NomenSoft data, without the ingestion work

How developers usually meet NomenSoft

As `TXT.zip` and `XML.zip`. The TXT archive is 63 MB compressed and 912 MB unpacked across 46 tables; the XML archive holds the same data at nearly twice the size, with one file exceeding a gigabyte on its own. Both come with `.xsd` schemas and a spreadsheet describing every column.

The tables are relational: codes in one, descriptions in another, amounts in a third, with the legal authority behind each fact in a fourth. Getting from the archive to "what does code 101076 cost today" is half a dozen joins and a careful reading of the date columns.

The parts that are easy to get wrong

The format has no quoting, so a text field containing a newline splits a record across two physical lines, and no CSV library can help — 14 records do this. Amounts use European notation, where `1.694,420000` is one thousand six hundred and ninety-four euros and forty-two cents. `31-12-2999` is not a date but an open-ended marker. And the update files express a change as a delete followed by an add, sometimes in that order in the file and sometimes not.

None of this is obscure once you know it. All of it costs an afternoon to discover and a production incident to discover late.

What this API does instead

Downloads each publication, verifies it by checksum, parses it, validates the row counts, and only then makes it visible. Amounts are exact decimals, the open-ended marker becomes `null`, and every response names the publication it came from.