Ship logs, not slide decks

Visibility is a drug. Status meetings, RAG slides, a shared folder of screenshots from staging. None of that is a product. A product is what a stranger can do on Tuesday without you in the call.

We measure a week by what reached main and stayed there. If it did not land, it did not happen — no matter how coherent the midweek update sounded. That sounds harsh until you have paid for six months of theatre and inherited a service that cannot be deployed by the people who now own it.

What we write down when something ships

Every engagement we stay on has a ship log. It is not a case study. It is an internal document the next engineer can open at 2 a.m. It says what changed, why, what we refused, and what will hurt later if nobody watches it.

• The outcome a user can name in one sentence.
• The constraint that made the week interesting — a date, a load, a two-person team, a cutover that could not take the floor down.
• The decisions we declined, so nobody “discovers” them again in a new ticket.
• The runbook: how to see it breaking, who to page, what not to restart.

We publish a public version of that habit on the work page: the product, the line, the stack. The private version is longer and less flattering. Both exist so the work can outlive the original Slack thread.

Main is the record

Pull requests are cheaper than steering committees. A commit message that says why the queue depth alert exists will still be true when the original PM has a new title. A slide that said “platform modernization — green” will not.

This is also how we use AI. Models draft. Humans keep the log honest. We will not let a generated summary become the only memory of a cutover. If the model hallucinates a migration step, the ship log is where that dies — before it becomes folklore.

What this costs you

It costs a crew that can write. Not “documentation as a phase.” Writing as part of shipping: the comment on the dangerous query, the note on the feature flag, the paragraph in the log that admits the first schema was wrong. That is slower than a demo. It is faster than rewriting the system in eighteen months because nobody remembers why the join is the shape it is.

If your last vendor left a Figma file and a wiki that 404s, we should start from the repo. The log begins on the first merge.