The only honest way to know if something actually helps you
A shorter gap before your first phone pickup might track lower recovery - but the arrow could point either way, and a bad night could be causing both. No amount of watching settles it. A coin flip does.
- Watching your data can show that two things move together. It cannot, alone, tell you which caused which.
- Reverse causation and a hidden common cause make most personal correlations impossible to interpret from observation.
- A small randomized experiment - some days on, some days off, decided by a coin flip - is the honest fix.
- An honest null ("for you, this changed nothing") is a real and useful result.
Where the honesty problem starts
Say your data shows that on mornings you reach for your phone sooner, your recovery reads a little lower. It is tempting to conclude that grabbing the phone hurt your recovery. But watching cannot prove that - there are at least three explanations and they look identical in the data.
One: the phone genuinely nudged your recovery down. Two: it is the reverse - you woke up already unsettled, so you reached for the phone sooner. Three: a hidden common cause - a short, broken night made you both wake anxious and recover poorly. Same correlation, three different truths, and no way to tell them apart by looking.
- Behaviour caused the outcome.
- The outcome caused the behaviour (reverse).
- A third thing caused both (confound).
The fix is a coin flip
The way science settles this is randomization, and it works just as well for one person. Instead of only watching, the model decides at random - some eligible days it nudges the change, some days it does not - and then measures what follows.
That single move does something observation never can: it breaks the confound and fixes the direction. Because the coin, not your mood, decided which days got the nudge, the two groups of days are comparable, and the difference between them is a real estimate of the effect - for you.
“Watching shows that things go together. A coin flip shows whether one helps.”
What a good self-experiment looks like
A trustworthy n-of-1 experiment is small and safe by design. It changes one thing, it never touches anything that could harm you (no one is going to randomize your sleep away), and it measures a clear, near-term outcome rather than a vague feeling weeks later.
It also tracks whether you actually did it. Adherence is where personal experiments quietly fail - if you only followed the plan half the time, the result is about that half. Sarenica records the coin flip, the delivery, and whether it landed, so the estimate is honest about what really happened.
The result you are allowed to trust
At the end, you get an estimate of what the change actually did for you - with honest uncertainty around it, and framed as the effect of trying the change, not a promise. Sometimes that estimate is a clean nothing: for you, this did not move the needle. That is not a failure of the experiment. It is one of the most useful answers it can give, because it stops you spending effort on something that was never working.
This is the line between the two things this whole product is careful to keep separate: "these go together in your data" and "this helps you." Only a randomized experiment earns the second sentence - and earning it honestly, for one person at a time, is the point.
