Your Nervous System Isn't Broken. It's Being Watched.你的神经系统没坏，它只是一直被盯着

Three findings, one pattern

In 1997 the psychologist Daniel Wegner ran a pair of simple experiments. Students were put under mental load, made to hold a long string of digits in mind or told they would be evaluated, and some of them were asked to relax. In the second experiment, under load, the ones told to relax ended up with higher skin conductance, a standard measure of physiological arousal, than the ones who had been told nothing at all. (The first experiment pointed the same way but wasn't statistically significant. The samples were small: 44 and 87 people.)

A year earlier, another team asked good sleepers to fall asleep as fast as they could while a march played in the background. They fell asleep more slowly than people given no instruction. With soft, soothing music the effect reversed: the instruction helped. Under load, trying backfired. Without load, it didn't.

Then, in 2024, a network meta-analysis in *JAMA Psychiatry* pulled apart the components of cognitive behavioral therapy for insomnia across 241 randomized trials and 31,452 people. Cognitive restructuring, sleep restriction and stimulus control earned their place. Sleep-hygiene education turned out not to be necessary. And relaxation procedures, the thing almost every sleep app leads with, were flagged as possibly counterproductive.

Three studies, different decades, one pattern. Under strain, the instruction to calm down can become part of what keeps you up.

Hold that next to the dominant story on social media. In that story, "tired but wired" means your nervous system is dysregulated, stuck in sympathetic overdrive, and the fix is to activate your vagus nerve, reset your dopamine, optimize your HRV and track your recovery. Search interest in "nervous system regulation" rose about 2.5-fold from 2023 to 2025. A market grew to meet it.

Our thesis. For many people, the state isn't mainly a broken switch. It's a supervision problem. Part of the mind is assigned to watch for failure: am I calm yet, am I asleep yet, is my heart rate normal. Under load, that watcher wins, and the watching itself keeps arousal high. The regulation boom mostly sells more watchers, in the form of scores, alerts and hacks that ask you to check. What tends to work is the opposite move: commanding only what you can command, letting the rest happen, and learning when to relieve the guard.

That's a strong claim, so here is the evidence, the limits, and what would change our mind.

1. The inspector in your head

Wegner's explanation became known as ironic process theory. When you try to control a mental state, two processes run. An operating process does the work: relaxing, focusing, falling asleep. A monitoring process scans for signs of failure: still tense, still awake, still thinking about it. The monitor is cheap and automatic. The operator needs resources. Under fatigue, stress or distraction, the operator weakens and the monitor keeps running, which brings back exactly what you were trying to avoid.

It's a lab-based theory, and it shouldn't be stretched too far. But it lines up with several independent findings.

Worry keeps the body paying after the event. A meta-analysis of 60 studies found that during worry and rumination, blood pressure, heart rate and cortisol are higher and heart rate variability is lower (in the experimental studies, roughly g≈0.45 for systolic pressure, 0.28 for heart rate, 0.36 for cortisol). These are small-to-moderate effects in healthy people, not proof that worry causes disease. The theory behind them, the perseverative cognition hypothesis, makes a specific point: what prolongs physiological activation is often not the stressful event itself but its continued representation in your mind after it's over. In one study of 73 teachers wearing heart monitors for four days, heart rate was higher and HRV lower in the hour they reported worrying and the two hours after.

Suppression mostly changes the face. A large meta-analysis of emotion-regulation strategies found that trying to suppress feelings or thoughts directly had effects close to zero (d+ of −0.04 and −0.12). Suppressing the outward expression did change the face, but physiological arousal, if anything, rose slightly.

Insomnia runs on monitoring. The leading cognitive model of insomnia describes a loop: worry about sleep leads to selective monitoring of sleep-related threats, the clock and body sensations, which leads to overestimating the problem, which raises arousal. In a small experiment, people with insomnia found it harder to pull their attention away from images of clocks. Our research notes put it in one line: you can decide when you get up; you can only allow yourself to fall asleep.

