This piece is built from one of my recent Monday-night livestreams, where I work through emerging research live and take questions from the audience. I have kept the science and dropped the names of everyone who asked questions. What follows is my read on three 2025 papers that, taken together, point at something I have been calling cognitive cadence.
What does "cognitive cadence" actually mean?
You already know the circadian rhythm, the roughly 24-hour cycle that governs sleep and wake. You may know the ultradian and infradian rhythms, the ones shorter or longer than a day. Your brain also runs on much faster rhythms, oscillations measured in cycles per second, and those rhythms have structure. Information processing is paced, not a constant blast.
The core claim: your brain cannot compute and clean at the same time. Heavy thinking and recovery appear to be mutually exclusive, almost like the run mode and sleep mode on a computer. When you force the brain to stay in compute mode too long, through chronic stress, sleep deprivation, or grinding without breaks, the system starts hijacking your waking life to do the maintenance you skipped.
I am calling this a cadence perspective because oscillation, not a fixed set point, is what keeps a biological system stable. Your body temperature does not sit at 98.6. It swings about half a degree across 24 hours, on purpose. Cortisol and insulin oscillate. The systems property that matters is homeodynamics, the ability to keep returning to a healthy rhythm, rather than homeostasis, the ability to hold a fixed value.
Three papers from 2025 mapped this at three timescales: half a second, ten seconds, and a few minutes. I will walk through each.
How does the brain reset attention in half a second?
A group at MIT, Batabyal and colleagues, recorded from dwelling electrodes in the prefrontal cortex of monkeys doing working memory tasks. They threw in irrelevant visual distractors and watched what hundreds of neurons did to get back on task.
The recovery is not the neural population popping straight back to the prior activation. The population traces a pattern, what the paper calls a rotating wave, and runs the phase a full cycle before it lands back where it started. That recovery takes about 500 milliseconds (Batabyal et al., 2025). In dynamical terms, the firing pattern circles back to its attractor, the configuration the circuit wants to return to.
This is a half-second delta-range reset of executive resources, and the timing is non-negotiable. When the next stimulus arrived before the 500 milliseconds were up, error rates climbed sharply. The wave needs its time to finish.
This is the mechanism behind why multitasking costs you, beyond mere inefficiency. Every context switch forces at least half a second of resynchronization. One of the senior authors, Earl Miller, has made the point that oscillations are energy-efficient. A pulsed reset, locked in when you need it, costs far less than holding the attention system rigidly on all the time. That economy is part of why I think you see this same effect in human attention. For the broader picture on attention and pacing, see my work on flow state and on procrastination, where approach-versus-avoid timing locks people up.
What is vasomotion, and why should you care about it?
Zoom out to ten seconds. Your brain's blood vessels do not hold a steady diameter. They flex on a rhythm of roughly one cycle per ten seconds, about 0.1 Hz. This is vasomotion, and we have seen it in imaging for decades without knowing what drives it. During slow-wave sleep the whole brain starts to pulse near twice per second, and cerebrospinal fluid moves through the tissue like water agitating in a washing machine, flushing metabolic waste.
Turner and colleagues at Penn State found the driver. A small population of neurons sitting in deep cortical layers releases nitric oxide in rhythmic bursts. These neuronal nitric oxide synthase neurons, the nNOS population, act as vascular pacemakers, timing the flex of the vessels (Turner et al., 2025).
When this reads on biofeedback tools I work with, it clicks into place. I run passive infrared hemoencephalography, pIR HEG, a biofeedback method where a sensor on the forehead reads vascular tone and you drive blood flow up with focus. Jeff Carman, who developed the device, reported that migraineurs show markedly more complete perfusion on infrared imaging after a single HEG session. If we are nudging the nNOS neurons to change how they pace the vasculature, that durable migraine effect makes mechanistic sense, which is otherwise hard to explain from simple blood movement alone.
Now the cost. When researchers killed these cells in mice, vasomotion power dropped by about 60 percent and cortical slow-wave activity dropped by about 30 percent (Turner et al., 2025). The effect is two to three times larger in non-REM sleep than in waking, which is exactly when waste clearance matters most. Without vasomotion, the pressure gradient that pushes CSF through the perivascular spaces collapses, and amyloid, tau, lactate, and damaged proteins accumulate.
Four weeks of unpredictable stress in mice killed 20 to 30 percent of the nNOS population, and the survivors fired less (Turner et al., 2025). This is one route by which chronic stress accelerates neurodegeneration: alongside the direct hippocampal atrophy from high cortisol, you get slow loss of the cells that keep clearance rhythms alive. The trash piles up. For how to protect this infrastructure, my pages on sleep, stress and fight-or-flight, and red-light photobiomodulation cover the practical levers.
What happens when you deny the brain its maintenance window?
Third timescale, a few minutes. Yang and colleagues at Boston University kept people awake for 24 hours, put them in an fMRI scanner, and ran a simple reaction-time task. Press a button when you see a cue. They watched for lapses.
