What forgetting looks like
Learn a list of unrelated items and test yourself at intervals, and the retention curve drops steeply at first and then flattens. The steep part is the first day or two. This shape has held up across more than a century of replication with different materials, though the exact numbers depend enormously on what the material is — meaningful, connected content that hooks into what you already know decays far more slowly than arbitrary pairings, which is one reason vocabulary in a language you already half-speak sticks better than vocabulary in one you do not.
The practical consequence is about timing rather than quantity. A review placed while the memory is still strong does very little, because there is nothing to reconstruct. A review placed after it is entirely gone is not a review, it is relearning from scratch. The productive window is in between, and it moves outward each time you hit it.
Why spacing works, stated bluntly
The strengthening comes from the difficulty of the retrieval, not from the number of exposures. This is the single idea that determines every scheduling decision below, with an unintuitive implication: an easy review is a wasted one. If you can produce the answer instantly with no sense of searching, that repetition contributed little, and the interval before it was too short.
Which means the goal of a review schedule is not to prevent forgetting. It is to catch each item at the edge of forgetting, over and over, pushing the edge further out each time. Systems that do this well feel slightly uncomfortable to use, because a well-tuned deck produces a meaningful failure rate — commonly somewhere in the range of one item in ten to one in seven. A deck you get entirely right is a deck whose intervals are too short.
Choosing intervals
There is no schedule that is correct for all material. What is reasonably well supported is the shape: short at first, expanding after each success. The tables below are starting points to adjust, not findings.
| Review | Gap after the previous one | What you are doing |
|---|---|---|
| 1st | Same day, a few hours later | Closed-book recall of everything, not a reread |
| 2nd | 2 to 3 days | Recall again; note which items you had to hunt for |
| 3rd | About a week | Only the items that were slow or wrong last time get full attention |
| 4th | 2 to 3 weeks | Spot-check the strong items, work the weak ones |
| 5th and after | Roughly double each time — a month, then two | Maintenance. Cost per item is now trivial. |
A rough heuristic that circulates is to set the first interval at something like ten to twenty percent of the time until you need the material, then expand. It is a reasonable place to start and it should not be treated as precise; the optimal spacing depends on the retention interval, the material and the learner, and the studies disagree about the details. What all of them agree on is that spacing beats massing, so getting the principle right matters more than getting the numbers exact.
When the deadline is short, the whole scheme compresses and some of it becomes impossible.
| Time to the exam | Workable schedule | Realistic expectation |
|---|---|---|
| 6 weeks or more | Full expanding schedule, five or more passes | Material is still there months later |
| 3 weeks | Day 0, day 2, day 6, day 13, day 19 | Solid through the exam, decays over the following term |
| 1 week | Day 0, day 2, day 5, night before | Enough for the exam, not much beyond it |
| 2 days | Spacing is barely available; two passes with a night between | Short-term recall only. Prioritize by weight on the exam. |
| Tonight | No spacing possible | Cram the highest-yield material and sleep. See the honest arithmetic in exam prep. |
Laying those dates out by hand across five subjects gets tedious fast; the spaced review planner generates the calendar, and word memorize planner handles the case where you have a fixed item count and a fixed deadline and need a daily quota.
Lapses
An item you get wrong after a three-week gap has not earned a two-week gap next time. The behavior that works is to drop it back close to the beginning of the ladder — to a day or two — and let it climb again. The instinct to nudge it slightly shorter and carry on is how a deck accumulates a permanent population of items you fail every single cycle.
If a specific item lapses three or four times no matter how you schedule it, the problem is usually the card, not the schedule. The common causes are that it contains more than one fact, that it is cued by wording you will never encounter outside the card, or that you never actually understood the underlying idea and have been trying to memorize the shape of a sentence. The last one is not fixable by repetition. Go back to the source, understand it, then rebuild the card.
