Most revision plans are built as calendars of coverage. The evidence on forgetting argues for something different: a schedule of retrieval, weighted by what a paper actually asks and closed by a standard that can be measured.
Hermann Ebbinghaus published the first quantitative account of forgetting in 1885, running the experiments on himself with lists of nonsense syllables and measuring how much relearning effort a list still saved him after an hour, a day, a week. The loss was steep at first and then flattened. The exact figures deserve little weight outside the laboratory: one participant, deliberately meaningless material, nothing to hang it on. What has replicated across a century of better-designed work is the shape. Newly learned material decays fastest in the period immediately after it is learned, which is precisely when most revision plans are not looking at it.
The distinction Robert Bjork and Elizabeth Bjork drew between storage strength and retrieval strength explains why the curve is not a record of deletion. Material that cannot be recalled today is usually still there, and is relearned far faster than it was learned the first time. What has fallen is access. A revision cycle is therefore not a reinstallation of lost content but a series of attempts to regain access at the moment access is becoming difficult. A retrieval that succeeds only after effort strengthens the memory more than an easy one does, and buys a longer interval before the next attempt is needed.
The standard plan is produced by dividing a syllabus by the number of weeks remaining and assigning the pieces to dates. It optimises for the property that is easiest to audit, namely that every chapter has been touched, rather than the property that decides the result: what can be recalled under time pressure on one particular morning. Two weaknesses follow. Such a plan states that a topic will be revisited but never when, relative to the point at which that topic is actually slipping. And because its passes are spaced by calendar rather than by performance, the second pass tends to arrive long after weak material has gone and well before strong material needed it.
Whatever the calendar says, the activity inside the slot is usually rereading, or the copying out of notes. Roediger and Karpicke's 2006 experiments on retrieval practice showed how badly that misleads. On a test taken minutes after study, repeated rereading looked the better method; on a test a week later the ordering reversed decisively, and the group that had practised recalling the material retained a great deal more. Dunlosky and colleagues' 2013 review of learning techniques reached the same verdict across the wider literature, rating practice testing and distributed practice as high in utility and rereading and highlighting as low.
Rereading survives because it feels effective. Work by Koriat and Bjork on foresight bias showed that predictions of future recall are inflated when the answer happens to be visible at the moment the prediction is made. Fluency in the moment gets mistaken for durability, and a plan built on that sensation will retire topics that have not been learned.
A related error is the premature drop. Karpicke and Roediger found in 2008 that removing an item from further retrieval practice once it had been recalled correctly badly damaged retention a week later, while removing the same item from further restudy did comparatively little harm. Revision schedules routinely tick a topic off after one good session, which is close to the manipulation that failed. Blocking is the third habit worth naming: Rohrer and Taylor's studies of mathematics practice found that mixing problem types within a session produced weaker performance during practice and stronger performance at test than working through one type at a time.
Frequency data settles part of the weighting question. The past-paper archive for JEE Main runs from 2002 to 2025 and holds 8,267 authentic questions. Of the topics represented in it, 117 carry enough questions to describe a stable pattern rather than a coincidence: 48 in Physics, 41 in Chemistry, 28 in Mathematics. They are not evenly loaded. Matrices and Determinants accounts for 8.8% of JEE Main Mathematics in recent papers, some 275 questions across the 23 years covered. Coordination Compounds accounts for 8.7% of recent Chemistry, 171 questions. Kinematics in one and two dimensions accounts for 6.6% of recent Physics, 180 questions. A cycle that hands equal time to every unit treats a topic appearing in nearly one Mathematics question in eleven exactly as it treats one that surfaces occasionally. Recurrence is not the only weight that matters, since a candidate's current retrieval strength on a topic matters at least as much, but a cycle that ignores both is only a reading order with dates attached.
The first property is that retrieval is the unit. A session that produces no attempt made with the material out of view has tested nothing and strengthened little. The second is that intervals expand, and are scaled to the date of the examination rather than to a fixed weekly rhythm. Cepeda and colleagues' 2006 meta-analysis of distributed practice, and their later work on the relation between the gap and the retention interval, found that the best spacing grows with how long the material has to survive. A candidate eight months out and a candidate eight days out should not be on the same schedule.
The third property is that the schedule is written by results rather than in advance. Items answered wrongly come back sooner, items answered securely go to the back, and the cycle becomes a queue with an ordering rule instead of a grid drawn up in a single sitting. The fourth is an explicit exit criterion. A coverage plan ends a topic when it has been read. A defensible cycle ends a pass on a topic when measured performance under timed conditions clears a stated bar, and schedules a return when it does not.
Negative marking supplies a floor for that bar. Break-even accuracy on a marking scheme is the penalty divided by the sum of reward and penalty: 25% for UPSC Prelims, 20% for NEET, 20% for JEE Main. The UPSC figure is exactly a blind pick from four options, which makes guessing there neutral in expectation. NEET and JEE break even at 20%, so a blind pick from four options is mildly positive. That arithmetic marks the point below which attempting a question destroys marks. It is a floor, not a revision target: a topic sitting a few points above the break-even line is barely worth attempting and nowhere near revised. Its use to a cycle is that it turns the word revised from a feeling of familiarity into a number that can be argued with.
Every property listed here makes the work harder while it is being done, which is a large part of why such cycles are rare. Bjork's phrase for the pattern, desirable difficulties, captures the trade: conditions that slow acquisition tend to improve retention. Kornell and Bjork's 2008 study of interleaved learning found participants performing better under interleaving and still judging the blocked method to have taught them more. A revision cycle cannot be validated by how it feels while it runs. It has to be defended on evidence and on measurement, because the sensation of learning will argue against it the whole way.
The full set, with progress tracking and five agent perspectives per question, is in the JupiteX app — browse the exam catalogue or browse the Learn library.