Back to the map
SuperhumanHealth AI  ·  A Day in the Loop
The experiment Does a 25-minute early-afternoon nap actually improve my recovery?
Meet MIST · A Day in the Loop

We showed you how it's built. Here's what it feels like.

Meet MIST — a health agent that knows your data and the evidence, and is honest about both. Follow one real day, and one real experiment, as I dogfood it on myself.

weights + context
The hazy memory and the exact one, from the architecture pages. This is that idea, alive.
Beat 1 · Day 0 — design

It doesn't hand me advice. It designs a test.

Before anything happens, MIST proposes a real two-week n=1 trial: a clear hypothesis, what we'll measure, what we hold constant, and the success criteria — set before we start, so I can't move the goalposts later.

Day 0 · evening · home office
Who
Me — the founder, dogfooding MIST on myself.
State
Curious but skeptical: "do naps even work for me?"
Doing
Reading MIST's proposed 2-week trial before I commit.
The point
Lock the success bar now, so I can't fudge it later.
MIST· your data + the evidenceonline
You keep asking if naps help. Let's actually find out — for you, not in general. Here's a test I'd trust.
Two weeks. I'll set the bar for "real" now, so neither of us can fudge it later.
complexity-resilience model confidence we'll find out
Experiment spec n=1 · 14 days
Hypothesis
A 25-min early-afternoon nap improves next-morning recovery.
Readout
HRV-based recovery (objective) + how I feel (subjective 1–5)
Held constant
Caffeine cutoff 12:00 · protein target each day · no other new variables
Success criteria
≥ 10% recovery lift on nap days, holding across ≥ 8 of 14 days — decided now, not after
Generic wellness app

"Naps are great for you! 😴 Tap to log your nap and keep your streak going."

MIST

"Here's a test that could actually tell you — for you. Two weeks. Here's exactly how we'll know if it's real."

It doesn't give you advice. It helps you find out.

Beat 2 · Day 3 — 7:42 am

It reads the night honestly — then asks how I actually feel.

The wearable gives the objective layer. But the number isn't the whole story, so MIST asks for the subjective one and fuses both into today's plan.

Day 3 · 7:42 am · bedroom, still in bed
Slept
6h41 · HRV 48ms · resting HR 54 — a notch below baseline.
Feeling
3/5. Wired, slept badly.
Ahead
A 2pm review I'm dreading.
Doing
Telling MIST the human half it can't read off a wrist.

MIST never pretends a good HRV means a good day, or a bad one means I'm broken. It holds both: the signal and the self-report. And it knows where its job ends.

The line we drew

An outlet — not a therapist. MIST listens and supports, and it will never diagnose you or play clinician. We drew that line on purpose.

MIST· morning check-inonline
Last night: 6h41 asleep, HRV 48ms, resting HR 54. Objectively a notch below your baseline.
Before I plan the day — how's your energy and mood, 1 to 5? And anything on your mind is fair game.
me3/5. Bit wired, slept badly, big review at 2pm I'm dreading.
Noted — I hear the dread, that's real. I won't try to fix your feelings; I'm not your therapist. What I can do is build today around that 2pm and protect your recovery. Want me to?
wearable: sleep + HRV your self-report

Two kinds of data, one honest read. It fuses the wearable and the human.

Beat 3 · Day 3 — 1:05 pm

It acts in the world. It never acts behind my back.

Today the nap is the intervention being tested, so MIST defends it — and keeps nutrition controlled so it can't skew the readout. But every real-world action is a proposal I approve, never a unilateral move.

25:00
Day 3 · 1:05 pm · at the office
Situation
A clean 25-min nap window — but it collides with a sync.
The move
I approve shifting the sync to 3:15; lunch is the 45g-protein bowl.
Why
The nap is the intervention; constant nutrition keeps it clean.
Rule
Nothing happens until I say yes.
MIST· planning your afternoononline
I found a clean 25-min window at 1:30 — but it collides with your sync. To protect the nap (it's the thing we're testing), I'd move the sync to 3:15. Your call.
For lunch I'd order the chicken-rice bowl — it hits today's 45g protein target and keeps nutrition constant, so it won't pollute the experiment.
protocol: protein held constantacts only on approval
MIST proposes → you approve nothing happens until you say yes
Move "Eng sync" 1:30 → 3:15protects the 25-min nap window Approved — moved & nap protected
Order chicken-rice bowl · 45g proteinkeeps nutrition constant for the trial Approved — ordered

It acts in the world — but you're always in the loop.

Beat 4 · Day 14 — the readout

Two weeks later, it tells me the truth — including what it doesn't know.

Day 14 · the readout
Data
14 days, nap days vs non-nap recovery.
Result
~8% better on nap days — but noisy, weekends confounded.
Verdict
Plausible, not proven (moderate confidence).
Next
MIST offers experiment v2 to tighten it.
MIST· experiment readoutonline
Moderate signal — recovery looks about 8% better on nap days. But the sample's noisy and weekends are confounded.
From the complexity-resilience model this is plausible — but I wouldn't call it proven. Here's experiment v2 to tighten it: control weekends, extend to 4 weeks.
complexity-resilience model confidence moderate · not proven
Morning recovery: nap days vs not+8% · noisy
no nap nap whiskers overlap → not conclusive

Real numbers, honestly drawn. The bars differ, but the error ranges overlap — exactly why MIST says "plausible," not "proven."

"Honest uncertainty is the product."
We win the same way the architecture does — by showing the work, not by sounding certain. A clean fake win would have been easy. This is better.
Beat 5 · ongoing — the flywheel

The result doesn't end the story. It seeds the next one.

What we learned updates my personal protocol → which sets tomorrow's smarter defaults → which generates fresh data → which seeds the next experiment. That loop, grounded in your own data and a pluralistic evidence base, is the whole thing.

Ongoing · the loop, on me
Updated
My protocol nudges toward an early-afternoon nap.
Tomorrow
Smarter defaults — still proposed, still approved.
Seeds
Experiment v2: control weekends, run 4 weeks.
The idea
Learn → Act → Data → Learn, tuned to me.
LEARN ACT DATA your loop
01

Learn — run the experiment

A rigorous n=1 trial with honest success criteria.

↳ today: the 2-week nap test & its readout
02

Act — execute the day

Smarter defaults, proposed and approved, that respect the protocol.

↳ today: protected the nap, controlled nutrition
03

Data — fresh signal

Every action you take becomes the input for the next question.

↳ today: HRV + self-report feeding v2

Most apps tell you what to do. MIST helps you find out what's true for you — and never pretends to be more sure than it is.