Successive relearning: mastery that sticks.

Ask any training team what happens to a topic after a learner passes it, and the honest answer is usually: nothing. The pass is recorded, the module locks green, and the system’s relationship with that knowledge ends on its best day. This article is about the protocol that refuses to end it there — and the evidence that the refusal is where durable knowledge actually comes from.

There is a moment in every quiz-based course where an item’s fate is decided. The learner answers it correctly, and the system checks it off. Mastered. The item never returns. Every flashcard app with a “learned” pile, every compliance module with a passing score, every onboarding checklist encodes the same assumption — that one correct recall is a permanent state rather than a perishable event.

The memory literature has been clear about this for decades: a single success is a snapshot, not a property. What the field needed was a protocol that takes the snapshot seriously and re-takes it — plus evidence for how much re-taking is enough. That protocol is successive relearning. Retrieve each item to a criterion of correctness — a set number of correct recalls — in a first session; then relearn it to criterion again in several spaced sessions. It is arguably the most directly usable finding in all of applied memory research.

How much practice is enough?

The founding study asked the question with unusual precision. Learners practiced recall of key-term definitions to varying initial criteria — one, two, three, or more correct recalls. They then completed varying numbers of spaced relearning sessions, with retention measured up to months later. Two results organize everything since (Rawson & Dunlosky, 2011).

First, raising the initial criterion helps, but the returns shrink sharply beyond about three correct recalls. Second, relearning sessions are where durability actually comes from — extra spaced sessions kept paying long after extra same-day successes stopped. The practical recipe — roughly 3× initially, then 3 spaced relearning sessions — has become the literature’s default.

Criterion-level work sharpened the picture. Repeated success within a session mostly buys speed of relearning later; the spaced returns buy retention itself (Vaughn & Rawson, 2011). This echoes a classic finding. Once an item has been recalled correctly, continuing to study it adds almost nothing — but continuing to retrieve it on later occasions adds a great deal (Karpicke & Roediger, 2008).

Efficiency, not just effectiveness

What makes this optimization work unusual for the field is that it counted costs. Most learning-technique studies report only outcomes. The successive-relearning program logged the minutes each schedule consumed, and asked which combination bought the most durable memory per unit of time. That framing produces conclusions a training budget can actually use.

Piling up extra successes in the first session is cheap but nearly worthless for retention — the gains evaporate along the forgetting curve like everything else learned in one sitting. A relearning session, by contrast, is remarkably cheap. Because the material was learned to criterion before, most items come back in one or two attempts. So a return visit costs a fraction of the original session while multiplying its durability. The protocol’s efficiency comes from an asymmetry the intuition never sees — initial learning is expensive and perishable; relearning is cheap and stabilizing (Rawson & Dunlosky, 2011).

This is also why the technique ages so well across a curriculum. The second relearning session is faster than the first, the third faster still, and items that keep surviving effectively pay rent in seconds. A knowledge base maintained this way settles into a long tail of nearly-free upkeep. That is exactly the shape a workforce’s compliance-critical or safety-critical knowledge needs — and exactly what one-shot certification cannot produce.

Out of the lab, into the GPA

The classroom translation was never going to be automatic. Real courses have competing demands, uneven attendance, and students who study however they study — which is what makes the field results worth their space here. The research groups embedded the protocol as a structured activity running alongside normal coursework. Key concepts entered a practice system; students retrieved them to criterion; the system brought each concept back across the weeks before the exam. Other course material was left to students’ usual habits as a within-course comparison.

The results are what separate successive relearning from most laboratory darlings. Embedded in an undergraduate psychology course, successive relearning of course concepts produced exam gains on the order of a letter grade, relative to students’ scores on business-as-usual material. The advantage was still visible on retention tests weeks after the exam (Rawson, Dunlosky & Sciartelli, 2013). In a difficult biopsychology course, the protocol improved scores on a genuinely high-stakes exam (Janes, Dunlosky, Rawson & Jasnow, 2020). These are not d-values from forty undergraduates recalling word pairs. They are grade-distribution shifts in credit-bearing courses.

An item’s life under the protocol

Following a single piece of knowledge through the schedule makes the machinery concrete. Say the item is a claims handler’s threshold rule: which cases must be escalated to a senior adjuster. On Monday, in the initial session, the learner meets the rule, then answers escalation questions until they have produced the correct judgment three times from memory — not in a row necessarily, and with errors corrected along the way. The item leaves Monday mastered in the only sense the protocol recognizes: mastered today.

Thursday, the item returns. The learner misses it once — three days of forgetting have done their work — relearns from the correction, and reaches criterion again in two attempts instead of Monday’s five. The following Wednesday it returns again; this time the first attempt succeeds, and the item’s next appointment moves two weeks out. By the fourth cycle the rule answers itself — the learner reads the case and the escalation judgment simply arrives, the subjective signature of knowledge that has stopped depending on the course that taught it. Total cost after Monday: perhaps four minutes, spread across a month. That is the trade the protocol offers on every item it tracks — minutes of scheduled return visits in exchange for knowledge that behaves like knowledge when a real case, months later, needs it.

Why it works: a package of the two best effects

Successive relearning needs no exotic mechanism, and that is its strength. The most complete review of learning techniques rated exactly two as high-utility across materials, learners, and settings: practice testing and distributed practice (Dunlosky, Rawson, Marsh, Nathan & Willingham, 2013). Successive relearning is those two techniques composed into a single protocol. Retrieval supplies the memory benefit of testing. The spaced returns make each retrieval happen after partial forgetting — when it is effortful enough to strengthen rather than merely confirm.

