Missing one day: what the evidence says

Phillippa Lally's 2010 paper is the one people cite when they want a number of days. It is also the one they skip when they want permission to catastrophise a blank square. In that study, people practised a daily eating, drinking, or activity behaviour in a consistent context and rated how automatic it felt. Automaticity rose along a curve. The average time to a plateau was on the order of two months, with a wide range. A missed day, in their models, did not throw the curve back to zero.

That last sentence is the whole argument against treating a streak like a fuse. Learning accumulated. A gap was a dip, not an explosion. If your app behaves as if one miss erases the person you were on Tuesday, the app is using a different theory of change than the study you keep quoting in group chats.

What the 2010 study did and did not claim

The participants were not training for a marathon under a coach. They were building ordinary actions, self-reported, in daily life. The outcome was a questionnaire measure of automaticity, not a brain scan and not a lifetime follow-up. Lally and colleagues were careful: habits form at different speeds, some people never reach a high automaticity score in the window they watched, and the famous "66 days" figure is a mean, not a law.

What the paper still gives you is a shape. Repetition in a stable context predicted a rise in automaticity. Missing an opportunity did not, on average, produce a collapse. That is the opposite of how a streak UI is built. A streak UI is built like a game life: one hit and the run ends. Game lives are fun because they are brittle. Habit learning is not supposed to be brittle.

Wood and Neal describe habits as associations between contexts and responses that take over when the cue is present, often without a fresh burst of goal-directed thought. A missed Tuesday does not delete the association between "after lunch" and "walk." It simply fails to add another repetition. Frequency still matters. Magical immunity to future misses is not what anyone found. The claim is narrower and more useful: one miss is not a personality event, and it is not a full unlearning.

Why your calendar still screams

Calendars and chains are self-monitoring tools. Self-monitoring changes behaviour partly because it makes the behaviour visible. The same visibility makes absence vivid. A filled month looks like a pattern. One hole looks like a stain. Humans overweight stains. That is not a published "streak shame trial" with a DOI you should invent. It is ordinary loss framing plus a UI that stores identity in consecutive integers.

Charles Duhigg, writing for a popular audience, retold the cue-routine-reward story in a way that made loops feel mechanical. Mechanics are comforting until the machine shows a red X. Then people assume the loop is broken. In the lab-adjacent sense Wood and Neal use, the loop is a tendency, not a sealed circuit. Tendencies survive a skip. Motivation stories often do not.

Gollwitzer's implementation intentions are one way to make the next day less dependent on mood: "If it is 12:40, then I walk to the gate and back." The plan does not care that yesterday was empty. Fogg's tiny version of the same idea is to lower the bar on the return day so pride cannot veto the repetition that actually updates the association.

Consecutive days versus a learning curve

Streaks measure consecutive calendar days that met a threshold you chose. Automaticity measures how little deliberation the action takes when the cue appears. Those can move together for a while. They diverge the moment your life includes travel, illness, or a Sunday that never contained the cue. Lally's participants who missed a day were not, on average, sent back to the start of the curve. Your streak was. That mismatch is the product design problem, not a moral one.

If you want a number that survives a miss, you want something like a rolling completion rate, or a count of repetitions per week, not a single unbroken integer. The integer is a motivational toy. Toys are allowed. They should not be mistaken for the science.

Claim you hear	What evidence supports	What it does not support
One miss ruins the habit	Almost nothing in Lally 2010	Treating a skip as unlearning
You must restart at day one	Streak software rules	Automaticity scores
Missing often still works	Frequency still trains the cue	Infinite skips with no context
66 days and you are done	A mean time to plateau in one study	A guarantee, or a finish line

A practical reading: protect the cue more than the integer. If the cue will not exist tomorrow (you are on a plane, you are in a different kitchen), write a replacement if-then tonight, not a punishment routine. If the cue will exist, do a smaller instance and log it. Partial credit, where an app offers it, is a way to record a reduced session without pretending the day was full or that it was nothing. Habit AI is one place that idea shows up as a product choice. The evidence still sits in the repetition, not in the badge.

None of this says skips are free. A week of skips is a different dataset from one skip. Patterns matter. Personality lectures do not. Count the week. Do not hold a funeral for Tuesday.

Read next

After the miss comes the morning: what to do the morning after a broken streak. For the number people confuse with a law, see how long it actually takes to form a habit. Product context: Habit AI.