the register nobody edits

A document that lists every decision a system rests on can only be trusted if it is complete, and a complete register written by hand costs a person's attention in proportion to how much has been decided. That cost is paid exactly when attention is scarcest: at the end of a long session, when the change has finally been understood and nobody wants to write it down a second time.

The tension is between complete and current. Completeness argues for writing everything down once, carefully, and then leaving it alone. Currency argues for editing the moment something moves. With a human as the writer these cannot both hold, and the usual compromise — a document maintained well enough — has no failure signal at all.

The decision was to stop writing it. The readable register is rendered from the entries the machinery already consumes: several hundred of them, grouped by the architectural scope each belongs to, each carrying the version it is at and the targets it applies to. Nobody chooses the grouping, the ordering or the formatting, because nobody writes them.

Two costs, both real. Nothing is filtered, so the register shows the whole mess, including entries that read badly out of context. And the output cannot be reviewed as a change: one amended entry can move hundreds of lines, so review has to happen on the entries and never on the page.

What it proves is that the document is a build product. Its accuracy is not a matter of anyone's diligence, which is the only kind of accuracy that survives a few hundred entries.