What 1 204 logs told me about my own weekends
Reading the weekday bias chart, and what to change because of it.
I exported my own data after fourteen months of tracking: 1 204 logged entries across five habits. The number I expected to look at was the completion rate. The number that turned out to matter was the weekday one.
The chart
Completion by day of week, all habits combined, over fourteen months:
| Day | Completion |
|---|---|
| Monday | 82% |
| Tuesday | 94% |
| Wednesday | 88% |
| Thursday | 76% |
| Friday | 70% |
| Saturday | 44% |
| Sunday | 52% |
The weekday average is 82%. The weekend average is 48%. That is not a dip, it is a different person.
Two smaller things fall out of the same chart. Thursday and Friday slide well before the weekend arrives, which suggests the decline starts on Thursday afternoon rather than Saturday morning. And Sunday is better than Saturday, consistently, by eight points.
The explanation I wanted, and the one the data supports
The story I told myself was rest. Weekends are for recovery, the drop is intentional, no action needed.
The time-of-day chart disagreed. On weekdays my logging peaks at 10:00 and again at 22:00. On weekends there is no peak at all, just a low flat smear from 11:00 to 23:00. That is not a person resting on purpose. That is a person with no structure, getting to some of it eventually.
The correlation view finished the argument. Reading is 34% more likely on days that started with meditation, and meditation is anchored to “kettle on”, which is a workday event. On Saturday I make coffee at a different time, in a different order, sometimes not at all. The anchor never fires, so the first habit never fires, so the chain of habits behind it never starts.
The drop was not about willpower or rest. It was one missing cue, and every habit downstream of it went with it.
What I changed
One thing, because changing several at once teaches you nothing.
I gave the morning block a second, weekend-only trigger: the walk. On Saturday and Sunday I leave the flat before I do anything else, and the block is anchored to coming back through the door rather than to the kettle. Same four habits, same order, different event.
I also cut the weekend target on the reading habit from 20 pages to 10. Not because ten is enough, but because a target I hit at the weekend produces an overshoot on Tuesday that banks a shield, whereas a target I miss produces nothing except a red square.
What happened over the next eight weeks
| Day | Before | After |
|---|---|---|
| Saturday | 44% | 71% |
| Sunday | 52% | 74% |
Weekday numbers moved by less than two points in either direction, which is the part I care about most: the change added structure to two days without borrowing anything from the other five.
Thursday and Friday did not improve. That is a separate problem with a separate cause, and I have not solved it yet. The honest version of a stats post includes the part that did not work.
How the numbers were produced
Worth stating, because a chart with no method behind it is decoration.
The export is a CSV of every entry: date, habit, value. Fourteen months, five habits, 1 204 rows where a value was logged. Days with no row are absent rather than zero, which matters: a habit scheduled three times a week should not be counted as four failures.
Completion per weekday is therefore logged days divided by scheduled days for that weekday, not divided by seven. Two habits were added in month three, so their first two months contribute nothing to either side of the ratio rather than contributing zeros.
Frozen days from break mode are excluded entirely. Twenty three days across the fourteen months were frozen, mostly in August, and counting them as misses would have made the summer look like a collapse that did not happen.
Shielded days are counted as misses in this analysis, even though the streak survived them. The shield protects the chain, not the statistics, and mixing those two would defeat the point of keeping an honest log.
How to read your own version of this chart
Three views, in this order.
Weekday bias. Look for a gap larger than 15 points between your best and worst day. A gap that size is almost always a missing cue rather than a missing virtue. Ask what happens on your good days that does not happen on your bad ones, and be specific about the event.
Time of day. A sharp peak means an anchor is working. A flat smear means the habit is happening whenever you remember, which is a habit waiting to be lost. Compare your weekday and weekend curves separately, because averaging them hides exactly what you are looking for.
Correlations. Find the habit that other habits depend on. That one is load bearing and it belongs at the top of your block. Protect it before you protect anything else.
One caution: fourteen months of one person’s data is an anecdote with a chart attached. Read your own numbers before you copy my conclusions, and be careful about reading anything into a difference of three or four points.
Getting the data out
Everything above came from a CSV export. In miniHabits that is one tap and it is in the free tier, because a tracker that will not give you your own history is a tracker you cannot audit.
The free heatmap tool takes the same CSV and renders the year as an image you can save, with no signup and no server call. If you want to check the shape of your data before committing to anything, paste a year in and look at it.
What I am looking at next
Two questions the current data cannot answer.
Thursday. The slide starts there and no cue explains it, because Thursday morning is identical to Tuesday morning. My guess is that it is a consequence of the week rather than a property of the day, which means the interesting variable is what happened on Monday and Tuesday. Testing that needs the correlation view to handle more than two habits at once, which is being built.
Sleep. Steps and sleep both come from HealthKit and both sit in the same log, so the question “does a short night predict a missed morning” is answerable with data I already have. I have not run it, partly because I suspect the answer will be inconvenient.
Neither of those changes the advice above. They are the reason to keep exporting.
The general lesson
The completion rate tells you how you are doing. It does not tell you anything you can act on.
The weekday chart, the time-of-day chart and the correlation list all answer a different question: where in your week does the structure disappear. That question has a fix attached, and the fix is usually one cue in one place rather than more effort spread evenly.
If you want the block structure that all of this hangs on, it is in the morning routine guide. If your weekend collapse looks less like a missing cue and more like a missing brain-day, the ADHD guide covers that ground more carefully.