The names that are also breeds, and the empty field that found four impostors

Set the name catalogue against the breed catalogue and the overlaps fall out immediately. — Savannah is a cat breed and a girl's name, Billy is a French hound and one of the most used dog names here. Others were never names in any ordinary sense: nobody calls a dog Havanese. The useful part is that the catalogue could tell those two groups apart without anybody deciding which was which, using a field that was simply empty on one of them.

What the test found

Four words came out of it that were breeds and not names: Havanese, Doxie, Pomeranian and Schnauzer, holding several hundred registrations between them. They're no longer in this catalogue. They were rejected instead of deleted — the rows keep their counts and their sources so the import can still be reconciled against the published register — but they no longer appear as names anywhere on this site.

Ten overlaps remain, and six of them carry a recorded sex: Billy, Savannah, Bambino, Somali, Pumi and Whippet. Those are names people chose, whose breed meaning is a piece of trivia.

Why the empty field is the tell

Sex isn't guessed on this site. It's read from how real animals of that name were actually sexed in the registers, so a name carries male or female because enough animals answering to it were recorded one way. A word that reached the name field because somebody typed the breed there has no such record behind it — the animals exist, but nothing tied a sex to that word as a name.

All four of the rejected words had an empty sex field, and all four arrived from the same two registers. Against a catalogue where about twenty-eight names in a hundred have no recorded sex, four out of four isn't a coincidence — and it's a test anybody can repeat, not an editor deciding that Havanese looks wrong.

The registration counts say the same. The six overlaps that carry a sex average more than ten times the registrations of the four that do not, which is what you'd expect if one group is names chosen repeatedly and the other is an occasional slip of the pen in a form.

The four that stay, and why

Four of the remaining overlaps have no recorded sex either: Brittany, Chantilly, Bombay and Chinook. They stay, and the reason is the rule instead of the evidence. Each is an ordinary word — a French region, a lace, a city, a wind — that a person could plausibly give an animal, and a missing sex on thirty registrations is thin. Where the evidence doesn't separate a label from a name, the name stays. A catalogue that deleted rows because they looked wrong would be making a judgement it couldn't show you.

Where this test stops, and what a second one found

The test matches a whole breed name or the last word of a two-word one, which means it can't see the first word. French and Golden are invisible to it, and both were sitting in the catalogue: French with a few hundred registrations, Golden with rather fewer, each the head of a breed and not a name. A separate scan for the same signature found them, and both have been rejected.

French is the clearest case in the whole catalogue, and it's worth saying why, because the reason isn't that it looks wrong. A name with its number of registrations appears in about ten of the source registers on average, and only one in a hundred appears in fewer than three. French appears in exactly two — the same two that leaked the others — and carries no sex where names of that size almost always do. Two independent anomalies pointing the same way is a finding; either one alone wouldn't be.

Which is also why four more breed words are still here. Lab, Chin, Hound and Chow are breed words with no recorded sex, and by description they look identical to the six that went. They stay because at their size the evidence doesn't exist: a missing sex is the norm and not the exception below about thirty registrations, absent on more than four names in ten there, against one in a hundred higher up. The empty field is only evidence where a sex would otherwise almost certainly be. Below that line a word stays whatever it looks like, and looking wrong has never been the standard here.

Why a breed ends up in the name box at all

It's worth picturing how this happens, because it isn't carelessness so much as a form doing what forms do.

Somebody at a counter is filling in a licence for an animal. There are two boxes near each other, one for the name and one for the breed, and occasionally the breed goes in both — or the person filling it in doesn't have the name to hand and writes what they can see. Multiply that by a few hundred thousand records and a handful of breed words end up in a column where they don't belong.

It also happens the other way round, which is harder to detect. A name that looks like a breed can be recorded as one, and that error leaves no trace in the name data at all — the row simply never appears. Nothing on this page can find those, and it's worth being clear that the cleaning only runs in one direction.

None of this is a criticism of the registers. They're administrative records published as a public good, not a dataset anybody built for this purpose, and a small rate of field confusion is the ordinary cost of using real paperwork instead of a survey. The job on this end is to notice it rather than to complain about it.

Why this matters beyond fourteen words

A few hundred registrations out of many hundreds of thousands is, on its own, not worth a page. What's worth a page is the method.

The interesting thing about this test is that it removed rows on evidence instead of on taste. Nobody looked at Havanese and decided it wasn't a name; a field that's populated on almost every well-used name was empty on all four candidates, from the same two sources, with registration counts an order of magnitude below the ones that stayed. That's a finding, and it's reproducible by anybody with the same data.

The same discipline is why four obvious-looking breed words are still in the catalogue. They fail the eye test and pass the evidence test, and the evidence test is the one that decides — because a rule that lets an editor delete what looks wrong is a rule with no limit, and the next thing it deletes is a real name somebody actually uses.

That's the general shape of how everything here is maintained: fields left blank when nothing measured them, figures recounted instead of asserted, and removals that can be explained without appealing to anybody's judgement. This page is one small worked example of it.

The wider set, where the overlap is a coincidence

Widen the test to names matching the last word of a two-word breed and twenty-nine more appear, holding several thousand registrations between them. This set is mostly coincidence and not contamination: Blue is the single biggest of them and is a genuine and popular name that happens to end Blue Heeler and Russian Blue. So are Rex, Laika and Chocolate.

What to take from it

If you like Savannah or Billy, the overlap costs you nothing — they're ordinary names with hundreds of animals behind them, and the breed meaning is trivia and not a warning. And if a name on any list looks less like a name and more like a label, the sex field is the quick check: a name people have been using for years will have been sexed by the registers long before it reached a site like this one.