Check every mapping table against the HCAT taxonomy

[ivorbosloper]

Every HCAT3_name in a mapping table has to carry that name's code. A wrong code is invisible today: the file validates, and a crop query quietly returns another crop — see #304, where 2,098 published fields do exactly that.

tests/test_hcat_codes.py walks every CSV under tests/data-files/ that has HCAT3 columns and checks both halves of the pair against tests/data-files/hcat3.csv (a copy of https://fiboa.org/code/hcat3.csv). Offline and instant.

The 23 rows that are wrong today are listed in KNOWN_BAD with the reason, so the test passes now and a new mistake fails. Injecting one of each kind into dk_2019.csv gives:

convert/dk/dk_2019.csv 1: spring_barley has 3301010403, the taxonomy says 3301010402
convert/dk/dk_2019.csv 2: 'spring_soft_wheat' is not an HCAT class

Two things found while writing it:

• sk/sk.csv in the fixtures is stale — 244 rows against 320 live, and it still has parsley and zucchini_courgette 3301140200 where the published list has been corrected to parsly and 3301140600. So the conversion tests exercise a table that is no longer what the converter reads. Not fixed here: refreshing it changes what the sk conversion test expects.
• The check only reaches tables that exist as fixtures — 23 of the 32 bad rows in 32 rows across the code lists disagree with the HCAT taxonomy #304. ch, pt and at have no fixture, so their errors are invisible to it.

🤖 Generated with Claude Code

https://claude.ai/code/session_01DVx9uQV2QPM8ecPAY3ZjXG