1. RNP
  2. Motor Learning
  3. From learning to transfer: building a movement that holds

Motor LearningSeries 1/6

SeriesFrom practice to the field: why a skill transfers or not1 of 6

From learning to transfer: building a movement that holds

Learning and transfer: a successful session proves nothing about what will last. The four conditions for a movement to hold, and how to check it.

LabO RNP·22 min read·reviewed on

A successful session proves nothing about what will last: performance during learning and delayed retention can come apart. A movement still holds three weeks later when it relies on an available input, when the decision itself has been trained, when the variability of the block covers the range of real use, and when the trace has been consolidated between sessions.

Hello to you, movement professional,

End of a technical block, on a Thursday, in a gym where the light falls sideways across a worn floor. A strength and conditioning coach watches his athlete perform, for the last time in the cycle, a single-leg braking step followed by a push-off in the opposite direction. It's clean: the pelvis doesn't collapse, the knee stays aligned, the eyes move before the foot. Six weeks earlier, that same step fell apart one time in two. The coach logs the session as a success, and he's right: what he sees is better than at the start of the block.

Three weeks go by. Game day. Same step, but with an opponent leaning into his shoulder, a ball arriving too early and a slicker surface. The movement is gone. Not degraded, not approximate: absent. The athlete falls back on the pattern he had on day one of the block, as if the six weeks had never happened.

This moment is often misread, and three explanations get passed around: run the block again, use drills closer to the game, he lacks the strength to control that braking. All three may be true, and none of them explains what happened, because they share the same assumption: that a movement executed well at the end of a block is a movement that has been acquired.

It isn't. What strength and conditioning calls transfer covers what the motor learning literature separates into two tests, retention and transfer: the availability of a motor solution at a distance from the time and place where it was built, including under conditions that were never rehearsed. That property can't be read at the end of the block. It can't even be read during the session where it is being built, and that is precisely what makes the job difficult.

An objection comes up straight away: if a movement only holds under game conditions, then all the technical work in the gym is wasted time, and you might as well just let them play. It's a fair point, and it sets the limits for everything that follows: work outside the game only has value if it installs something the game alone would not, and that value must be verifiable at a distance, on a date written down in advance. Technical work whose delayed retention nobody checks is a belief, not a method. Each condition therefore comes with its own failure criterion.

So what exactly holds, and under what conditions?


Learning and transfer: what a successful session proves, and what it never proves

The most robust finding in this field is also the most counterintuitive: performance observed during learning and performance observed in delayed retention do not necessarily move together (Schmidt and Bjork 1992; Soderstrom and Bjork 2015). The conditions that make a session look good are often the ones that degrade long-term retention. Transfer of learning is measured on a delayed test, in a situation that was not trained, and it is the only measure that matters on the field.

Three levels of evidence coexist in this article, and it's better to separate them right away. That acquisition performance and retention performance can dissociate, to the point of moving in opposite directions, is an established fact, replicated in dozens of laboratory protocols across a wide range of motor tasks since the contextual interference effect described by Shea and Morgan in 1979. That the mechanism behind this dissociation is the effort of retrieving a solution rather than its repeated execution is a debated theoretical framework, and there is real tension between schools behind it. That the holding of a movement can be reduced to four conditions read in a fixed order is a LabO RNP field framework, built to support decisions, and never validated as a model.

One appealing framework has to be dismissed here in a single line, because it has polluted this subject for thirty years: the learning styles typology, visual, auditory, kinesthetic. The idea that instructions should be matched to the learner's style does not survive the protocols that have tested it (Pashler et al. 2008). What changes whether a movement holds is what the athlete has to build for themselves during the trial, not the channel through which the instruction arrives.

And this is where the usual counter becomes a trap.


Why the rep counter doesn't measure learning

The reflex on the field is to count. Number of trials, number of sets, number of sessions in the block. The counter is reassuring because it's objective, and it's objective because it measures the one thing that doesn't determine the outcome. Repeating a step a thousand times under constant conditions produces execution that becomes more and more consistent under those exact conditions, and says nothing about what will happen anywhere else.

The reason is mechanical, and it has nothing to do with the athlete's motivation or attention. A repetition under identical conditions lets the athlete replay the solution found on the previous trial without having to rebuild it. The athlete keeps the solution in short-term memory, reapplies it, refines it, and the session visibly improves. But the trace being strengthened is the trace of application, not of construction. Three weeks later, in a situation where the solution can't be replayed as is, there is nothing to apply.