And one more piece, which matters later: the "willpower is a fuel tank" idea failed two large preregistered replications, across 23 labs (d=0.04) and 36 labs (d=0.06). Effort is real. Fatigue is real. What seems to change after sustained effort is less a depleted tank and more your response caution: people become riskier and less willing to wait.

Put together, the picture is consistent. You can command actions. You cannot command outcomes like calm or sleep. When you try, you create an inspector to check whether the outcome has arrived, and under load the inspector becomes part of the problem.

2. The market sells more inspectors

Now look at what the regulation market offers. Much of it is measurement: rings, watches and apps that turn your body into a dashboard of scores. Measurement can be valuable. But the evidence on what these devices can actually tell you, day to day, is sobering.

- Daily HRV is noisy. For the same person under the same conditions, morning HRV (lnRMSSD) commonly moves 10–20% from one day to the next.

- The noise can swallow the signal. In observational data from about 28,000 app users, next-morning HRV was about 12% lower after a night of heavy drinking and about 10% lower during illness. Both sit inside ordinary daily variation.

- Composite scores are unvalidated. "Readiness" and "body battery" scores have no independent validation. In one preregistered study, a wristband's stress score predicted people's self-rated stress only marginally. Plain heart rate predicted it best.

- Few wearable metrics hold up. An umbrella review covering more than 150 device models found only heart rate and atrial-fibrillation detection had relatively strong support.

- Sleep stages are a guess. Wrist devices agreed only modestly with lab sleep staging (κ about 0.21–0.53) and correctly identified wakefulness just 29–52% of the time.

- A "better" number can be a false positive. In a small study, 160 mg of caffeine at bedtime raised nighttime high-frequency HRV and lowered heart rate. Caffeine also costs sleep: a meta-analysis estimates that a typical coffee (about 107 mg) should be finished at least 8.8 hours before bed.

None of this means the devices are useless. Overnight readings from a good chest strap or ring track ECG closely, and weekly averages are far more informative than single days. The problem is resolution. Most people read a daily score at a resolution finer than the noise, and then act on it.

That is where the inspector problem returns. Clinicians even have a name for the extreme case: orthosomnia, anxiety from chasing perfect sleep data, first described in case reports of patients who lay in bed longer to make the tracker show more sleep.

There's a deeper reason measurement can backfire, and it has to do with how well we sense our own bodies. In one study of 572 people, reported heartbeat counts correlated with actual heartbeats at only r=.16, and more than 95% of scores were undercounts. Across 55 studies, anxiety had no relationship with how accurately people perceive their heartbeat. What anxiety does track, across 71 studies, is negative appraisal of body signals and attention to them.

A 2026 review of panic disorder takes this a step further. It argues that monitoring the body narrows the range of heartbeat sensations that feel acceptable, which can itself trigger panic. A classic 1988 experiment showed the mechanism in miniature: panic patients shown false feedback that their heart was racing developed real increases in heart rate, blood pressure and skin conductance, if they believed it.

The body isn't lying to you. It's guessing, and it's a dutiful but timid sentry. Give it more alarms to watch and it will find more reasons to sound one.

3. The vagus nerve is becoming the new “qi”

The second pillar of the regulation story is the vagus nerve, and here the gap between the feed and the literature is wide.

The popular frame comes largely from polyvagal theory, formally published in 1995. In February 2026, 39 scholars of vagal physiology, evolution and vertebrate social behavior published a paper titled "Why the Polyvagal Theory is Untenable" (Grossman et al., *Clinical Neuropsychiatry* 23(1):100–112). Stephen Porges published a long reply in the same issue, arguing the critics had assessed a reconstructed version of his theory. A medical news site then ran the story under the word "debunked," without mentioning the reply.

The honest summary: the market says "activate," the headline says "debunked," and the science says the argument isn't finished. The clinical phenomena that made the theory popular, feeling safe, freezing, dissociation, are real. The dispute is about the mechanism.

