Sleep Does Not Repair the Previous Day's Actions, It Reinterprets Them

Hello to you, the movement professional,

Sleep is not the recovery of the sensorimotor loop, it is its processing stage. During the night, sleep spindles focus on the cortical regions used during the day and engrave motor skills. A sensorimotor correction not consolidated by a quality night does not hold up to retesting at 48 hours.

This is the position we defend. The rest of this article serves to support it, and especially to show its limits, because it's the part that no one publishes.

What the field says about sleep, and why it doesn't help you

In the world of performance, sleep occupies a strange place. Everyone agrees that it is fundamental, and almost no one makes it operational. You're told you need eight hours. You're sold blackout curtains, weighted blankets, apps that rate your nights out of a hundred. Then we move on to the real subject, which would be the session.

This discourse is accurate, and it gives you no grip. The sleep duration of a person you accompany depends on their work, their children, their commute, their life. You can't move it. An instruction you can't move always ends up as a closing sentence of the session.

There is a deeper reason. This discourse treats sleep as a state of restoration, just like tissue repair. Yet what happens at night in the nervous system is nothing like healing. It's a processing, in the informational sense: the system sorts what it keeps and redistributes the rest to other regions.

As long as you think recovery, you categorize sleep in the same box as nutrition and stretching. The day you think processing, it changes category: it joins the assessment. And it joins, more precisely, the reading of [the sensorimotor loop](/fr/blog/boucle-sensori-motrice), this circuit that goes from sensory input to motor output and decides the movement before the person is aware of it.

Sleep doesn't work everywhere, it works where you have worked

The result that changes everything came from a nap study with high-density electroencephalography, meaning a dense grid of sensors that allows mapping activity by cortical region rather than globally.

Right-handed subjects learn a motor sequence with the right hand, thus with the left Motor Cortex. Then they take a nap, during which their sleep is recorded. Result: the activity of sleep spindles, these bursts of rapid waves around twelve to fifteen hertz that mark stage 2, increases locally on the left sensorimotor cortex. On the right, the one of the hand that did nothing, nothing changes. The lateralization of sleep physiology exactly follows the lateralization of learning (Nishida and Walker, 2007).

So sleep does not uniformly water the brain. It seeks out the circuits that have worked and works on them, like someone who, instead of ironing everywhere, first asks where it sticks.

This is the tipping point for your practice. If sleep targeted indiscriminately, it would remain a contextual variable, a kind of weather. Since it targets, it becomes the direct continuation of what you did in the session. What you engaged in the person in the morning is what their night will take up again.

And this processing is not distributed uniformly over time either. Stage 2 spindles are rare at the beginning of the night and peak in the last quarter (Fogel and Smith, 2011). In other words, the densest window for motor consolidation is precisely the one most easily sacrificed, the one of waking up two hours early for the morning session. Cutting the end of the night does not randomly remove a bit of sleep. It preferentially removes the part that engraves the movement.

Falling asleep is not turning on sleep, it's accepting to turn off wakefulness

The mechanism requires precision because the subject is saturated with shortcuts.

Sleep does not activate like a light. It appears when the wakefulness system deactivates. The main actor of this deactivation is a small nucleus in the brainstem, the locus coeruleus, which distributes norepinephrine to the entire cortex. In wakefulness, it fires around two hertz, and much more in a threat situation. In slow sleep, it slows to about one hertz. In REM sleep, it turns off completely.

This extinction commands everything else. In the presence of norepinephrine, a synapse can only do one thing: strengthen. Weakening and erasing, on the other hand, require norepinephrine silence. This is the thesis defended by Gina Poe: without this silence, the hippocampus does not empty, and a system that erases nothing eventually can no longer encode (Poe, 2017).

Here, we must explicitly dismiss an appealing framework. You will often be told that this shift is explained by the polyvagal theory. We do not use it. The mechanism documented at the brainstem level is noradrenergic, and it can be very well described without borrowing from a contested model. Dismissing a weak framework is a matter of work hygiene, not caution.

Honesty about the level of evidence: Poe's thesis is solid and defended, it is not unanimous. A competing model, the synaptic homeostasis of Tononi and Cirelli, explains the same sorting by a global downregulation of all synapses, without invoking norepinephrine. Both stand up. We give you both.

