don't let your sleep score keep you up at night

there's a patient type sleep clinics started documenting a few years ago: someone who sleeps, by every objective measure, fine, but who has become so fixated on perfecting their wearable's sleep score that the fixation itself is wrecking their sleep. researchers coined a name for it: orthosomnia. the pursuit of perfect sleep, as measured by a device, producing bad sleep.

the tool you bought to improve the thing is the reason the thing is broken. i can't think of a cleaner parable for where the quantified-founder culture has drifted.

builders were always going to fall for this

i understand the appeal completely, because it's aimed at people like me. builders instrument systems. dashboards are how we love things. so when the industry offered dashboards for the body, rings, straps, patches, scores, we adopted them the way we adopt observability tooling: more metrics, more control.

but the analogy smuggles in an assumption that's false for bodies: that observation is free. in software, a metrics exporter doesn't slow the service. in humans, the observer is inside the system. a number you check with anxiety changes the system that produces tomorrow's number. the dashboard is a component of the machine, and it can be the broken one.

what the instruments actually measure

before the psychology, the engineering, because the accuracy story alone should lower the temperature.

consumer trackers are genuinely good at one thing: telling sleep from wake. sensitivity for detecting sleep runs above 95%. they're mediocre at the thing people actually obsess over: sleep staging. against polysomnography, the lab gold standard, four-stage classification accuracy runs roughly 65 to 80%, and the devices systematically overcall sleep when you're lying awake. the "you got 43 minutes of deep sleep" readout that ruined your morning is, a meaningful fraction of the time, simply wrong.

the glucose patch is a purer case. continuous glucose monitors are life-changing instruments for diabetics. in metabolically healthy people, the evidence for benefit is close to empty: the trials showing improvements are in overweight or prediabetic cohorts, short, and about behavior nudges; there is no rct showing cgm improves any hard outcome in healthy users. the "glucose spike" content economy is running far ahead of its data, selling healthy people a disease model of their own lunch.

so the stack, honestly labeled: one metric family that's solid (sleep/wake timing, resting heart rate trends), one that only looks precise (staging scores), and one that's evidence-free theater for healthy users (the patch). that's what's underneath the confidence of those crisp morning numbers.

the anxiety is measurable now

the psychological cost stopped being anecdote. the orthosomnia literature has grown from case reports into validated scales, and the correlations are exactly what you'd guess: it travels with perfectionism, health anxiety, and obsessive tendencies. a 2025 study of wearable users documented the emotional underbelly: guilt, frustration, and compulsive checking when the numbers disappoint. cardiologists report patients spiraling over their watches' heart alerts.

notice who that profile describes. perfectionist, control-seeking, symptom-googling at 1am, personally identified with performance metrics. that's the founder demographic, which is to say: the population most drawn to body dashboards is the population most vulnerable to being harmed by them.

and there's a subtler cost than anxiety: outsourced interoception. i've watched people check their wrist to find out whether they feel rested. the skill of actually reading your own state, the oldest instrument you own, atrophies like anything else you stop using. i've written about offloading thinking to machines; this is offloading feeling, and the same rule applies: offload storage, never the sensing itself.

even the world record holder walked things back

the extreme end of measurement culture has run the experiment for us. bryan johnson, the most-instrumented human alive, a man i genuinely find instructive as a frontier experiment, publicly quit rapamycin after years of dosing, citing side effects and new evidence suggesting it might have been accelerating the very aging markers he tracks. this year he cut his supplement protocol from over a hundred pills to around thirty.

read that trajectory honestly: the most measured man on earth, with the best data pipeline in the field, keeps discovering that interventions he was sure about, backed by his own dashboards, were net negative. that's not an indictment of him; walking things back publicly is the most scientific thing he does. it's an indictment of the certainty the dashboards sell. if his telemetry couldn't protect him from false positives, your ring's readiness score is not the oracle you're treating it as.

measure to intervene, or don't measure

i'm against unactionable measurement, because in a system where observation has costs, a metric must earn its anxiety.

the filter that survives: every metric you track needs a pre-committed if-then. if resting heart rate trends up for a week, i cut training load and look at alcohol and stress. if sleep duration trends down, i audit what's stealing the hour. trend, decision, action. that's instrumentation.

what fails the filter: the nightly staging readout (you have no deep-sleep lever to pull, and the number is a third wrong anyway), the readiness score before a day whose shape you can't change, the glucose patch on a healthy metabolism. those are anxiety with a subscription fee.

three rules, if you keep the devices:

• trust trends, never nights. weekly averages of duration and timing are the only numbers these devices deliver reliably. single-night scores are noise wearing a font.
• no morning verdicts. check the data weekly, at a set time, like a real review. a score you read before coffee becomes a mood, and the mood becomes the day.
• calibrate the instrument you were born with. before looking at the number, write down how you feel. if your guesses and the trend agree for a month, you didn't need the device. if they never agree, either the device is wrong or your interoception needs the reps more than your dashboard does.

the founder version of health is not maximum telemetry. it's a body you can hear without electronics, checked against honest instruments, feeding decisions you actually make. everything beyond that is a dashboard optimizing you.