Meanwhile, the hacks:

- Ear-clip vagus stimulation. A Bayesian meta-analysis of 16 sham-controlled studies found acute stimulation had no reliable effect on vagally mediated HRV (g=0.014).

- Cold water on the face. It triggers a real reflex. Cardiac vagal activity rose only during the cold exposure (g=0.59), not afterward (g=0.11). It's a reflex, not a switch.

- "Fake" breathing. In a 2023 blinded trial, 400 UK adults spent four weeks doing either slow breathing at about 5.5 breaths per minute or a matched placebo breathing at 12 per minute. Both groups improved on stress, anxiety and wellbeing. There was no difference between them, and expectations were equal in both arms. A 2024 trial of fast breathing with long breath-holds against a sham found the same: both improved, no significant difference.

- But breathing still beats nothing. A meta-analysis of breathwork trials found small-to-moderate reductions in self-reported stress (g≈−0.35) against controls.

Here is one way to read that combination, and it's our inference, not a finding. A lot of the benefit may come from doing a short, structured, deliberate action at all, an action you can actually command, rather than from hitting a precise frequency or "activating" a nerve. That would also explain why the gadgets underperform: they promise to change an outcome directly.

4. The page torn out of every meditation app

The third pillar is meditation, sold almost everywhere as pure calm. The research on its side effects contains one of the most striking measurement results we found.

In a US sample of 434 meditators, researchers asked one general question: had meditation ever led to a challenging, difficult or distressing experience? 32.3% said yes. Then they asked the same people ten specific questions. The figure rose to 50.0%. Same people, same practice. Only the questions changed. 10.6% said it had affected their daily functioning, and 1.2% reported impairment lasting a month or more.

A separate study of 886 US meditators found 31.4% had experienced something difficult or distressing and 9.1% had functional impairment. Another study, of an eight-week mindfulness course, found that one open-ended question identified only 26 of the 80 people who had any meditation-related effect, missing more than two-thirds of the adverse effects that specific questioning found.

In Hong Kong, 21% of practitioners in one survey reported negative effects, yet 63% believed meditation was completely risk-free. For mainland China, a 2022 review concluded that research on this question was essentially absent.

Balance matters here. Adverse experiences are not unique to meditation, most are transient, and in one digital program about nine in ten people who reported one were still glad they had learned. The point isn't that meditation is dangerous. It's that the rate of difficulty you see depends on whether anyone asks, and the business model rewards not asking. Median 30-day retention for mental-health apps is around 3%. Calm sells. Warning labels don't.

A manual without a side-effects section isn't safer. It's just easier to sell.

5. The system underneath

Step back and the pieces form one system.

1. Load goes up. Work, screens, notifications, markets, news. Each adds something to watch.

2. The inner inspector works harder. Am I calm? Am I asleep? Is that sensation normal?

3. The market adds external inspectors. Daily scores, stress alerts, sleep grades, each read at a resolution finer than its noise.

4. The acceptable range narrows. More checking means more deviations noticed and more alarms raised. The body, a timid sentry, obliges.

5. Arousal stays high, and the loop asks for more control. Which means more monitoring.

Our research notes contain a line that explains why this loop is profitable: what can't be measured can't be compared, and what can't be compared can be priced. Calm is hard to verify, so claims about it are cheap to make and easy to sell.

This loop connects to the other two pieces in this series in a way that surprised us.

Our technology piece looks at a different checker: the human who reviews what an AI agent wants to do. There, the problem is the opposite. Users approve 97% of permission prompts, and in one company study people caught 13.6% of dangerous commands. The external checker is asleep at the switch. Here, the internal checker won't go off duty.

We think these are the same faculty, misallocated. Vigilance is a limited budget, not in the sense of a fuel tank that drains, but in the plain sense that attention pointed at one thing isn't pointed at another. Our research files put it as: attention isn't "focused," it's "priced." If you spend that budget scanning your heart rate, your sleep score and your feed, there's less of it left for the few moments that actually need it: the command that would wipe a database, the contract you shouldn't sign, the trade you'd regret.