The coach in the gym saw six weeks of continuous progress. His counter said one hundred and fifty clean trials. What the counter didn't say was how many times his athlete had to rebuild the solution instead of recalling it. The answer, in a block with constant conditions, is close to zero.

To understand what needs to be built, you have to look at what actually gets engraved when a movement holds.


The RNP training space, free for life

Get access to RNP training, no strings attached.

Welcome! Introduce yourself

SecureNo cardNo password

By continuing, you agree to our Terms of Use and our Privacy Policy.

What gets engraved when a movement holds, and where

An available movement looks less like a muscle sequence stored somewhere and more like a set of parameters that make it possible to regenerate a solution suited to the situation at hand, and those parameters live in the sensorimotor loop, distributed across identifiable structures.

The cerebellum, cortex and deep nuclei, builds and updates the model that predicts the sensory consequences of a motor command, according to the internal models framework (Wolpert, Miall and Kawato 1998). This predictive motor control is the centerpiece of whether a movement holds: what carries over from one context to another is the ability to predict accurately what this movement will produce, far more than the movement itself. The basal ganglia, and in particular the putamen within the striatum, handle the selection and chaining of the sequence's segments, and they are where the shift happens between a solution that demands attention and a solution that runs in the background (Doyon et al. 2009).

Two signals feed this update. The efference copy, sent in parallel with the motor command, lets the system know what it has just requested before receiving the feedback. Reafference brings back the actual feedback, and it is the gap between the two, more than success, that forms the raw material for updating the model.

The consequence for running a session is unsettling. A trial with no gap between what was predicted and what happened teaches almost nothing, since there is nothing to correct in the model. A block in which everything succeeds is a block in which the model has barely moved.

Still, you need to understand why this engraving happens before the movement, and never during it.


A fast movement has no time to be corrected, so it has to be predicted

On a high-speed braking step, the window in which the movement plays out is shorter than the time needed for sensory information to travel up, be processed, and come back down as a useful correction. Real-time control of a fast movement, in the sense of voluntary correction mid-action, does not exist, and no cue given during the action can make up for it. This is a wiring constraint, not a lack of training.

Two mechanisms cover this gap. The first is feedforward: the command goes out already parameterized for the expected situation, based on the predictive model, without waiting for sensory feedback. The second is anticipatory postural adjustment: before the limb moves, the trunk and stance organize themselves to absorb the perturbation that the movement itself will create. Both are set before the movement starts, and setting them is exactly what a learning block installs or fails to install.

A third route asks nothing of anyone. The spinal reflex arc responds far faster, but in a stereotyped way: its task-dependent modulation exists (Stein and Capaday 1988), but it remains narrow. It guarantees that something happens, never that this something is the right solution for the situation.

In other words, everything that will make the movement hold on game day is already written at the moment it launches. So the question becomes: under what conditions does what is written at launch stay written three weeks later?


The four conditions under which a movement holds

What follows is a field framework, and it should be read as such. No study validates this four-part breakdown, nor the order in which it is presented. It exists because it lets you decide, at the end of a block, which of the four was missing, rather than running the same block again and hoping for better.

Four conditions, then. That the input the movement relies on is available at the time of learning. That what is trained is the decision and not just the output. That the variability encountered during the block covers the range in which the movement will have to serve. And that the trace is consolidated between sessions, which is not something you decide within a session.

They are read in this order, and the order isn't decorative: the second is useless if the first is missing, and the third creates disorder if the second isn't in place. It's the same reading-order principle that governs the whole method, applied this time to durability rather than correction.

Each one is read through a different indicator, starting with the input.


The first condition: a movement only holds on an available input

This is where most technical blocks go astray, and it is read before the first session is written. The proprioceptive chain that carries information about stance and position has several relays, each of which can be tested separately, and each of which can deliver poor information without anything showing in the output. Muscle spindles for stretch length and velocity, Golgi tendon organs for tension, joint receptors and plantar cutaneous receptors for position and pressure: if one of these channels returns blurred information, the system still builds a solution, with whatever it has.

The problem only shows up after the block, because the solution that gets installed contains the compensation. An athlete who learns to brake with poor plantar information learns a braking pattern that does without that information, often by locking up higher in the chain. That movement holds up as long as conditions stay those of the block, and it breaks as soon as the situation calls for fine modulation of the foot contact.