Every two to five minutes, like clockwork, the sleep-deprived subjects zoned out (Yang et al., 2025). The lapse ran on a predictable sequence. About 12 seconds before the error, the pupils constricted. Around 10 seconds before, heart rate slowed into parasympathetic dominance. About 5 seconds before, the EEG shifted into slow-wave dominance, delta and theta surging like sleep onset. At zero, the error, a massive CSF outflow from the ventricles (Yang et al., 2025). Then about 10 seconds after, pupils redilated, heart rate climbed, CSF returned, and the person came back online.
I have watched a version of this in my own work. On a 25-minute continuous performance task, put a camera on the subject and almost everyone, around the 20-to-25-minute mark, stares blankly for 15 to 20 seconds and then resumes. Ask them afterward and they report they were present the whole time. They were briefly offline and could not tell.
This is a forced state transition driven by adenosine. As the sleep-promoting waste builds past a threshold, the brain shuts down arousal and inserts a micro-maintenance window. The subjects in the study were paid to stay alert and could not suppress it. This is why drowsy driving is as dangerous as drunk driving. You are intermittently offline, and you cannot predict when.
What does this look like in a QEEG brain map?
I did not put up maps on the stream, but these mechanisms interact with signatures I see constantly in QEEG brain mapping. The two big markers of stress colliding with fatigue are alpha speed and delta speed.
Alpha is the brain's idle frequency, like the idle speed of a car engine. In a healthy brain the alpha index runs at a consistent speed across a hemisphere, drifting only slightly front to back. When the timekeeping degrades, alpha slows globally and spreads out within the hemisphere, so language tissues that need to hand off information at the same speed fall out of sync. That produces word-finding trouble, tip-of-the-tongue, slowed processing, and what people report as memory problems but is usually retrieval speed or attention. My deeper treatment of this is in decoding alpha waves.
A practical rule of thumb: within one person you can tolerate roughly half a standard deviation of alpha-speed variation across a hemisphere before it shows up as delay. Higher individual alpha frequency has been associated with better cognitive performance in the research, especially in language-dominant tissue. You want alpha fast and synchronized within the hemisphere.
Delta is the heartbeat of the brain, near twice per second, and it is central to clearance. When you are a little tired or sick, delta speed climbs to about one standard deviation above typical while the brain surges recovery modes. Sit in that elevated state long enough and delta collapses to slow delta, around 1.5 Hz, the state of being half-asleep while awake and unable to truly sleep at night. That collapse is regulatory failure, and chronic stress plus sleep debt is the path into it. This is the same territory I cover in biohacking brain fog.
Can neurofeedback and biofeedback retune the cadence?
Some of this is tunable, which is the actionable part.
For raising peak alpha, you can train SMR, 12 to 15 Hz, rather than pushing alpha directly. Alpha resists being driven and tends to rebound when you reward it up. If slow alpha is slow because beta is unstable, stabilizing SMR is the more reliable route. I will also reward slow alpha in a narrow band, roughly 6.5 to 9.5 Hz, while inhibiting 4 to 7 Hz, to surge slow alpha without dragging up fast alpha. The principle is to get other rhythms out of alpha's way rather than standing on the gas. The full reasoning is in my SMR neurofeedback guide.
For the vascular rhythm, HEG biofeedback appears to act through the nNOS-to-vasomotion pathway, which is the most plausible explanation for durable migraine and brain-fog recovery. A clean test would be to tag nitric oxide and image the nNOS response with PET or SPECT before and after HEG. That study is waiting for a graduate student.
A few honest caveats on evidence strength. The three core papers are well-controlled animal and human studies, but the cadence framework that ties them together is my synthesis, and the HEG-to-nNOS link is extrapolation I find compelling rather than established. Alpha-theta training, by the way, does not undo the physiological costs of sleep deprivation. It sits too close to a sleep phenomenon and tends to erode sleep quality; SMR and left-side beta training are the better tools there.
What should you actually do with this?
You cannot work at the point of failure. Willpower does not override adenosine, and being paid to stay alert does not stop the forced lapse. Interventions have to target the underlying oscillations: circadian rhythm, blood-sugar swings, the stress cycle, and the recovery windows you are skipping.
Cognitive decline, at the cellular level, involves losing specific neurons that keep the rhythm alive, the way you lose Betz cells and watch elders go downstairs stiff-legged because the spring is gone. The nNOS population may be one of those critical nodes. That helps explain why poor sleep and chronic stress predict dementia risk decades later. You are degrading the maintenance infrastructure, not just feeling tired.
Your brain is a biological system running nested rhythms, half a second, ten seconds, a few minutes, 90-minute ultradian cycles, the 24-hour sleep-wake loop. Healthy cognition is the right states in the right sequence at the right amplitude. Defend your sleep, manage stress before it kills the pacemaker cells, stop trying to multitask through the half-second reset, and take real recovery breaks before the brain forces them on you. If you want to see your own alpha and delta speeds, get a brain map and start from data.
Primary sources: Batabyal et al. (2025), Journal of Cognitive Neuroscience, https://doi.org/10.1162/JOCN.a.2410; Turner et al. (2025), eLife, https://doi.org/10.7554/eLife.105649.3; Yang et al. (2025), Nature Neuroscience, https://doi.org/10.1038/s41593-025-02098-8.