Making cards that work
| Rule | Why |
|---|---|
| One fact per card | A card with four facts is scored as one, so you cannot tell which part failed, and the schedule ends up wrong for all four. |
| Cue it the way you will need it | If the exam gives you a symptom and asks for a cause, the card should not go cause to symptom. Retrieval practice is specific to the direction you practice. |
| Avoid answers that are lists | Partial recall of a seven-item list has no clean pass or fail. Split it, or turn it into a cloze with one blank. |
| Do not build cards for things you do not understand yet | Repetition on an un-understood item produces a memorized string with no anchor, and it lapses forever. |
| Include a source or a page number on the back | When the card is wrong — and some will be — you want to find where it came from without re-reading a chapter. |
| Prune | An item you will never need is a permanent tax on every future session. Deleting cards is allowed and underused. |
Paper works. The Leitner box — a row of physical dividers where a correct card moves one compartment forward and a wrong card goes back to the front — is a mechanical implementation of everything above and requires no software at all. The flashcard maker will lay out a printable set if you want the paper version without the cutting.
Mnemonics, and where they stop earning
Deliberate memory techniques are real and they are narrower than their reputation. A method of loci — placing items along a familiar route through a building you know — reliably improves recall of ordered arbitrary lists. Keyword techniques, where a foreign word is linked through a same-sounding word in your language to a vivid image, measurably speed initial vocabulary acquisition. Acronyms and rhymes hold short fixed sequences.
What none of them do is make you understand anything. A mnemonic converts an arbitrary list into a retrievable one; it does not help you reason with the contents, and time spent constructing elaborate imagery for material that has an internal logic is time taken from learning the logic. The rule of thumb: use a mnemonic where the ordering genuinely is arbitrary — cranial nerves, resistor color codes, the order of operations, the sequence of steps in a procedure — and skip it where the material has structure you could learn instead.
They also fade. A memory palace holds for as long as you keep walking it, and stops holding when you do not. Mnemonics are a way into the material, not a substitute for spacing.
Sleep is part of the schedule
Consolidation continues after study, and a large share of it appears to happen during sleep. This is why a review before bed and a check the next morning is a sensible pairing, and why a schedule that runs on five hours of sleep for a fortnight underperforms an identical schedule that does not. The practical version is unglamorous: protect the sleep and the same number of review minutes buys more. Sleep hygiene has the specifics, and the sleep cycle calculator is a blunt but useful way to pick a bedtime that does not have you waking mid-cycle.
One last habit, worth more than any interval tuning. Once a week, open the deck and read your own failure log rather than your card count. Cards reviewed is a measure of diligence. Items that failed twice in a row is a measure of what you do not know, and it is the only column that tells you where tomorrow should go.
Questions people ask
How many cards should I review a day?
The number is set by your deck and your intervals rather than chosen. If you are adding 20 new items a day on an expanding schedule, expect daily reviews to settle somewhere between roughly 100 and 200 items within a couple of months, and to keep growing while you keep adding. The lever people reach for is longer intervals; the lever that works better is adding fewer new items and deleting ones you do not need. A deck that has grown past the time you actually have gets abandoned entirely, which is worse than a smaller deck you keep up with.
What do I do with a card I keep getting wrong?
Send it back to a one or two day interval rather than shortening it slightly. If it still fails after three or four cycles, stop blaming the schedule and look at the card: it probably contains more than one fact, or it is cued by phrasing you will never see outside the card, or you never understood the idea and are memorizing a sentence shape. The first two are fixed by rewriting. The third is fixed by going back to the source material, which repetition cannot do for you.
Is there a correct spacing formula?
No. The consistent finding is that spaced beats massed and that expanding intervals work well; the optimal specific gaps depend on how long you need to retain the material, what the material is, and who is learning it, and the studies disagree on the details. A workable default is same day, then 2 to 3 days, then a week, then two to three weeks, then roughly doubling. Adjust by outcome: if you are getting nearly everything right, lengthen the gaps, because an easy review is doing almost nothing.
Do memory palaces and mnemonics actually work?
For arbitrary ordered material, yes, and the effect is not subtle — competitive memorizers use these techniques because they work. The limit is what they produce. A mnemonic gives you retrievable access to a list; it does not give you understanding, and it does not survive without maintenance any better than anything else. Use them for genuinely arbitrary sequences, keep using spacing on top, and do not spend an hour building imagery for material that has an internal logic you could have learned in that hour.
Is it worth reviewing right before sleep?
It is a reasonable habit, and the reason is consolidation: a meaningful part of memory processing appears to happen during sleep, so material studied close to bed gets processed rather than sitting through a long waking day of interference first. Treat it as a small edge rather than a technique in itself. The much larger effect runs the other direction — cutting sleep to add study hours reliably costs more than the extra hours return, both in what gets consolidated and in how you function the next day.