The spacing meta-analysis’s central lesson — longer gaps buy longer retention (Cepeda, Pashler, Vul, Wixted & Rohrer, 2006) — sets the relearning intervals. The criterion rule guarantees no item exits a session below mastery.

Bahrick saw the deep structure early. In his studies of knowledge maintained over years, what predicted very-long-term retention was not how well something was first learned. It was how many spaced relearning cycles it had survived (Bahrick, 1979).

Designing the schedule

Turning the protocol into a running system involves a handful of decisions, and the literature has something to say about each. The criterion: “correct” should mean produced from memory — typed, spoken, or selected against genuinely competitive alternatives — not merely recognized as familiar. And the default of three correct recalls in the first session sits at the knee of the diminishing-returns curve. Raise it only for knowledge whose failure is expensive (Rawson & Dunlosky, 2011).

The gaps: the spacing literature’s central trade-off applies. Longer intervals cost more failures at the start of each relearning session, and buy more durability after it. The best gap scales with how long the knowledge must last (Cepeda et al., 2006). Expanding schedules, where each survival earns a longer reprieve, fit the protocol naturally.

Failure handling is where naive implementations go wrong. An item missed during a relearning session has not fallen back to zero — Bahrick’s point — but it has shown that its current interval outran its stability. The correct response is to relearn it to criterion within the session and shorten its next gap. Do not exile it back to the start of the queue, which wastes the accumulated cycles. Do not wave it through, which defeats the criterion.

And scope deserves an explicit decision. The protocol earns its cost on knowledge that must be produced reliably months from now — definitions, procedures, thresholds, contraindications — not on every fact a course happens to mention. Choosing what enters the relearning pool is a curriculum judgment the schedule cannot make for you.

Why nobody does it by hand

If the protocol is this good and this simple, why is it rare? Because its bookkeeping is inhuman. Running successive relearning by hand means tracking, for every item, whether this learner has reached criterion, how many relearning sessions the item has survived, and when it is next due. Multiply that by hundreds of items, interleaved across topics.

Left to their own devices, learners do the opposite of the protocol. They drop items after the first success — which feels efficient and measurably damages retention (Kornell & Bjork, 2008) — and they mass their practice the night before the deadline. Self-regulated study defaults to exactly the schedule the evidence says to avoid (Kornell, 2009). Criterion tracking is a job for software.

The standard objections, answered

“We don’t have time for return visits.” This objection measures the protocol against a fantasy. The real comparison is against retraining: companies that skip upkeep do not escape the cost of forgetting. They pay it later — in refresher courses that restart from zero, in errors on the floor, and in audits failed on material everyone covered last year. Four minutes per item across a month, set against relearning the item from scratch next year, is not an added cost. It is the cheap branch of a fork the company is already on.

“Our people are professionals — they’ll retain what matters.” The forgetting curve was replicated on motivated adults. The flashcard-dropping studies were run on university students with grades at stake. Expertise in a domain slows forgetting of its core schemas. It does nothing for the peripheral, arbitrary, and recently changed material that compliance and product knowledge mostly consist of (Kornell & Bjork, 2008). Professionalism changes what people want to retain. It does not change the decay function.

“Won’t constant re-testing annoy them?” The protocol’s arithmetic runs the other way. Relearning sessions only include items that are due, and due items shrink as stability grows. So a mature schedule touches each person for a few minutes a week — less interruption than one scheduled refresher webinar, delivered in slivers. What annoys learners, the survey data in the classroom studies suggests, is not brief retrieval. It is discovering at exam time that a semester’s “completed” material has evaporated (Rawson, Dunlosky & Sciartelli, 2013).

What the evidence doesn’t show

• The evidence base is strongest for definable knowledge components. Key terms, concepts, procedures, facts — material where “correct” is checkable. Whether the protocol transfers wholesale to complex judgment and open-ended skills is plausible but less directly tested.
• It is not free. The protocol costs real minutes across real weeks. The optimization literature exists precisely to spend those minutes well — 3×-then-3-sessions is a diminishing-returns knee, not a magic number, and the right settings shift with the stakes and lifespan of the knowledge.
• Criterion ≠ understanding. Retrieving a definition to criterion guarantees the definition, not the ability to apply it. Successive relearning secures the knowledge floor; application-level practice still has to be designed on top of it.
• Classroom trials share a lab lineage. Much of the classroom evidence comes from the research groups that developed the technique, in psychology courses. Independent replications across domains are accumulating, but the base is younger than, say, the spacing literature.

Where the evidence stops

1. 1The evidence base is strongest for definable knowledge components
2. 2It is not free
3. 3Criterion ≠ understanding
4. 4Classroom trials share a lab lineage

© 2026 FUTURE PROOF™

What this means for practice

Abolish the “learned” pile. Define mastery as a criterion — answered correctly, from memory, on separate occasions — and make re-demonstration the default rather than the exception. Space the relearning sessions at growing intervals. Let items graduate only after surviving several, and shorten the leash on anything that stumbles. Decide on purpose which knowledge earns a place in the relearning pool. The protocol’s cost is real, and its value is concentrated where reliable recall months later actually matters.

Then budget for it honestly, and defend the budget with the arithmetic the literature provides. Return visits are cheap; they get cheaper with each cycle; and they are the only line item that turns training from an event into a durable capability. A curriculum that gives zero minutes to relearning has decided, implicitly, that forgetting will not happen. It will — on the standard curve, at the standard rate, starting the afternoon the course ends. The companies that act on this are not the ones that spend more on training. They are the ones that spend a modest fraction of the same budget after the applause.