Spaced repetition: why review on day 4, not tomorrow
Elman Huseynov · 2026-09-01 ·
Learn a pattern today and review it on day 4, not tomorrow. Then day 11. Then day 25.
A review that arrives too early is close to free. It also teaches close to nothing.
The memory was still sitting there. Nothing had to be rebuilt.
The curve underneath this
Ebbinghaus measured his own memory in 1885. He drew the forgetting curve.
Retention falls fast at first, then flattens.
Every successful review resets the fall and makes the next one shallower.
The best review timing is later than feels safe.
Where the numbers come from
Cepeda and colleagues ran the largest version of this test in 2008.
Cepeda and colleagues varied the gap between study and review. Then they tested people much later.
The whole body of work goes under the spacing effect.
The best gap scaled with how long the memory had to last. Roughly 10 to 20 percent of the target interval.
Interviewing in 3 months needs 9 to 18 days between reviews.
Interviewing in 3 weeks needs 2 to 4 days between reviews.
Reviewing binary search every evening for a month wastes the month.
The ladder
1, 3, 7, 16, 35 days.
Recall a pattern correctly and it moves up a rung. Fail it and it drops to day 1.
The Leitner system moves patterns up or down by rung. One property matters.
Your weak patterns come back constantly. Your strong ones get out of the way.
The first rung is short on purpose. A pattern learned an hour ago is fragile.
One recall the next day makes the longer gaps survivable.
The mistake that hides all of this
Most preparation is organised by topic. A week on trees. A week on graphs. A week on dynamic programming.
Every review inside that week happens while the material is still warm. So every review is easy, and easy means cheap.
At the end you have finished eight topics. You can reliably do the last one.
Mixing patterns in one session fixes it.
Mixing patterns feels much worse. Now you have to work out which pattern this is before you start.
That step is the one an interview tests.
What to do this week
Keep one list of patterns with a date next to each.
Each day, take the ones that are due. For each, close everything and rebuild the core from memory.
Correct, push the date out. Wrong, back to tomorrow.
Ten minutes a day holds a lot more than a three-hour Sunday.
What you do inside those ten minutes is active recall, not rereading.
Where the schedule sits in a week: how to prepare.
Where this runs
AlgoPath schedules this for you. Recognition attempts feed a Leitner scheduler.
The patterns you keep missing keep arriving. The solid ones back off.
You do not keep the list. The track does.
The first 33 steps are free.