This is the direct application of the read-the-outputs, train-the-inputs logic. RNP places this reading before any decision about content, and it does so for exactly this reason: you cannot engrave cleanly on a medium that renders poorly. On a day-to-day scale, the athlete's state that day modulates the same question, by deciding which level can be demanded this morning.

Failure criterion for this condition: if the input reading never changes the content of the block, it serves no purpose and should be dropped.

The second condition, by contrast, plays out in how each trial is run.


The second condition: exercise the decision, not just the output

The trial has just gone off, imperfect, and this is where the craft tips one way or the other. Fixing the visible output right away, by naming what was wrong, buys immediate quality at the cost of delayed retention. What the position on correcting the movement argues applies here in its temporal version: a correction applied from the outside installs an execution, it does not install the ability to find that execution again without help.

What needs to be exercised is the segment that comes before the movement. The motor decision plays out in a very short window, before the athlete is aware of the movement, and it is what selects the solution from a space of possibilities. A block that hands over the solution in advance removes that selection: it trains the output and leaves the decision idle.

What supports this selection is the representation the system has of its own body. The body schema, that unconscious and constantly rewritten map, is what the decision relies on to set the parameters of the command. A precise map makes it possible to rebuild a solution in a new situation. A coarse map makes it possible to replay a known solution, and nothing else.

Failure criterion: if the athlete cannot produce the movement correctly without being given the cue again before each trial, the condition is not met, whatever the level of the output.


The third condition: the variability encountered must cover the space of use

This is the core of what ecological motor learning has brought to the field, and it is also what gets done worst in practice, because varying is easy and varying well is not.

What carries over from one context to another is a relationship between what the situation offers and what the body can do, rather than a trajectory. That is what affordances are: the action possibilities an environment presents to that particular body. An athlete who learns to read them in a single setting has learned a setting. The right way to vary them is through design rather than instruction. The constraints-led approach means modifying the task, the space or the rule so that the target solution becomes the most natural one, without naming it.

Two safeguards frame this variability, and forgetting them produces costly noise. The first is representativeness: a representative learning design keeps the information sources that drive the decision in the real situation, instead of replacing them with artificial cues (Pinder et al. 2011). A light signal on the floor and an opponent's shoulder do not call on the same reading. The second is the distinction between useful variability and disorder: structured motor noise explores the solution space, unstructured noise just adds error.

A further lever comes into play, more debated than people admit. A focus of attention directed toward the effect of the movement rather than the body segment producing it is associated, across a large body of literature, with better outcomes in both performance and retention (Chua et al. 2021). The effect should be presented for what it is: a more recent meta-analysis that corrects for publication bias estimates that this advantage has been overstated and that its true size remains uncertain (McKay et al. 2024). It is a lever to try and to check on the delayed test, like everything else.

Failure criterion: if the variability introduced degrades the session without changing anything on the delayed test, it was disorder.

The last condition, for its part, is not decided during the session.


The fourth condition: what is not consolidated is not acquired

Between two sessions, the trace that was built does not sit idle. It is replayed and stabilized, and part of this work apparently happens overnight, linked to slow-wave sleep and the spindles that punctuate it (Walker et al. 2002). The size of this overnight gain is debated, with some of the improvements attributed to sleep stemming from protocol biases (Pan and Rickard 2015). That is still enough to take overnight consolidation out of the lifestyle-hygiene aisle: it is part of the moment when what was built becomes available, or fails to.

The practical consequence concerns how the block is scheduled. Learning work placed on a late match evening, followed by a short night, produces a decent session and uncertain engraving. The same work placed elsewhere in the week should produce the same session and better engraving: this is a field hypothesis, which the failure criterion below lets you check. The content is identical; the delayed result may not be.

Failure criterion: if delayed retention is identical regardless of where the learning work sits in the week, this condition carries no weight for this athlete, and you should stop accounting for it in their case.

One domain is supposed to escape this reasoning, and yet it follows it to the letter.


Why strength is no exception

The widespread idea is that strength is a physical quality, so it carries over on its own, unlike a skill. The four conditions say the opposite, and explain why.

In a strength block, contractile capacity improves, and along with it, often mostly, an organization: the intermuscular coordination that decides which muscles contract, in what order and with what relative intensity to produce that force in that position. This organization is a learned motor solution, engraved in a context, on a given input, with a given variability, and it obeys the same retention conditions as a braking step.

That is why a gain measured in the weight room can hold there for a long time and never show up on the field. Add to this the temporal dimension: a transfer condition met at the end of a block is not a transfer condition met three weeks later.


