cars were going. They changed one word. People asked how fast the cars smashed into each other gave higher speed estimates than those asked how fast they hit — and a week later, the smashed group was more likely to "remember" broken glass that was never in the film. The question had rewritten the memory.
Why is this true?
If memory rebuilds itself each time, why does it feel like a faithful recording?
Because the reconstruction is seamless and fast, and it comes wrapped in confidence. You never feel the rebuilding happen, so the finished memory arrives feeling whole and original — the same reason you never notice your brain filling the eye's blind spot.
So memory is neither a vault nor a video. It is a three-stage process — in, held, out — that loses most of what it will lose in the first day and then holds the rest stubbornly, and that rebuilds every memory it hands back, sometimes editing as it goes. This is not a design flaw. A system that recombines the past is also the one that lets you plan a future you have never seen. The cost of that flexibility is that you must hold even your surest memories a little loosely, and let a well-timed review, not a single cram, do the keeping.
Practice — new ink and old, interleaved
1.From an earlier folio: prolonged stress floods the body with cortisol. Given that chronic cortisol can affect memory systems, what would you expect in someone under relentless, unresolved stress?
2.Given the forgetting curve, when is the best time to review something you want to keep?
3.From an earlier folio: the lesion method uses injuries that nature, not the researcher, assigns. Why does that make it weaker than a true experiment?
4.From an earlier folio: the lesion method links a region to a job by what is lost when it is damaged. Some patients, after damage to a deep brain region, can no longer form new long-term memories. What does the lesion method conclude?
5.Gage's memory, speech, and movement were spared, but his planning and restraint were not. What does this pattern suggest about the frontal lobe?
6.Name the three stages of memory and say which one fails in a tip-of-the-tongue moment.
Encoding, storage, and retrieval; a tip-of-the-tongue moment is a retrieval failure — the memory is stored but momentarily inaccessible.
How close were you? Grade yourself honestly — it sets your review date.
7.Which chemical drives the fast, within-a-second part of the response, and where is it released?
8.From an earlier folio: both perception and memory are described as "constructed." What do the two ideas share?
9.In one sentence, explain what the word "reconstruction" means when applied to memory.
10.What is the best description of inattentional blindness, as shown by the gorilla study?