3 August 2026

A lab notebook for infrastructure work

Wikis are for facts that are already settled. Incident work is the opposite: half-formed hypotheses, commands whose output mattered for ten minutes, and a decision that has to be explained tomorrow.

So I keep a lab notebook instead: one plain text file per week, in a folder under version control. Every entry starts with a timestamp and the question I was trying to answer. Then the commands, the relevant output (trimmed), and one line at the end — what I now believe, and how confident I am.

The format is deliberately ugly. Plain text greps fast, diffs cleanly, and survives every tool I've switched to in the last decade. Tables and diagrams live next to it as files, not pasted screenshots.

The habit that made it work was writing the question first. Without it, an entry becomes a transcript of typing rather than a record of thinking, and six months later it answers nothing.