Most records projects start with choosing software. Ours starts with the shelf. Before anything is installed we agree where each kind of record lives, who owns it, and what happens to it at the end of its life. The software is the easy part.

The reference deployment runs on eighteen root containers. Twelve of them are departmental areas, matching the parts of an administration that actually generate and hold records. Beneath those sit eighty second-level function folders, one for each recognisable function rather than one for each person or each project.

That distinction matters more than it sounds. A folder named after a function survives the person who ran it. A folder named after a person does not. Staff turnover is the most common way a records system quietly falls apart, and structuring by function is the cheapest defence against it. Team folders in this system are built to outlast the team.

What the folder tree cannot carry, tags do. The deployment uses 34 tags across six namespaces, covering things like record type, sensitivity and status, so that a document can be found by what it is as well as by where it sits. Search is filtered by permission, so a person only ever sees results they are entitled to see.

The trade-off is time. Information architecture is a separate, priced piece of work because it takes real hours with real staff, and it happens before the appliance is racked. We would rather charge for it openly than pretend the structure appears on its own.