133 pages

Pioneering

Notes from active builds: the design decisions, war stories and how-it-works of whatever I'm making right now.

Three models vote on what happened

I ran one job past three AI providers and had them vote. The voting was the easy part. Deciding what counted as agreement was not.

The lock I took off

That morning I put a lock on a bucket. By teatime I'd taken it off for good, because it guarded the risk I'd already covered and locked in the ones I hadn't.

The runner fleet that scales to zero

The homelab finally ran out. Not disk this time, capacity. So the overflow went to AWS, on the condition that it costs nothing when nobody's pushing.

The bill for fifty-one modules

Change one line in the tls module and you've signed up for twelve releases. That's not a bug in the architecture, that's the architecture.

Exit 0 is the strongest claim a program can make

The job told me, very politely, that it hadn't done its job, and then exited 0. Two questions were enough to undo the spec written to fix it.

Four lines cost a whole website

To append four lines to one YAML file, the release job was cloning this entire website into memory. The first fix for it failed without making a sound.

The extraction playbook only got real when chat left

I wrote the process document before doing the hard one. Then the hard one went through it and rewrote a step.

The packages that didn't want to leave

Everyone writes up the modules that came out. The decisions that mattered were the four packages I decided to leave exactly where they were.

The builder had been archived for a year

I went looking for reproducible CI images and found the builder had been archived fourteen months earlier. Swapping it out turned out to be the easy part.

I graded every package for how badly it wants to leave

Twenty-seven packages, four marks each, out of ten. The interesting column wasn't the one I expected, and one row explained the whole programme.