What you can observe in a session, and the test nobody runs

The useful distinction is between the cues available during a session and the cue that only exists at a distance, and it is sharper than people think.

During a session, you do not see retention. You see three things that, in our field reading, inform it indirectly, without any correlation having been measured. How many times the athlete had to rebuild their solution instead of recalling it, which shows in how long they take to get going when a variable changes. Whether they produce the movement without a prior cue, which you test by removing the cue. And whether their solution withstands a change of surface, approach speed or information source, which you test by changing one variable at a time. These three readings tell you whether the athlete is recalling or rebuilding, never to grade them.

At a distance, and only at a distance, you see retention itself. The delayed test is run in a situation that has not been trained, after a delay long enough for the immediate memory of the session to have faded, with no cue reminder and no technical warm-up on the target movement. A test run the next day, in the block's conditions, after a reminder of the cue, only measures the memory of the week.

One reading comes before all of this, at the start of the block, before the first session: the reading of the input for the target movement, on the relay that carries it. It decides whether the block starts by engraving or by making the input available. Without it, you are engraving on a medium you know nothing about.

Two columns are enough to keep track of this over a season: what was worked on, and the date on which its retention will be checked. The second column is the one that changes the craft, because it makes the judgment falsifiable, and because it is written on the day the block starts, not on the day it suits you.

So the most reliable cue arrives after the next decision: you choose the following block without knowing whether the previous one held. Hence how easy it is to fall into the trap that follows.


Correlation is not causation: the trap specific to a learning block

The strength coach in the gym observed clear progress over six weeks and concluded that his block worked. The reasoning is spontaneous and wrong, and it always takes the same form: some content was applied, an improvement followed, so the content produced the improvement. In a learning block, three alternative mechanisms produce exactly the same observation.

The first is the context effect. The end-of-block test is run in the block's conditions, so it measures adaptation to those conditions, not acquisition. The more homogeneous the block, the more flattering this test, and the wider the gap with the match will be.

The second is regression to the mean, combined with the choice of starting point. A technical block almost always starts because something was going wrong, so at a low point in a fluctuating series. Part of the observed gain is simply that series returning to its usual level, and that part does not depend on the content.

The third is concurrent maturation. Over those six weeks, the athlete also played, progressed through their season, changed load and context. The block was not the only thing going on.

The tipping criterion is explicit, and it includes going back. To attribute retention to the content, you need a comparison with alternative content at equal volume, a delayed test in an untrained situation, and a separation that survives removal of the content. If the separation disappears when you stop, it was a form effect, and you have to say so.


What nobody can promise you yet

The four conditions are a way of putting things in order, not a validated model. Nobody can tell you which one weighs most for a given athlete, or in what proportion, or whether there is a threshold below which one missing condition cancels out the other three. The literature supporting the dissociation between acquisition and retention was built on laboratory tasks, often simple, often with non-athlete participants, and carrying it over to a braking step in a match is an extrapolation declared as such. It is solid on the direction of the effect, not on its size.

Nor can anyone give you a specific number of days to wait before testing, for the reasons given above. Giving one here would be making it up.

The falsifiable prediction, on the other hand, is clear. Take two groups with identical content, volume and duration. In the first, the solution is given before each trial and conditions stay constant. In the second, the athlete has to rebuild their solution on every trial and one variable changes with each set. If what this article argues is right, the first group will be better at the end of the block, and the second will be better on a delayed test in an untrained situation. The reasoning is wrong, and we should say so, in any of these cases: the second group's advantage already shows up in technical quality measured at the end of the block, it shows up only on a test repeated under training conditions or in the athletes' self-reported feel, or it is missing on the delayed test.


Frequently asked questions about learning, transfer and whether a movement holds

Should you stop correcting a movement during the session?

No, but you need to change when you do it. A correction given before the trial builds an execution that depends on the cue. Feedback given after a set, and framed around the effect produced rather than the body segment, leaves the rebuilding to the athlete while still guiding their search. The question remains who builds the solution.

How long should you wait before testing whether it holds?

Long enough that the athlete's memory of that week's practice no longer supports the movement. That is all we can honestly say: no published delay covers the range of sporting tasks. The practical rule is to write the test date down at the start of the block and stick to it.

Does this logic apply to a complete beginner?

Only in part, and that is a real limitation. For an athlete who has no available solution at all, the priority is to let one emerge, so more stable conditions make sense at the very beginning. Variability becomes the dominant lever once a solution exists, not before. This is the idea behind the challenge point framework proposed by Guadagnoli and Lee (2004): the right level of difficulty depends on the learner's level.