Decisions are where this gets expensive. In one lab study, a single night without sleep did not make people riskier or more impulsive. Five nights of five hours did. Behavioral finance has its own version of the monitoring problem: investors who check their portfolios more often see more losses and take less risk than their goals justify. The finance piece in this series follows that thread into who ends up carrying the risk of the AI build-out.

6. What holds up

None of this means giving up on regulation. It means aiming at what can actually be commanded. These are working proposals, labeled with the strength of evidence where we have it, not treatment. Persistent sleep problems, panic or heart symptoms belong with a doctor; the first-line treatment for chronic insomnia is CBT-I delivered by a professional.

Command actions, not outcomes. Relaxation isn't a command; it's a result. What you can issue are actions: a minute of slow breathing done as an action rather than a demand for calm; naming a feeling in a single word (labeling reduced amygdala responses in lab studies, grade B); a fixed wake-up time, which is the biggest lever on sleep you actually control.

Read the week, not the day. If you track, set a personal baseline band from a 7-day average, interpret change only after a few weeks, and ignore single nights. For sleep, trust three simple numbers: when you went to bed, when you got up, how long you slept. In motor learning, people given feedback on half their attempts retain a skill better than people given feedback every time. Whether that transfers to breathing or body-reading hasn't been tested, but it suggests scaffolding is meant to come down. Take measurement holidays.

Know the base rates. Healthy adults in sleep labs wake roughly 21 to 43 times a night; our inference is that most of these brief wakings are never remembered. In a study of 1,248 good sleepers, 27% developed acute insomnia within a year; 72% of those recovered, and over the year 1.8% of the whole sample developed chronic insomnia. One bad night is noise, not a verdict. Regularity also matters: in a UK cohort of about 61,000 people, everyone else had 20–30% lower all-cause mortality than the least regular fifth of sleepers after full adjustment, and regularity predicted it better than duration (an observational result).

Remove some of the alarms. In a randomized crossover study, 467 adults blocked mobile internet on their phones for two weeks. 91% improved on at least one outcome among wellbeing, mental health and sustained attention, and they slept about 18 minutes more a night. Only about a quarter complied fully, which tells you how strong the pull is.

Gate irreversible decisions. A rule we use, which is a design rule and not a research finding: after a night under six hours, postpone decisions you can't undo by 24 hours, or turn them into a version you can undo. It's deliberately stricter than the evidence: in the study above, one sleepless night alone didn't raise risk-taking, while five short nights did.

Interrogate any number. Who asked? How was it asked? What was counted? How big was the denominator? A percentage without a denominator is an advertisement, not data.

7. Limits, and what would change our mind

The evidence for the core mechanism comes mostly from laboratory studies with modest samples, and one of Wegner's two experiments was not significant on its own. The sleep-instruction effect reversed under low load. Monitoring isn't always harmful either: a meta-analysis of 138 studies found that monitoring progress toward a goal improves goal attainment (d=0.40), especially when it's recorded and shared. Our claim is narrower: monitoring outcomes you can't directly command, at a resolution finer than their noise, tends to backfire.

We would change our view if well-powered trials showed that daily score-based feedback improves sleep or stress outcomes over weekly or no feedback, if sham-controlled studies showed specific breathing rhythms or vagal devices outperforming matched placebos on lasting outcomes, or if large surveys showed meditation difficulty rates that don't depend on how the question is asked.

Keep the best-supported parts of measurement: heart-rate and atrial-fibrillation alerts exist for good reasons.

The guard can be relieved

The test our research team set for any practice is simple: does it still work when you take the instrument off? A skill that only exists while a device is watching hasn't been learned yet.

Your nervous system probably isn't broken. In many cases it has been put on permanent watch duty, by circumstances, by habits and by products that profit from the watching. The work isn't to install a better alarm. It's to learn which things you can command, which things you can only allow, and when to let the inspector go home.

Real quiet isn't failing to hear any sound in the mountains. It's still being able to hear yourself in the middle of the market.

See the system. Build leverage.