I counted 1,890 daily-note lines. My memory was wrong.

For six months I kept every fleeting thought in one file: a single daily note, one line each, - HH:mm then a verb then the thought. In July I exported the whole thing and counted. 1,890 lines across 181 days. Almost nothing I believed about my own days survived the count.

The short version: my mornings are not for thinking, my evenings quietly do most of the real work, and 94% of what I wrote down never became anything. The file I'd been treating as a weak knowledge base turned out to be an excellent buffer, graded on the wrong exam.

Why count your own notes at all?

I wasn't trying to optimize anything. I'd just noticed I kept saying things about how I work ("I think best in the morning," "I mostly track tasks") that I had no evidence for. And there was the evidence, six months of it, timestamped, sitting in a file I opened every day.

Counting took an afternoon. I split each line on its leading timestamp, bucketed by hour, and grouped by the first verb. No dashboard, no app, one throwaway script and a spreadsheet. The method only worked because the format made it work: every line already carried a time and a verb, so the structure I needed was a side effect of how I'd been capturing all along. Had the lines been paragraphs in a journal, none of it would have been countable. Consistency at capture time is what buys you countability later.

Here is what I found, next to what I would have told you before I looked:

What did the timestamps actually say?

The day has a shape, and it wasn't the one I pictured. Two hours, 21:00 to 22:59, held 430 of the 1,890 lines. Nearly a quarter of six months of thinking happened in a window I'd have described as winding down. The single busiest hour was 21:00 with 224 lines; 22:00 came second at 206.

Mornings were busy too, 454 lines between 07:00 and noon, but busy with a different kind of line. The 07:00–09:00 block ran on todo, call, email, reply. Logistics. The hours after midnight were almost empty: 22 lines total from midnight to 06:00, most of them the one or two nights I couldn't sleep.

Split by verb, the gap sharpened into something I couldn't wave away. Of my idea and note lines (the 620 where I was thinking rather than administering), 41% landed after 21:00. Of my todo and call lines, 9% did. Same file, same person, and I had the two halves of the clock backwards in my head.

Where does the thinking actually happen, then?

I'd been guarding my mornings for deep work and spending the evening on scraps. The count says I had it inverted. My mornings clear the queue; the thinking I claim to value shows up late, compressed, after the day's obligations are spent. Once I saw it I could feel it was true. The 21:00 lines are the ones with questions in them, half-formed, the ones actually worth keeping.

The averages misled me in a quieter way. A packed day, I'd assumed, produced more lines — loosely true and mostly useless. The mean was 11.3 lines a logged day; the median was 9. A handful of incident days, one shipping day at 39 lines, drag the average up past any ordinary day. If I'd only ever looked at the mean, I'd have pictured a life that basically never happened. The median is the honest number, and it's the one I'd never bothered to compute.

The 94% nobody warns you about

Here is the finding that reorganized how I think about the whole setup. Of 1,890 lines, 113 (6%) ever became something else: promoted to their own file in the vault, turned into a task I actually finished, folded into a piece of writing. The other 1,777 sat there dated and untouched forever.

My first reaction was that 94% is a failure rate. Then I sat with it. Those lines were never meant to graduate. They were meant to get out of my head so I could keep moving, and they did that the instant I typed them. Capture was the entire job. I'd been judging a buffer as if it were a library. This is the flip side of something I argued a couple of weeks back, that the worst moment to organize a note is the moment you take it: if organizing can always wait, it turns out that most of the time it waits forever, and the line is no worse for it.

The number I don't trust

One caveat I have to say plainly, because the tidy 41% flatters me. I only counted what I bothered to write. The thoughts that were too fast, too dull, or too half-formed to log are invisible to the spreadsheet, so "41% of my ideas arrive after 21:00" honestly means "41% of my logged ideas." It's entirely possible my mornings are full of thinking I never capture because I'm heads-down doing it. The count corrects my memory, but it can't see its own blind spot. I trust the shape of the day more than any single percentage in it.

What changed after I saw the numbers?

Not much, and all of it small. I stopped scheduling anything that needs a clear head before 10:00; that slot goes to the queue on purpose now, and I feel less guilty about it. I started protecting the 21:00–22:30 window the way I used to protect mornings: no chores, no inbox, just the file open. And I quit apologizing for a messy midday, because the messy midday is exactly where logistics are supposed to live.

The count answered one question and handed me a sharper one. I know when a line gets written. I have no idea how long it then sits before I act on it — whether the 21:00 idea gets touched the next morning or rots for three weeks. So I've begun stamping a second time on the lines I finish, a small done HH:mm at the end, and in a few months I'll have a distribution for time-to-action instead of time-to-capture. My guess is it will embarrass me again. That's the whole point of measuring the thing instead of remembering it.

FAQ

How long did the counting actually take?
About an afternoon, most of it spent fixing lines that didn't match my own format. The counting was a dozen lines of script; the mess was six months of me being inconsistent with myself.

Do I need a special app for this?
No. Any log where each line starts with a timestamp will do, and plain text is plenty. The only real requirement is that you were consistent enough that a machine can split the lines cleanly.

Isn't 1,890 lines too small to conclude anything?
For strong claims, yes. I'm not publishing a study, I'm correcting my own memory, and the bar for that is just "better than a vibe." A number I can re-check beats a story I told myself, even a small number.

If you keep any kind of timestamped log, run one query before you reply: which hour holds the most of your lines, and does it match the hour you'd have said you do your best work? I'm collecting the mismatches. Mine was three hours off.

I'm a solo dev who keeps one dated line per thought; on the days I curate, those lines flow into a plain markdown vault. I write here every few days about the parts of working alone I only believe once I've counted them.