Declaring what a file contains — column names, types, units, descriptions, table metadata and quality-control bounds — either in a MAML file on disk or built directly in code. A MAML can also declare how a file is read: which columns get renamed, which rows are kept, and in what order they come back. The group ends on two worked examples that put the pieces together.
fields:
section and its data types, deferring a col_size/array_size until write time with auto,
renaming columns on the way out with col_map: and on the way in with extra: remap:, declaring
a read-time filter and sort order with extra: filter:/extra: sort:, and how a MAML's
table-level keys become the parquet file's own metadata.schema%init/schema%add_field: the
same schema without a file on disk, choosing which of its columns a given program writes, adding
and clearing table metadata at run time, and printing a column listing.qc: min:/max:/miss: bounds: what each one checks,
building a qc-maml in code, deferring the declaration to read time with parquet_read_qc, and
write-side and read-side enforcement.