Sleep, Wakefulness and the Programmable Brain

Two people reach three in the morning having been awake for the same twenty hours. One is heavy-lidded, irritable, and useless for anything requiring judgment. The other finishes a chapter, checks a spreadsheet, and feels mildly fine. Elapsed waking time is identical. What differs is the state of two internal systems that were never designed to agree with the clock on the wall: an accumulating sleep pressure and a circadian rhythm that swings wakefulness up and down across the day. When people ask how much of sleepiness could one day be dialed up or down on purpose, they are really asking how far those systems can be pushed without breaking the machinery they regulate.

That question is now answerable in three separate registers, and it is worth keeping them apart. Some parts of sleep regulation are demonstrated, causal, and mechanistically understood. Some are plausible engineering, with working prototypes in the laboratory or in early clinical trials. And some are scenarios that depend on capabilities that do not yet exist. Most public confusion about “programmable sleep” comes from sliding between those registers inside a single sentence.

The two clocks behind the feeling of sleepiness

The foundational framework is the two-process model introduced by Alexander Borbély in 1982 and revisited by him and his collaborators in a 2022 review in the Journal of Sleep Research (Borbély, 2022). The model is almost embarrassingly simple, which is part of why it has survived four decades. A homeostatic process, Process S, rises while you are awake and falls while you sleep; it is indexed empirically by slow-wave activity in the sleep EEG, the large slow oscillations that dominate deep non-REM sleep. A circadian process, Process C, oscillates roughly every twenty-four hours and is set primarily by light. Sleepiness at any moment is a function of where you are on both curves.

The model’s grip on reality is easy to demonstrate. Keep someone awake past their usual bedtime and slow-wave activity in the recovery sleep rises, then decays exponentially across the night. Nap in the late afternoon and the evening pressure is partly discharged, so bedtime arrives without its usual force. These are among the more robust regularities in human physiology.

What is less widely appreciated is how much the model does not yet explain. In a 2025 mathematical reappraisal in npj Biological Timing and Sleep, Anne Skeldon and Derk-Jan Dijk laid out the model’s internal machinery — two thresholds that a combined homeostatic and circadian signal must cross to trigger sleep onset and wake, plus a set of roughly eight parameters that govern the shape and timing of the curves (Skeldon & Dijk, 2025). They also noted something important for anyone promising precise sleep control: no physiological correlate has yet been identified for the threshold levels themselves, or for the amplitude of the circadian drive. The model predicts behavior beautifully while its central parameters remain abstract. That is a genuine gap, because a control system needs to know which knob it is turning.

Skeldon and Dijk make a related point that cuts against the fantasy of a master dial. The threshold that governs sleep onset is not fixed. It responds to light exposure, to caffeine, and to something more slippery — the level of engagement or motivation a person brings to staying awake. Extended versions of the model in which light feeds back into the circadian oscillator can predict the timing effects of real interventions, but only within a narrow band and only when the individual’s parameters are known. The practical consequence is that sleepiness is a moving target whose calibration shifts with context.

Where the arousal switch actually lives

If the two-process model describes the pressure, a small cluster of neurons in the lateral hypothalamus describes the valve. Orexin, also called hypocretin, is a neuropeptide produced by a few tens of thousands of cells in the dorsolateral hypothalamus. In a 2002 review in Nature Reviews Neuroscience, J. Gregor Sutcliffe and Luis de Lecea described the system’s reach: orexin neurons project widely to the noradrenergic locus coeruleus, to histaminergic and cholinergic nuclei, and to the arousal machinery that keeps the cortex awake and responsive (Sutcliffe & de Lecea, 2002).

The clinical evidence that this system is not decorative comes from narcolepsy type 1, in which orexin-producing neurons are lost, most likely through an autoimmune process. People with the disorder do not merely feel sleepy; they collapse into REM sleep at inappropriate moments, experience cataplexy, and show a fragmented night despite overwhelming daytime sleepiness. The lesson is that orexin does two jobs at once. It stabilizes wakefulness against the constant pressure to drift into sleep, and it gates the boundary between wake and REM, keeping the transitions clean.

This is the strongest demonstrated lever on human arousal that exists. It is also, for anyone imagining elective control, a caution. Killing orexin neurons produces a devastating neurological disease. A system that gates the wake–sleep boundary is not a volume knob.

