It probably depends on how you define "schema-less". I can't speak for all brands, but in Dynamic Relational, a draft RDBMS idea, if you issue "SELECT madeUpCol

It probably depends on how you define "schema-less". I can't speak for all brands, but in Dynamic Relational, a draft RDBMS idea, if you issue "SELECT madeUpColumn FROM myTable", you'd get nulls for that column (assuming myTable has rows). Depending on settings, the table could also be made up (unused), in which case you'd get zero rows back. You'd never get errors like "column doesn't exist" or "table does not exist" (unless you add constraints, which would be common as projects mature to ensure data integrity). You can only see if there is non-null data in a named slot. In such a case, there is no formal centralized schema. Any notion of schema would come from a survey of what data exists. Columns and tables are essentially "create on write" (UPDATE and INSERT).