How Dog Name Matching Works, Explained Step by Step

The Dog Name Compass is seven adaptive screens, roughly two minutes, that end in twelve names split into three groups — six Best Matches, three Practical Alternatives, three Worth a Look — each with a plain-English reason drawn from your own answers. Nothing about the scoring is written by a language model at runtime: the matcher is a deterministic, versioned system, meaning the same answers against the same dataset and scoring version always produce the same ordering. A short household sound check runs entirely in your browser and never reaches a server. Breed is used, but deliberately weighted low. The rest of this page walks through each piece.

Each of the seven screens exists to narrow the field on one specific dimension, not to collect data for its own sake. Here's what's asked and why it's on the list.

Scoring is deterministic and versioned. That means two things in practice. First, deterministic: no language model generates or ranks names at the moment you take the quiz — the same set of answers run against the same dataset version and the same scoring version will always produce the same twelve names in the same order. Nothing is randomized and nothing depends on server load or the time of day. Second, versioned: both the underlying name dataset and the scoring logic carry version numbers, so if either changes later, that's a traceable update rather than an invisible drift in results. Fit labels — Best Match, Practical Alternative, Worth a Look — are relative to the strongest candidate produced in that specific run, not fixed thresholds a name either clears or doesn't. The labels describe how a name did against its actual competition for your answers, not an absolute score compared across every run anyone has ever taken. Two people with almost identical answers can reasonably see different Best Matches, because their small differences in cues, size and style shift which candidates compete hardest within their own run.

The household sound check compares candidate names against the names of people and other pets in your home, and against the cues you already use, to catch collisions before you do — a new puppy called something too close to your cat's name, or to 'sit', is a problem worth catching in advance. Those words are processed entirely in your browser. They are never sent to a server, never logged, never included in a URL, and never fed into analytics. The server that generates your twelve results only ever receives controlled option IDs — a size category, a style selection, a cue flag — not the actual words you typed for people or pets in your household. This is a deliberate design choice, not an incidental one: household names are personal, and there's no reason a naming quiz needs to store them anywhere. If you close the browser tab partway through, whatever you typed for the sound check simply disappears with it, rather than sitting somewhere waiting to be resumed from a server-side record.

Breed is included as an input, and it's deliberately capped low in how much it can move a result. It is never used to infer temperament — the Compass doesn't reason that a working-line dog needs a serious name because of assumptions about the breed's personality. What breed does contribute, in a small way, is a nudge toward naming conventions common for that breed's typical role or size, nothing more. There's no photo upload anywhere in the flow and nothing is inferred from a dog's appearance. Breed exists in the quiz because some owners expect to see it asked, and because it can slightly inform size-related weighting when size itself isn't specified — not because breed predicts what kind of dog you actually have. Two dogs of the same breed can and often do land on very different result sets, since style, size and the household sound check all carry more weight than breed does.

The Compass currently holds one approved lexical source. Its licence terms cover only the plain, present-day sense of an ordinary English word, and they explicitly exclude etymology, the origin of a personal name, the meaning of a word in any language other than English, and cultural or religious significance. That's a narrow grant, and it rules out the kind of claim most naming sites make without a second thought: what a name is said to mean in another language, what a surname is said to trace back to, who a mythological figure was understood to be. Rather than publish a guess dressed up as fact, every result card currently shows 'Meaning not verified' until a multilingual dictionary source is registered and has been through the same licence review this one went through. That isn't a promise of a date — it's a description of why the field is empty right now, and it stays empty for as long as no properly licensed source exists to back it. No meaning shown anywhere on the Compass is written by a language model; an AI-generated etymology can sound confident and still be wrong, and a naming decision isn't the place to risk that.

Popularity, when it eventually appears, will be scoped rather than national. The intended source is public dog-licence registration data from New York City and Seattle — real records of names actually registered to dogs in those two cities, not an estimate or a survey. That's a genuine observation about two specific places, never a claim about the most popular dog name in the country, because the Compass has no dataset that would support a national ranking. As of this page, that dataset is still being processed, so popularity bands are not live in the quiz results yet. What's written here is the intended method and its scope, stated honestly ahead of the feature landing, rather than a number presented before the underlying data is ready to support it. When bands do appear, expect them to be labeled clearly as observations from those two cities specifically, not restyled into a ranking they were never built to support.

Being clear about what the Compass refuses to do is as much a part of the design as what it includes, and every item below was a deliberate decision rather than something left out for lack of time. None of the following is in the tool, on purpose.

If a detail on a result card looks wrong — a size category, a style tag, a cue flag, or an unverified label you believe should be sourced — it's worth reporting rather than working around. Email the address listed on the site's contact page with the name in question and what looks incorrect, and it's reviewed against the source registry behind that entry before anything changes — nothing is edited on a report alone, and nothing is added just because a reader is confident about it. This is also the right channel to flag a name that seems to be missing a detail it should plausibly have. Corrections that check out are applied to the dataset, not just the display, so a fixed entry stays fixed the next time anyone else's quiz run happens to surface that name.