The variable you can measure without equipment

Now, let's cross the two plans. If falling asleep consists of letting wakefulness turn off, then the time it takes a system to fall asleep measures something precise: its ability to stop giving importance to what happens to it from outside.

This ability has a functional name, sensory gain control, and it is exactly the operation that the nervous system must know how to execute to accept a recalibration.

Look at what you do in a session. You change an input, a support, a gaze orientation, a cervical afference. You wait for the system to revise its map, its body schema, and produce a different output. This revision also affects the efference copy, this prediction that the system emits at the same time as the command to know in advance what it should feel. It is what makes the motor decision different in the next movement. For this, it must accept to deprioritize the version it held as true. A system that remains alert, on the other hand, locks its version: it continues to treat each input as a potentially dangerous signal, and it revises nothing.

Sleep latency measures this speed at which the system accepts to lower its guard, and nothing else. Unlike duration, you can note it with one question, without any device, with any person you accompany.

What our data shows, and what it doesn't show

Before the first figure, the limits, because they condition the reading of everything that follows.

Our data is declarative. There is no actimetry, no sleep recording, no follow-up of [heart rate variability](/fr/blog/systeme-nerveux-autonome/hrv-readiness). There is no control group, which means that the expectation effect is not separable from the protocol effect: it has been known since Draganich and Erdal (2014) that the mere belief of having slept well modifies the next day's performance. The protocol is finally composite, it combines [slow breathing](/fr/blog/systeme-nerveux-autonome/respiration-lente), reflex micro-mobility, and sensory refocusing, and nothing allows attributing the result to one of these components rather than the others. Thirteen days, finally, is a short window.

That said, the declarative nature calls for a perimeter precision rather than an excuse. Subjective well-being measures outperform objective measures for tracking training response (Saw, Main, and Gastin, 2016). And Gina Poe's lab has shown that different brain regions can be in different sleep stages at the same time, the cortex sleeping while deep structures are still awake. Fifty years of sleep science rest on a surface recording that only sees the cortex. The person who tells you they slept poorly may be reporting a state that the device does not measure.

Here are the eight indicators tracked in twenty-eight participants over thirteen days, NIT Sleep protocol of LabO RNP, paired comparison by Wilcoxon test. The two non-significant ones are in the table, in their place.

Estimated effect sizes between 0.45 and 0.70.

Six lines move, and yet the result that matters lies in the gap between the first and the penultimate. The protocol did not make people sleep longer. It halved the time it took them to turn off. What moved is the axis of wakefulness.

Why a Short Latency is Not Automatically Good News

We must immediately set a safeguard, otherwise this article becomes exactly what it criticizes.

In sleep medicine, sleep onset latency is measured to assess sleepiness. The alert threshold is low: a latency of less than five or eight minutes is a classic marker of sleep debt, not of a calm system. Someone who collapses in three minutes hasn't calmed down; they are exhausted.

Latency is thus read as an inverted U-curve, never as a score to minimize. The group's endpoint, a little over nine minutes, is in the healthy zone, above the suspicion threshold. If the same protocol had brought the group to three minutes, we would have had a problem, not a result.

Remember this reading rule, it is more useful than the number: a latency extending beyond twenty minutes signals a system that cannot turn off its wakefulness, a latency dropping below five minutes signals a system in debt. Both deserve a conversation. It's the middle interval that is comfortable.

What It Changes on Your Test-Retest

The practical consequence lies in a reading rule, and it's the only part of this article we ask you to apply.

You work on an input, you measure before, you measure after, and you get a change, say on a Romberg test or any outcome you can objectively assess. Three days later, you recheck and it's back to how it was. The default reading is that the correction didn't hold.

We argue that there are two hypotheses, not one. Either the correction didn't take, or the night didn't happen. Until you've ruled out the latter, the retest is uninterpretable. And ruling it out costs one question in your anamnesis: how many minutes does it take this person to fall asleep in recent weeks. Beyond twenty minutes, you document, you don't conclude.

It's a modest shift and it needs to be properly bounded. It concerns the calibration of a sensory input, which is what Neuro-Postural Reprogramming does when it reads and then reignites a sensor. It does not concern the acquisition of a sports skill in a situation, for which the framework of ecological motor learning does very well without this hypothesis. Two different objects, two different frameworks, and we do not mix them.

