Smart Node has documentation now. Eight pages at smart-node.app/docs: architecture, install, security and data, rules and approvals, integrations, plus a changelog, a limits page, and what actually runs.
The product page talks about a crew and a front desk, because that is how the person buying it thinks about the work. An engineer reads the same page and has no idea whether there is a system underneath or one clever prompt. Two different readers, and one page cannot serve both. So now there are two.
The arithmetic was the fun part. Before writing anything I counted the internal documents on this project: 1,177 of them in six months, spread across the memory system, the document store, the node itself and the repository, plus 580 tasks I did not count because a task is a record rather than a document. Eight pages came out of that. Writing them was the easy half. Deciding what a stranger never needs to know was the other half, and it took longer.
The part I enjoyed was the architecture map. I did not draw it. There was already a generator on the node that builds the internal map from a list of boxes and a list of arrows, with a note at the top saying never to edit the output by hand. The public map is the same generator with a different list. An afternoon instead of a week, and the two cannot quietly drift apart.
A few small decisions in it I would make again. Everything inside the machine is grey and the single arrow leaving it is the only thing with colour, so your eye finds it without being told. A message can take one of four routes depending on how it arrives, so there is a switch above the map that lights one at a time instead of a single line pretending to be the whole story. And the channels that are not built yet are on the map, dashed and labelled, because leaving them off would look better and help less.
There is also a limits page. What it does not do, where it is thin, what is still an open question. That reaches a customer either way. The only thing I get to choose is whether it comes from me first.