Turning the wake drive down

The first practical exploitation of that lever is the class of drugs called dual orexin receptor antagonists, or DORAs. These molecules block orexin’s receptors and thereby lower the wake drive rather than broadly depressing the brain the way older sedatives do. Daridorexant, developed by Idorsia, is the clearest example, and its phase 3 results were reported by Emmanuel Mignot and colleagues in The Lancet Neurology in 2022 (Mignot et al., 2022). Two trials randomized a combined 1,854 adults with insomnia disorder to drug or placebo for three months. At the 50 mg dose, wake after sleep onset fell by about 23 minutes relative to placebo at month one and about 18 minutes at month three; latency to persistent sleep improved by roughly 11 minutes. Self-reported total sleep time increased by around 20 minutes, and participants reported less daytime sleepiness.

Those numbers are real and, in a field where sleep drugs have a long history of disappointing results, meaningful. They are also modest in absolute terms, and the trial itself lists limits: participants were mostly White, people with significant comorbidities were excluded, and the study ran only three months. The FDA approved daridorexant in January 2022 on the strength of this package. The drug’s significance is conceptual as much as clinical. It demonstrates that a specific arousal signal can be turned down with a pill, improving sleep and some measures of daytime functioning in the trial. Next-day impairment remains a possible adverse effect and must be assessed for each patient.

But note precisely what was demonstrated: symptom relief in a diagnosed disorder. The drugs are approved for insomnia, at doses tested for that purpose, with a risk profile assessed in that population. They are not approved as a tool for making a healthy person sleep on demand, and nothing in the data says they would work that way.

Turning the wake drive up

The mirror-image experiment is more interesting for the question of deliberate alertness, because it tests whether an underactive wake system can be restored rather than merely propped up. In 2025, Yves Dauvilliers and colleagues reported in the New England Journal of Medicine a phase 2 trial of oveporexton, an oral drug that selectively activates the orexin receptor type 2, in 112 people with narcolepsy type 1, of whom 90 received oveporexton and 22 received placebo (Dauvilliers et al., 2025).

The results were substantial for a condition with few good options. Participants who received the drug stayed awake longer on the maintenance of wakefulness test, with improvements in sleep latency far exceeding the change in the placebo group, and their scores on the Epworth Sleepiness Scale fell by roughly nine to fourteen points depending on dose, against a drop of about two and a half points on placebo. In two of the doses, the frequency of cataplexy — the sudden loss of muscle tone that defines the disorder — was reduced.

The side-effect profile is the part that should temper any story about a future alertness pill. Insomnia occurred in about 48% of participants, though it usually resolved within a week; urinary urgency and frequency appeared in roughly a third. These are the fingerprints of a drug that is pushing the arousal system hard. Even in a population whose problem is not having enough wake drive, the body pushes back. The trial was phase 2, in a rare disease, with 112 participants. It established that the approach can work. It did not establish anything about tuning alertness in people whose orexin system is intact and who simply want more usable hours.

What stimulation can and cannot do to the sleeping brain

A third line of work bypasses pharmacology entirely and tries to sculpt the sleep itself. The slow oscillations of deep sleep are not just a marker of recovery; they coordinate the replay and consolidation of memory. If they could be amplified, the reasoning went, the memory benefits of sleep might be amplified too.

In 2006, Lisa Marshall and colleagues reported in Nature that applying a weak oscillating electrical current to the scalp at 0.75 Hz during early non-REM sleep increased slow-wave activity and improved retention of hippocampus-dependent declarative material the next day (Marshall et al., 2006). A control condition at 5 Hz suppressed the oscillations and produced no memory benefit. In 2013, Hong-Viet Ngo and colleagues in Neuron showed something more precise: delivering brief sounds in phase with the up-state of the ongoing slow oscillation enhanced not only the oscillation but phase-coupled spindle activity and declarative memory consolidation, while out-of-phase stimulation did nothing (Ngo et al., 2013). The studies were small — around ten participants each — but the mechanism was specific and the controls were tight.

Then came the correction that every honest account of this literature has to include. In 2019, a team led by Justus Henin and colleagues reported in eNeuro a double-blind replication of closed-loop acoustic stimulation (Henin et al., 2019). The stimulation worked in the narrow sense: it did enhance slow oscillation and spindle power. It produced no behavioral memory benefit, in either a spatial-navigation task or a word-pair task. A technique that had looked like a demonstrated tool for improving sleep’s function now looked like a way of changing the EEG signature of sleep without reliably changing what sleep accomplishes.

