Telemetry — Week by week
The habits moved, and I didn't notice
The other telemetry pages average 138 days into one answer. An average hides the thing I actually wanted to know, which is whether the answer has been changing the whole time.
It has. The working day now starts 65 minutes later than it did in March and ends at the same hour it always did, the nights are 97 minutes longer, and 17 hours a week have gone somewhere. None of that was a decision.
Readings through rolled up page built
First half against second
The run split down the middle, with each half reduced to its median week. Medians rather than means, so the week I spent away from the desk doesn't drag its whole half down with it.
| Measurement | Mar–May | May–Aug | Change |
|---|---|---|---|
| Hours at the keyboard | 77.2 h/wk | 60.1 h/wk | −17 h |
| Day begins | 8:20 AM | 9:25 AM | +65 min |
| Day ends | 12:38 AM | 12:10 AM | −28 min |
| Median night | 6.9 h | 8.5 h | +97 min |
| Terminal | 44.5 h/wk | 27.7 h/wk | −16.8 h |
| Browser | 28.7 h/wk | 30.1 h/wk | +1.4 h |
| App switches | 4.42/h | 4.32/h | −0.1/h |
A fifth of the working week, gone
Hours the machine was actively in use, by week. The median week ran 77.2 hours through the spring and 60.1 since, a drop of about 22%. The dashed line is the earlier figure.
The floor is the week of Jul 13 at 29.7 hours, which was a holiday and reads like one: the same week is 94% away from home.
2 partial weeks at the ends of the run are left off every chart here. A week the recorder only saw four days of is a hole in the data, and plotting it as a short bar would make it look like a hole in the work.
Bars are weeks, beginning Monday. Hover any of them for the week's figures.
The morning slipped; the night held
First and last active reading of each day, taken as the median across the week. The start has moved from about 8:20 AM to 9:25 AM. The end has not moved at all: 12:38 AM then, 12:10 AM now.
So the shorter week is not an earlier finish. It is a later start against a fixed bedtime, which shortened the working day from 16.4 hours to 15.2 without any single day feeling different.
A day runs 05:00 to 05:00 here rather than midnight to midnight. Cutting at midnight would file a session that ran to one in the morning as the next day's very early start, and the chart would show a dawn habit I do not have.
And the hour went into sleep
The median inferred night has gone from 6.9 hours to 8.5. Almost all of it came from the morning: waking 69 minutes later, against a bedtime that moved 27 minutes earlier.
That is the same 65 minutes the working day lost off its front, arriving in a different account. The caveat matters here: sleep is inferred from a locked screen, so a week spent away from the laptop reads as an excellent week of rest. These numbers are a proxy, and the proxy is at its worst exactly when I'm travelling.
The gap between the lines is the night. Weeks with fewer than three inferred nights are left as gaps rather than guessed at.
All of it came out of the terminal
Foreground hours per week, by category. Terminal time has fallen from a median 44.5 hours a week to 27.7. Browser time went from 28.7 to 30.1, which is no change worth the word.
The whole reduction in the working week is one line on this chart. I would have guessed the time was being spread thinner across everything; it isn't, and I would not have found that out by looking at the four-month total, where the terminal is still comfortably in front.
“Everything else” is comms, editors, meetings, games and media added together. Individually they are each about an hour a week, which on this scale is a flat line at the bottom.
The day breaks in half at dinner
The same categories against hour of day, summed over the whole run. The peak is 10:00. Then the total falls away through the afternoon to a clear trough at 20:00, and climbs back to a second peak that runs past midnight.
Two working days with dinner in between, in other words, and the evening one is nearly as large as the morning one. The heat grid on the rhythm page shows the same shape as a smear of colour; stacked against a scale it is a much less flattering picture.
Total hours spent in each hour of the day across 138 days.
One number that refuses to move
Every measurement above has drifted. This one hasn't. Counting how often the foreground application changes between one reading and the next gives 4.34 switches an hour of active use, and it has sat between 3.17 and 4.88 every week since March — 4.42 in the first half against 4.32 in the second.
Fewer hours, later starts, longer nights, and the texture of the hours themselves is unchanged. Whatever moved, it wasn't that.
This is a floor, not a count. Readings are 5 minutes apart, so switching to a browser and back between two of them is invisible. The real rate is higher. It is still worth plotting, because the blind spot is the same width every week, and a trend can be read off a consistent undercount.
Where the laptop was
Share of readings taken somewhere other than my own network. It is near zero for most of the run and then 94% in the week of Jul 13, which is the one week the other charts on this page should be read with suspicion.
The steadier number underneath: the median week has the machine unplugged for 83% of readings. For a laptop that spends almost all its time at one desk, it is very rarely on mains power.