I Kept Building and Stopped Writing It Down

I made dump because I kept learning things and then letting them disappear.

A strange bug would make me understand something differently.

A small project would become much less small.

I would spend an evening following one question into several worse questions, finally reach an answer and then move on without writing any of it down.

So I built a place specifically for those stories.

Then I spent roughly two weeks building things and stopped writing.

Excellent system.

01.There was no shortage of material

Sweep kept growing beyond the tiny file-moving script I first wrote about.

jobs-mcp became increasingly serious engineering work, despite being a private tool that began with a fairly ordinary problem. At one point I managed to give different parts of one operation slightly different interpretations of the same deadline, which is a very efficient way to discover that I did not understand time budgets as well as I thought.

I also started experimenting with deliberately programming without AI assistance.

Not because I suddenly think AI is evil.

I wanted to notice what changes when I cannot hand the confusing part to something else. Where do I slow down? Which documentation do I read? Which parts do I only think I understand because I am used to receiving a plausible next step?

Then I started CS50 while already studying Software Engineering and already building software.

Apparently the reasonable response to having several unfinished projects is to begin an introductory computer science course.

None of this left me with nothing to write about.

It left me with too many possible starting points.

01.1I kept waiting for the finished version

Every time I considered writing something, the work still had another obvious step.

Sweep had more rules to define.

The deadline problem was fixed, but it belonged to a larger set of retry, cancellation and pacing decisions.

The experiment without AI had barely started.

CS50 was still new enough that I had more first impressions than conclusions.

I kept thinking I would write after the next part was done.

Then the next part created another part.

This is not particularly surprising. Most of the things I build are unfinished in some meaningful way. Orbit still does not schedule anything. Yanawa still cannot compile anything. Sweep is not wandering around my filesystem unsupervised, which remains good news for my files.

If I wait until a project has nothing left to ask me, I will either never write about it or I will only write after I have forgotten what made the problem interesting.

The existing posts already knew this.

The person who was supposed to write them apparently needed a reminder.

01.2This is not a changelog

The answer is probably not to document every commit.

I do not want dump to become a list of repository activity with paragraphs around it. Git is already very good at remembering that I changed a file. It is less good at remembering why a tiny decision bothered me for an afternoon or which assumption finally stopped making sense.

Those are the parts I want to keep.

The moment a personal script starts needing an interface.

The moment one deadline turns out to be several.

The moment beginner material finds a gap underneath something I already know how to build.

They do not need the rest of the project to be finished.

They only need enough of a before and an after that I can tell what changed in my head.

So this is not the complete account of the last two weeks.

It is just me writing something down again before the next interesting problem convinces me to wait.