The most valuable things in a three-day study plan are the two nights in the middle of it.
That sounds like a dodge, or like advice to work less. It is neither. There is a randomised experiment that isolates exactly this, and the size of the effect is larger than almost anything else you could change about how you use the three days.
Why three sessions beat one long one
Stéphanie Mazza and colleagues had 40 people learn word pairs to perfect recall, twice, twelve hours apart. One group learned at nine in the morning and relearned at nine that evening. The other learned at nine in the evening, slept, and relearned at nine the next morning. Identical gap. Identical criterion. The only difference was whether the gap contained a night.
Getting back to perfect recall in the second session took the sleep group an average of 3.05 attempts. It took the wake group 5.80. Nearly twice the work for the same result.
A week later, the sleep group recalled 15.2 of 16 pairs. The wake group managed 11.25, and not a single person in it recalled all sixteen. The authors' summary is the line to build a plan around: "Sleeping after learning is definitely a good strategy, but sleeping between two learning sessions is a better strategy."
Three days gives you two nights. That is the resource. An all-nighter on day three throws away the only structural advantage you have.
Your practice score is going to lie to you
Before the plan, the finding that makes people follow it.
Doug Rohrer and Kelli Taylor had undergraduates practise finding volumes of unfamiliar solids, either blocked by type or all mixed together. During practice, the blocked group was scoring 89 percent. The mixed group was at 60 percent, and would have felt like they were doing badly.
A week later, on the test, the mixed group scored 63 percent and the blocked group scored 20.
Whatever your accuracy is during a study session, it is not a forecast. Practising one topic at a time inflates it, because you never have to work out which method applies. Mixing topics deflates it, and it is the deflated version that predicts the exam.
The same paper found something else worth knowing before you plan your three days: one group did twice as many practice problems as another and scored the same, with a p-value of .8. Past a point, more problems on the same topic buy nothing.
What "cramming doesn't work" actually means
This gets repeated so often and so loosely that it is worth being precise, because the honest version is more useful.
The big meta-analysis on spacing pulled together 839 assessments. Hold total study time equal, and spreading it out beat compressing it at every retention interval they measured. Across the whole set, 47.3 percent against 36.7. In the two-to-seven-day range, which is where you are standing, 45.4 against 31.1.
But the authors add a caveat that most people quoting them leave out, and it deserves quoting in full: "a few hours of cramming would typically involve repeated noncontiguous study of individual bits of information, rather than literal massing as examined in the studies noted." Lab massing means back-to-back repetition. Your all-nighter is not quite that.
So the accurate claim is narrower and still decisive. Massed study does produce a real advantage on an immediate test. Rawson and Kintsch found exactly that, and found the advantage had disappeared by a two-day delay. Your test is not immediate. By the time you sit down, the window in which cramming wins has already closed.
The three days
Day one: find out what you actually cannot do. Take a mixed set of problems covering everything on the test, closed book, no notes, no examples. It will feel bad. That is the point, and it is the same reason the mixed group above looked worse while learning more.
Do the whole set before you check anything. Then mark it.
That ordering is not arbitrary. Andrew Butler and Henry Roediger compared feedback timing and found that checking answers at the end of the session beat checking immediately after each question: 56 percent against 45 percent on a test a week later, with no feedback at all trailing at 31. Marking as you go feels more efficient and is worse.
Day two: work your errors, but re-test everything. This is the day people get wrong, usually by only doing the topics they missed.
The distinction that matters comes from Karpicke and Roediger's work on what to drop. Restudying only the material you got wrong, while continuing to test yourself on all of it, produced the same retention as restudying everything, at roughly 26 percent less total work. But dropping the material you already knew out of the testing more than halved recall, from around 80 percent to the mid-thirties.
So the rule has two halves and both are load-bearing. Re-study only your mistakes. Keep re-testing the whole syllabus.
Day three: one full mixed run under time, then stop. Sit a complete practice set with a clock, in test order, no notes. Mark it at the end. Fix only what it exposes, and keep that session short.
Then stop early and sleep. The second night is the last piece of consolidation you are going to get, and trading it for two more hours of problems is the worst exchange rate available on day three.
Where the solver fits, and where it does not
On day one it does not fit at all. The entire value of that session is finding out what you cannot do unaided, and a solver open on the desk destroys the measurement.
Day two is where it earns its place, on the specific errors you are repairing. How you read a worked solution decides whether it helps you, and there is a method to that: go straight to the step you failed on instead of reading from the top, and name the rule behind every line. Then reproduce the whole thing on a blank page, or work a fresh one through a full step-by-step solution and rebuild it from nothing. That is the version that sticks.
On day three, use it only to confirm answers after your timed run, never during it.
Two honest limitations
The spacing research does not actually cover a three-day window. The shortest retention interval in the main study is seven days, where the best gap tested was one day. Scaling that ratio down to three days points at roughly a one-day gap between sessions, which is what the plan above uses, but that is an extrapolation and I would rather say so than dress it up.
And three days is a repair job, not a substitute for a term. Everything here is about extracting the most from the time you have, which is a different thing from the time you needed.
If day one goes badly in a particular way, worth knowing which problem you have. Being unable to do the problems at your own kitchen table, in silence, is not the same as being able to do them there and freezing in the hall, and those two need opposite fixes. A three-day plan solves the first one. It does very little for the second.
The short version
Three sessions, two nights, mixed problems every time, whole set before marking, restudy only errors while re-testing everything, and finish day three early enough to sleep properly.
Notice how little of that is about working more hours. The students in every study above spent the same total time as their comparison group. What changed was the arrangement, and the arrangement is the only part you still control three days out.