The days you forgot to log

An experiment needs days, not intentions

HRV, resting heart rate and sleep stages move night to night for reasons that have nothing to do with what you’re testing — a hard session, a glass of wine, a bad day at work. That scatter is why a self-experiment needs two to four weeks per phase: a real effect has to have time to separate from the noise. Nine tagged days out of twenty-one isn’t a short experiment. It’s a broken one, and the honest answer it produces is a shrug.

A missing day is not a neutral day

This is the part people miss. If the days you forgot were scattered at random, you’d just have fewer of them — annoying, survivable. But you don’t forget at random. You remember to log on the days something stood out: the night you slept badly, the morning you felt sharp, the day the supplement seemed to be doing something. The days that felt unremarkable are exactly the days that go unlogged. So the sample you end up with leans toward the memorable ones, and it will tell you the change did more than it did. Fewer days weakens a result. Biased days invent one.

Coverage beats diligence

The fix isn’t to remember better. Nobody remembers better. The fix is to stop asking a daily question when what you mean is a continuous fact: you didn’t take magnesium on eleven separate occasions, you started taking magnesium. Recorded as a period with a start date, that statement is still true tomorrow morning with nothing running and nothing to remember — and it covers the unremarkable days as faithfully as the vivid ones, which is the whole point.

Backdating matters just as much, and it isn’t cheating. Realising in September that you began in early August is a fact about the past, not an edit to your data. Being able to say so gets you a month of coverage you had genuinely earned and would otherwise have lost — and a month is often the difference between a result and a shrug.

Some things you should never have to log at all

Then there’s the category that shouldn’t be your job in the first place. If you nap in the afternoon wearing a ring, the nap is already recorded — start time, length, the lot. What the ring doesn’t do is call it a nap: its own labels only separate the main night from everything else, so a real afternoon sleep carries the same tag as a twelve-minute doze on the sofa at nine in the evening.

Which is why naming every one of them a nap would be worse than naming none. Dozing off in front of the television before bed is the common case, and a label that lands on three days in four cannot correlate with anything at all — it’s just weather. The useful version reads the clock as well as the label: a daytime session, long enough to count, is a nap; the pre-bed doze is not. That distinction is what turns it from a curiosity into something your data can actually answer questions about.

What this looks like in Vitra

Both of these landed in this release. A tag can now run as a period — mark it ongoing, backdate the start to when you actually began, and every day in between counts from the moment you save it. It’s stored as a period rather than a row per day, so stopping it or fixing the start date never touches a day you tagged by hand. And naps are now read from the sessions your ring already synced, so they show up in your tag correlations and self-experiments without you logging anything. Your history fills in the moment you update, and all of it is computed on your own machine.

Frequently asked questions