A variety name is not a variety identity

This is a note about a matching problem that looks solved and is not.

Official plant variety registers publish a denomination — the variety's registered name — alongside a species, a holder, a status and dates. If a corpus holds a cultivar called "Cadenza" and a register holds an entry called "Cadenza", it is tempting to treat them as the same variety. The names match exactly. The register is authoritative.

What the registers actually returned

Across four official registers — Great Britain and Northern Ireland variety lists, the Canadian register of varieties of crop kinds, United States plant variety protection, and Australian plant breeder's rights — 42 entries matched a cultivar name exactly.

Eighteen were the same variety. Twenty-four, or 57%, were a different botanical species:

Register	Name matched	What the register actually held
United Kingdom PBR	Cadenza (bread wheat)	Fragaria × ananassa, a strawberry
United States	McIntosh (apple)	Triticum aestivum, a common wheat
United States	Roma (tomato)	Phaseolus vulgaris, a garden bean
United States	Gala (apple)	an onion, a tall fescue and a potato
United States	Marquis (wheat)	Phaseolus vulgaris
United Kingdom	Riesling (grape)	Solanum lycopersicum, a tomato
United Kingdom	Carmel (almond)	Spinacia oleracea, a spinach

Every register reproduced the trap independently. It is not a quirk of one system's data entry.

Why this happens

Variety denominations are required to be unique, but the scope of that requirement is a species group — not the whole of botany. "Roma" may be a tomato in one group and a bean in another without any rule being broken. The registers are behaving correctly. It is the name-only matcher that is wrong.

What makes the error dangerous is that it is convincing. A false match is not obviously junk: it carries a real register identifier, a real country, a real grant date and a real status. Published without checking the species, it looks exactly like a verified record.

The correction, and a second one

Match on the denomination and the register's own species field. AgricultureID now records the register's species verbatim alongside every registration, and the match basis vocabulary has exactly one member, so a weaker basis cannot be named and then quietly used.

Reviewing the rejected pairs individually surfaced the opposite error too: two genuine matches had been discarded because the register publishes durum wheat at species rank as Triticum durum Desf. where the corpus uses subspecies rank, Triticum turgidum subsp. durum. That is one documented taxonomic equivalence, written down as an exception — not a fuzzy rule, because a fuzzy rule would re-open the door the species check just closed.

What this means for reading any variety claim

If a source tells you a variety is registered somewhere, the useful question is not whether the name matches. It is which species the register says the entry is for. Where AgricultureID cannot answer that, it publishes no registration at all — 128 of 251 candidate matches were discarded on this basis in an earlier pass, and the discard list is the part of that work we would defend hardest.