Does this call cycle planning into question?

No. You still plan the stages and the direction of the cycle in advance. What changes is how you run the trials within each session, plus a delayed check date added alongside each piece of learning content.


What the coach sees on returning to the gym

Go back to that Thursday, the light coming from the side, the clean plant at the end of the block. With the four conditions in hand, that session reads differently. The plant entry was never questioned, so the solution that was built may contain a compensation. The cue was repeated before every trial, so what got practiced was execution, not selection. Conditions stayed constant for six weeks, so the space covered was a single point. And the technical work fell on heavy evenings, so it was consolidated in poor windows. Four gaps, in a session that still looked successful.

A movement that disappears in a match points to a solution that never existed outside the place where it was built. Read where it's dark, switch back on what isn't coming through, and make it hold: that last step is the one training skips most often, because it means waiting before you can be sure you were right.

What remains open is the question of dose. How much rebuilding can an athlete handle per session before exploration turns into pure disorder, and does that limit shift with their level of expertise? We don't have the answer, and as far as we know, the longitudinal follow-up that would provide it does not yet exist in this field.

A movement that never had to be rebuilt was never learned. It was only shown.

👉 I want to understand why, not just what.

RNP is not a medical procedure: it complements health care, it does not replace it.

Sources

Shea JB, Morgan RL. Contextual interference effects on the acquisition, retention, and transfer of a motor skill. Journal of Experimental Psychology: Human Learning and Memory. 1979;5(2):179-187. doi:10.1037/0278-7393.5.2.179

Schmidt RA, Bjork RA. New conceptualizations of practice: common principles in three paradigms suggest new concepts for training. Psychological Science. 1992;3(4):207-218. doi:10.1111/j.1467-9280.1992.tb00029.x

Soderstrom NC, Bjork RA. Learning versus performance: an integrative review. Perspectives on Psychological Science. 2015;10(2):176-199. doi:10.1177/1745691615569000

Pashler H, McDaniel M, Rohrer D, Bjork R. Learning styles: concepts and evidence. Psychological Science in the Public Interest. 2008;9(3):105-119. doi:10.1111/j.1539-6053.2009.01038.x

Wolpert DM, Miall RC, Kawato M. Internal models in the cerebellum. Trends in Cognitive Sciences. 1998;2(9):338-347. doi:10.1016/S1364-6613(98)01221-2

Doyon J, Bellec P, Amsel R, Penhune V, Monchi O, Carrier J, et al. Contributions of the basal ganglia and functionally related brain structures to motor learning. Behavioural Brain Research. 2009;199(1):61-75. doi:10.1016/j.bbr.2008.11.012

Stein RB, Capaday C. The modulation of human reflexes during functional motor tasks. Trends in Neurosciences. 1988;11(7):328-332. doi:10.1016/0166-2236(88)90097-5

Pinder RA, Davids K, Renshaw I, Araújo D. Representative learning design and functionality of research and practice in sport. Journal of Sport and Exercise Psychology. 2011;33(1):146-155. doi:10.1123/jsep.33.1.146

Chua LK, Jimenez-Diaz J, Lewthwaite R, Kim T, Wulf G. Superiority of external attentional focus for motor performance and learning: systematic reviews and meta-analyses. Psychological Bulletin. 2021;147(6):618-645. doi:10.1037/bul0000335

McKay B, Corson AE, Seedu J, De Faveri CS, Hasan H, Arnold K, et al. Reporting bias, not external focus: a robust Bayesian meta-analysis and systematic review of the external focus of attention literature. Psychological Bulletin. 2024;150(11):1347-1362. doi:10.1037/bul0000451

Walker MP, Brakefield T, Morgan A, Hobson JA, Stickgold R. Practice with sleep makes perfect: sleep-dependent motor skill learning. Neuron. 2002;35(1):205-211. doi:10.1016/S0896-6273(02)00746-8

Pan SC, Rickard TC. Sleep and motor learning: is there room for consolidation? Psychological Bulletin. 2015;141(4):812-834. doi:10.1037/bul0000009

Guadagnoli MA, Lee TD. Challenge point: a framework for conceptualizing the effects of various practice conditions in motor learning. Journal of Motor Behavior. 2004;36(2):212-224. doi:10.3200/JMBR.36.2.212-224

The LabO RNP team

€0Your free RNP PathsStart