That single result deserves more weight than it usually receives. It is the difference between being able to move a number the brain produces and being able to move the thing that number is supposed to stand for. Closed-loop stimulation of sleep is, on the current evidence, an open question rather than a settled capability.

Demonstrated, plausible, speculative

Laid side by side, the three lines of evidence grade very differently.

Demonstrated: the two-process structure of sleepiness, the causal role of orexin in stabilizing wake and gating REM, the ability of orexin-receptor antagonists to reduce wake time in insomnia, and the ability of an orexin-receptor agonist to restore wakefulness in narcolepsy. These rest on replicated physiology and randomized trials.

Plausible engineering: individualized sleep timing that uses a person’s own data to schedule light, caffeine, and naps; closed-loop stimulation tuned to the individual rather than the group average; and scheduling systems that predict performance decline before the person feels it. Each of these is an extension of demonstrated mechanisms, but none is yet a validated product, and the failure of the acoustic-stimulation replication is a reminder of how often plausible extensions do not survive contact with controls.

Speculative: a general-purpose dial for alertness in healthy people, or the ability to replace sleep’s restorative function with a signal. The pharmacology argues against the second in particular. Orexin agonists in narcolepsy produce insomnia in half the participants because wakefulness and sleep drive are coupled; raising one raises the other’s shadow. The idea that sleep is a single quantity that can be compressed is not supported by anything in this literature.

Suppressing fatigue is not restoring capacity

The most consequential distinction in this whole area is between making a tired person feel awake and making a tired person recovered. They are different interventions with different risks, and the difference is easy to blur.

A stimulant, or an orexin agonist used outside its approved setting, can raise alertness while the underlying debt remains unpaid. The person feels capable and is not. Performance on simple tasks may hold steady while judgment, emotional regulation, and the formation of new memories degrade, and the subjective sense of impairment is the first thing to disappear. Alertness without recovery also carries the risk of masking a medical problem — sleep apnea, depression, a thyroid disorder — that would otherwise announce itself through daytime sleepiness.

Sleep researchers have an unusually defensible answer to this. Sleep’s restorative functions accumulate a debt that cannot be discharged by anything except sleep, and the slow-wave activity that the two-process model uses as its index is the visible trace of that debt being paid. Technologies that raise alertness without letting the books settle are borrowing against an account that must eventually be repaid with interest. Any honest assessment of a “programmable” arousal system has to ask whether it improves the underlying capacity or merely hides the evidence that capacity is running low.

Who gets to set the target

Even a safe, effective arousal technology raises a question that engineering cannot answer. Alertness only becomes a tool once someone decides how alert a person should be, when, and for whose benefit.

The commercial version of this problem is familiar from other domains. A system optimized to keep a workforce productive will find that suppressing fatigue across a night shift improves the metric it was given. It may even be right that the workers feel better in the short run. But the shift schedule that created the fatigue, the staffing level that made it unavoidable, and the question of whether the same work could be arranged differently all become invisible once a technological fix is available. The intervention narrows the space of questions that get asked.

Consent is the other weak point. It functions differently when an employee, a student, a patient, or a gig worker has a realistic alternative than when the choice is between opting in and losing the position. It weakens again when the intervention itself alters the person’s preferences, because the person who agreed beforehand and the person making the decision afterward may not be the same court of judgment. For any system that changes arousal, reversibility, the ability to disconnect, and periodic re-consent at a fresh baseline are not bureaucratic flourishes. They are part of what makes the choice a choice.

Military and safety-critical contexts illuminate the stakes from a third direction. There, the goal is to sustain performance through conditions that human physiology was never meant to handle, and the case for intervention is stronger. The same capability, pointed at a factory floor or a classroom, serves a much less defensible purpose. The technology does not change; the objective does.

What would settle the question

The evidence that should change anyone’s mind is fairly specific, and it is not the kind that makes headlines.

First, an intervention that improves objective performance and later sleep and next-day recovery, rather than trading one for the others. Second, independent replication, ideally across laboratories, of any stimulation effect on memory — the acoustic-stimulation story shows why a single positive result is not enough. Third, adverse-event reporting that includes the effects people would rather not advertise, such as rebound insomnia, urinary symptoms, or a growing dependence on the tool. Fourth, evidence that any benefit persists after the intervention is withdrawn.

The two-process model supplies a final standard. If someone claims to have made sleep programmable, the question is which parameter was moved, and whether the body’s other parameters stayed where they should. The model’s own incompleteness — the missing physiological correlate of its thresholds — is a warning against treating the curves as if the knobs were already labeled.