August 2, 2026 · 2 min read
On my current programme I have written more than forty-seven standard operating procedures from nothing. Not improved, not consolidated — written, because none existed.
Most of what I now believe about documentation I learned by getting it wrong in the first ten.
The first documents I wrote were accurate. They described what the platform did, what the constraints were, what a competent operator should do about them. They were also close to useless, and I did not find out until someone in another country tried to use one.
They were accurate to me. Every one of them had a gap my own knowledge silently filled: an assumed default, an unnamed prerequisite, a step that only makes sense if you already know why it is there. I could not see the gaps because I was the one thing making them invisible.
That is the difference between an artefact and an interface. An artefact is a record of what you know. An interface is something another person can operate without you. Only one of them is worth the time it takes to write.
The rule I ended up with is blunt: can someone in another country follow this without asking me?
The important word is while. Applied at the end, that test produces a review comment. Applied during writing, it changes what you write — you stop describing and start instructing, because you are picturing a specific person who cannot ask a follow-up question and is on a different clock to yours.
Anything that failed the test got rewritten. Not annotated, not appended to. A procedure with a clarification bolted onto the end is a procedure that will be misread again, just somewhere else.
The economics are lopsided. A gap found while writing costs a paragraph. The same gap found during an audit costs a finding, a remediation, and a conversation about whether the operating model is real.
There is a subtler cost too. Documentation written to be filed reads differently from documentation written to be used, and operators can tell within a page which one they are holding. The first kind quietly teaches them that documents are not where answers live. Once a team believes that, they ask people instead — which is exactly the dependency the documents were meant to remove.
The measure I trust is not how many documents exist. It is whether the receiving organisation signed off the training built on them — in this case, in all three countries — and then ran the platform without me.
Everything before that point is a hypothesis. Forty-seven procedures nobody has followed are forty-seven hypotheses with a version number.
Write for the person who arrives after you have gone. Not as a courtesy — as the design constraint.
They are the only reader whose experience you cannot repair by being available. And in a programme built to be handed over, they were always the only reader who mattered.
