Streak lengths are not normally distributed, and that changes the goal

Take a year of habit data, split it into runs of consecutive completed days, and plot how long the runs are. The shape is always the same: a huge pile of very short streaks, then a rapidly thinning tail, then one or two long ones far out on the right. It is closer to a geometric distribution than anything symmetric, which is what you get when each day has a roughly fixed chance of breaking the chain.

That shape has a consequence people get wrong. Your longest streak is not a measure of your typical behaviour, it is the maximum of a skewed sample, and maxima are noisy. Beating a personal best of 141 days is mostly a question of running the experiment for long enough, not of being more disciplined than you were during the 141.

The number that does describe your behaviour is the daily break probability, and it falls straight out of the same data. If you complete 85 percent of days and misses are roughly independent, your median streak is short and your expected longest streak in a year is calculable. It usually lands close to what you actually got, which is a mildly deflating but useful result.

This page shows how to compute the distribution from an export, what the shape means, why streak records are a bad target, and what to aim at instead.

Outline

Status: outline. The introduction above is finished copy. Everything below is the section plan for the full draft, targeted at about 1 500 words.

1. Computing the distribution. From a CSV export to a histogram, in ten lines.
2. Why the shape is geometric. Independent misses and a constant break probability.
3. The longest streak is a maximum. Why it is noisy and mostly measures elapsed time.
4. Break probability. The number that actually describes you, and how to get it.
5. When misses are not independent. Clustering, and what it says about the cue.
6. A better target. What to aim at instead of a record.