The database ecology, family by family.

The WORKSPACES catalog is an ordinary x64 table with an ordinary memo field. Each row is one saved workspace: a unique id, the human name you load it by, the flavor and format measured at save time, size, lineage to the row it superseded, who saved it and when, and the roots its tables lived under. The memo field holds the payload. What varies is how much of the database that payload is:

• A posture -- which tables, which indexes, which tag orders, which relations, and where each cursor sat. Roughly a kilobyte for a thirteen-table system. The tables stay on disk; the row describes how to stand them up.
• A mini-database -- the posture plus every table's bytes and every attached index's bytes, in one binary-safe container. Ninety-four kilobytes for the same thirteen-table system. Nothing stays on disk; the row is the database, and it can be stood up into memory with no disk source at all.

That is the inversion worth sitting with. A database normally contains tables; here a table contains databases, versioned by the same append-only history the engine gives any other table, attributed to a real member, and verified byte-for-byte when written. Ordinary database machinery -- rows, a memo field, a record lock, an append -- turns out to be enough to hold databases, because nothing in the memo layer ever asked what it was storing.