The Hypothesis We Do Not Present as a Fact

There is an extension that interests us, and we present it for what it is: a working hypothesis of the LabO RNP, not a demonstrated mechanism.

The measured fact first. About sixty-eight percent of elite French footballers exhibit a non-integrated Asymmetric Tonic Neck Reflex (Bastiere, 2024). In a performing adult, a residual reflex is read as a locked resource rather than as a problem. Residual primitive reflexes signal which sense to recalibrate.

The inference then, clearly separated. A residual reflex is a short loop still commanded by the brainstem, a reflex arc that bypasses higher revision. Falling asleep also depends on the quality of descending inhibition. We hypothesize that both read the same axis. No study demonstrates this to date, and we do not claim anywhere that integrating a reflex improves sleep.

What Duration Advocates Would Reply, and Why We Maintain

The serious objection exists and it comes from sleep science itself. Quantity and quality do not substitute for each other. Four excellent hours are not worth eight hours, and eight fragmented hours are not worth eight structured hours. Making latency the master variable is selling a shortcut to people who sleep five hours.

It is correct, and we do not circumvent it. We do not say that latency replaces duration. We say that between the two, only one is a test you can administer every night, without equipment, and read as a sign. Duration remains the foundation. Latency is the sensor. Confusing the two would be exactly the error we criticize in the current discourse.

What No One Can Promise You Yet

Our data do not demonstrate that the protocol acts through its sensorimotor component. They also do not demonstrate that a reduction in latency improves the retention of a correction: this link is our thesis, it is not measured here. The resting heart rate decrease, less than two beats, is a signal and not proof of a shift in the [autonomic nervous system](/fr/blog/systeme-nerveux-autonome). And we have no satisfactory explanation for the Toe Touch remaining unchanged: either the protocol does not affect baseline tone, or a global flexion test over thirteen days does not measure it. We do not know.

Only one thing would settle it, and we name it. A replication with sixty participants with an active control group, same duration and frequency, reduced to slow breathing alone. If the latency gap between the two arms is less than five minutes, the sensorimotor component adds nothing and our thesis falls. We will publish in that case too.

Sleep does not work everywhere. It works where you have worked. That's why it belongs to your assessment and not to your end-of-session advice.

👉 I want to be among the professionals who read the loop upstream: labo-rnp.com/fr/pros

LabO RNP is neither medical nor paramedical. Our approach complements a care pathway, never replaces it.

By the LabO RNP team

Frequently Asked Questions About Sleep and the Sensorimotor Loop

Does sleep act on the entire sensorimotor loop or only on certain parts?

It acts in a targeted manner. Sleep spindles increase locally on the regions of the sensorimotor cortex solicited during the day, and not on others (Nishida and Walker, 2007). What you worked on in a session is what the night takes up. Sensorimotor loop and sleep are therefore not two neighboring subjects: the second deals with what the first has produced.

Why talk about sleep onset latency rather than sleep duration?

Because duration depends on the person's life and is not a test. Latency measures the speed at which the nervous system agrees to lower its alert level, it can be assessed with one question, and it changes under intervention. In our data, it went from 20.52 to 9.33 minutes without the duration changing significantly.

Is a very short sleep onset latency a good sign?

No, not below five to eight minutes. At this level, a short latency is a classic marker of sleep debt. Latency is read as an inverted U-curve: too long signals a system that doesn't turn off, too short signals an exhausted system.

What to do when a sensory correction does not hold at the 48-hour retest?

Before concluding the correction failed, check the night. Two hypotheses coexist, the absence of effect and the absence of nocturnal consolidation. A declared latency over twenty minutes is enough to make the retest uninterpretable. Motor decision and nocturnal consolidation are read together: what hasn't been engraved at night will not weigh on the movement the day after tomorrow.

Does the vagus nerve explain falling asleep?

It is not the mechanism we retain. The deactivation of wakefulness is explained at the brainstem level by the decreased activity of the locus coeruleus and the withdrawal of norepinephrine (Poe, 2017). We do not use the polyvagal theory, which remains contested.

Do self-reported sleep data have value?

They measure something other than objective recording, not less well. Subjective measures surpass objective measures in tracking training response (Saw, Main, and Gastin, 2016), and part of what a person reports about their night escapes surface cortical recording. However, they remain sensitive to the expectation effect, which our data do not allow us to exclude.