The blog I expected to take a few days

I expected this blog to be a few days of work. Twenty-eight days after the first commit it has one published post.

The building was the easy part. What sits behind it now is a loop: something that drafts only from raw material I supply, a critic that returns objections and is forbidden from rewriting a single line, and a fact-checker that traces every number in a draft back to where it came from. All of that worked. On my own reckoning it is faster than writing the posts by hand and better than what I would have produced, because it takes the tedious work away and it does more than I would have done.

What it does not take is the half with my name on it — deciding what I am happy to publish, and whether it is true.

That half is not small. The rules and the skills that govern how a post gets written run to 10,873 words. The one published post is 2,557. Before either of those mattered I had to hand over 6,552 words of raw material: what happened, in what order, with the numbers and the logs. A drafter that invents nothing can only arrange what it is given.

Where that supply runs out, the machinery stops rather than guesses. The second post I started is still where it stopped. It was drafted the day after the site went up, it runs to 1,614 words, and it carries thirteen open questions. Its central section, the one the format makes mandatory, is this in full:

## What I tried first

[[TK: this section is the post. How many versions of the written process were there before the
checks moved into the pipeline, and over what period?]]

[[TK: what was each attempt, in the order it happened? A page in the wiki, a pull request template, a
runbook, a checklist in the PR description, an AI-assistant instruction file — which of these, and in
what order?]]

[[TK: what was the reasoning at the time for each one? Not the hindsight reasoning — what made it
look like the right next move when it was made?]]

[[TK: what did each attempt cost? Time to write, time to maintain, or a specific incident it failed
to prevent.]]

[[TK: did any of them make things worse? A longer document that got read less, or a checklist so
detailed people started copy-pasting the ticks?]]

[[TK: what was the moment it became clear that documenting it harder was not going to work?]]

Six questions and no prose. Every one of them is answerable, and every one of them is answerable only by me.

The catch is not the iterating. It is that iterating needs direction, and direction competes with everything else that wants attention. That post has sat untouched for twenty-seven days. In the same twenty-seven days, forty-five commits landed on this repository across seven working days, including the drafting and publishing of a different post. I was not away. This blog is not important, so it was easy to side-line.