The Data Desk now has a data dictionary at /data/dictionary. For every dataset offered on Downloads it lists each field of the JSON, by its path, and each CSV column, with its type, its unit, what it holds and what an empty value means there. An empty value has five names, each defined once: not disclosed, where the source could have stated it and did not; not read, where the desk has not read or recorded it; not applicable, where the field has no meaning for the row; not in this dataset, where a file leaves something out on purpose, as the difficulty file leaves out the reward ledger; and none listed, for a list the record leaves empty. None is ever written as 0. Six marks say when a figure is more limited than it looks: a floor on disclosed positions, a cut decimal, a filing's own approximately, a holder's own statement, a labelling rule, or chain data read from one operator. The dictionary is held to the files by the site's tests: a field with no entry, an entry no file or loader holds, a value of another type or unit, or an empty value in a field said to be never empty fails them. Downloads links each dataset to its fields, each dataset's metadata file gives their address, and the method page links the dictionary. A dataset with no file has no fields to define and says why, as it does on Downloads. No figure changed.
Data Desk