Mood tracking that actually tells you something

A mood score compresses an entire day into one value, and the compression throws away precisely the part you would need to act on. "Tuesday: 3/10" tells you Tuesday was bad. It does not tell you whether it was bad because of the meeting, the sleep, the argument, or the fact that you had a low-grade cold you have since forgotten entirely.

And you will forget. That is the whole problem. Within about ten days the context evaporates, and you are left holding a chart that you can only interpret by guessing — which means you will interpret it according to whatever story you already believed about yourself. Mood data, uncoupled from context, mostly confirms priors.

The fix is unglamorous: record two or three sentences alongside the score, at the moment you record it. Not a diary entry — a caption. What happened, and what you made of it.

This converts every point on the chart into something you can click. Six months later, "bad Tuesday" is one click from the reason, in your own words, written before you knew how the week would end. That is why mood tracking in Compline is a property of an entry rather than a separate feature — a mood log and a journal are the same artefact, and splitting them destroys most of the value of both.

Given all that, the feature list a mood tracker actually needs looks different from the one most of them ship. It needs the mood entry and the writing to be the same action, not two taps in two different screens, because friction between them is exactly what causes people to log the number and skip the sentence. It needs the writing searchable by mood later — "show me every low day this year" has to be one query, not an afternoon of scrolling.

It needs to resist the urge to summarise too early. A weekly average smooths away the specific bad Tuesday that was actually informative, in exchange for a number that looks tidier on a dashboard. And it needs to show you the entry text next to the score when it shows you a pattern at all, because a pattern with no evidence attached is a claim, and a claim you cannot check is not something you should act on.

None of this is exotic. It is mostly the absence of shortcuts — the app not doing the one thing that would make the chart look more finished and the data less honest.

Cycles you cannot feel. Weekly rhythms, monthly ones, seasonal drift. These are invisible from the inside because you only ever experience one point at a time, and they are the clearest win in the whole exercise. Discovering that your Sunday evenings have been consistently poor for two years reframes a lot of Sunday evenings.

The gap between remembered and recorded. Pick a fortnight you remember as uniformly awful and count the good days in it. There are almost always three or four. This is the single most reliably useful thing mood data does, and it works from month one.

Associations with what you do. Which habits move together with how you feel, across enough days to mean anything. Note the phrasing: move together. Not cause.

When an app tells you "your mood is 30% higher in weeks you exercise", the arrow is genuinely ambiguous. Exercising may lift your mood. Feeling better may be what got you out of the door. Some third thing — sleep, workload, a person being away — may be driving both.

This matters practically, not just pedantically, because the two readings imply opposite actions. Under the first, forcing yourself to run during a bad week helps. Under the second, it adds a failure to a week that was already hard. Any tool that reports the correlation as though it were a cause is handing you a plan built on a coin flip, which is why Compline states sample size and stops short of claiming causation.

The way you resolve it is not statistics. It is reading the entries — the ones you wrote at the time, before you had a theory. Which is, again, why the writing has to be attached to the number.

Say six months of Sundays cluster low — nothing dramatic, just consistently a point or two under the weekly average, every single week. On its own that observation is nearly useless: a fact about a day of the week, filed and forgotten.

The entries change that. Reading the Sunday-tagged ones together shows they are not about Sundays at all — most of them mention Monday, specifically the first meeting of the week, three or four sentences before the mood tap. The pattern was never "Sundays are bad." It was "the anticipation of one specific Monday meeting reliably ruins the day before it," which is a completely different, and completely actionable, fact. Move the meeting, or change how you prepare for it, and the low Sundays should follow — a prediction you can now go and test, rather than a mood you can only endure.

None of that is visible in the ninety-day chart. The dip is there, but the chart cannot tell you it is caused by a meeting instead of, say, a standing Sunday habit you have grown to dread, or nothing at all — noise the eye insists on organising into a shape. Only the entries, read together, can tell the difference, which is the entire argument for attaching writing to the number in the first place rather than treating mood as a metric that stands on its own.

Track the mood, write two sentences, and leave it alone for a month. Do not analyse a fortnight of data; there is nothing in it and the noise will convince you otherwise. Once you have a season's worth, look for the cycles first, the remembered-versus-recorded gap second, and treat everything else as a hypothesis to check against the entries themselves.

A mood tracker that tells you something is just a journal with the dates lined up. The chart is the index, not the insight.