← Blog

The Interstitial Journal blog · Built on evidence, not hype

What Your Journal Patterns Can and Can't Tell You

Open the Insights screen after a few weeks of entries and something will look like a finding: a cluster of entries late in the afternoon, one tag suddenly dominating, a mood that dipped on the same days as a particular kind of meeting. The pull is to read it as a fact about you. The research on how people read small samples says to slow down — not because the pattern is meaningless, but because it is a better question than it is an answer.

By comun.ai · 21 September 2026 · ~6 min read

What the screen actually shows

To be precise about the thing we are discussing: the app's Insights screen shows summary tiles (entries, active days, focus time), a times-of-day chart of when you write, entries broken down by tag, and mood over time, across the last week, the last month, or everything. All of it is computed on your device from your own entries. It is a mirror of your log. Nothing on it is a score, a prediction, or an assessment.

A mirror is only as good as what stands in front of it, and a personal log has three properties that make it easy to over-read: the numbers are small, you chose when to write, and the act of writing is part of the data.

Small numbers wobble more than intuition expects

Amos Tversky and Daniel Kahneman found that even researchers trained in statistics expected small samples to resemble the population they came from far more closely than chance allows — what they called the belief in the law of small numbers.¹ Strong The statistics are not in dispute; the intuition that ignores them is a well-documented bias.

A week of entries is a small sample. If a tag doubles from one week to the next, that can be a real shift, or it can be the ordinary wobble of a handful of counts. The same view across a month, or all time, is steadier simply because there is more in it. The week view is useful for remembering what happened; it is weak evidence for what usually happens.

Extreme weeks are usually followed by less extreme ones

Regression to the mean is a statistical certainty rather than a psychological effect: when a measure varies partly by chance, an unusually high or low reading tends to be followed by one closer to the average, with nothing having caused the change.² Strong Tversky and Kahneman's classic example involved flight instructors who concluded that praise made trainees worse and criticism made them better, when what they were seeing was mostly regression after unusually good and bad landings.³

You will see the pattern you expected to see

Loren and Jean Chapman gave observers materials in which the features they believed went together had been deliberately paired at random. The observers still reported the relationships they expected — an illusory correlation.⁴ Moderate The study was about clinical judgement, not journals, but the mechanism generalises uncomfortably well: if you already suspect that a certain project drains you, the entries that fit will be the ones you remember.

And noticing is not causation. Two tags that rise together in the same fortnight may share a cause — a deadline, a season, a manager's calendar — rather than one driving the other.

Your log is a participatory instrument

You write when there is a seam to write in, and you probably write more when unsettled than when absorbed. So the times-of-day chart shows when you write, which is related to, but not the same as, when you work well. Recording can also change what is recorded; we covered measurement reactivity and experience sampling in What a Mood Tag Is, and What It Isn't, and will not repeat it here.

Research on self-tracking describes this terrain well. Ian Li, Anind Dey and Jodi Forlizzi's stage-based model separates collecting data from reflecting on it, and found that problems in one stage cascade into the next.⁵ Early Daniel Epstein and colleagues' lived informatics model adds that people routinely lapse and resume tracking, so real logs have gaps.⁶ Early Eun Kyoung Choe and colleagues, studying people who presented their own self-tracking projects, described pitfalls that include tracking too many things, not capturing the context around the numbers, and drawing conclusions without the rigour an experiment would need.⁷ Early These are descriptive studies of how people track — useful maps, not outcome trials.

How to read the screen, then

Start with the longest range. Check the all-time view before you trust a week. If a pattern only exists in seven days of data, treat it as weather, not climate.

Ask "compared to what?" A heavy afternoon in the times-of-day chart may mean your afternoons have more seams, not that they are harder.

Discount the rebound. After an unusual week, expect the next one to look more normal on its own. Do not credit the change you made until it has held for a while.

Turn the pattern into a question, then write the question down.

Where the evidence stands

What this means for this app

This is why the Insights screen stays deliberately plain: counts, times, tags, and mood over time, with no score, no grade, and no streak to protect. It does not tell you what a pattern means, because the data cannot support that, and it stays on your device, because the only person entitled to interpret it is you.

The calibrated one-sentence version: a pattern in your own log is a good reason to ask a question and a poor reason to reach a conclusion.

This is reflection, not treatment. A mood trend in an app is not a clinical measure; if something in your entries worries you, that is worth raising with a qualified professional. More on our science page, how your notes stay private on our privacy page, and the method itself in the interstitial journaling guide.