How we compute a match.

Most "which dog should I choose" guides online fall into one of three traps. The first: they confuse popularity with compatibility. You get offered a Labrador because it is the most adopted breed, not because it matches your home, your budget, your activity level or your experience.

The second: they reason in binary categories ("small dog", "family with children", "guard dog") where reality is continuous. A 7 kg Border Terrier is not a lap-sized "small dog" — it is a terrier that needs 90 minutes of exercise a day. The word "small" hides the number that matters.

The third: they never explain their criteria. You read that "the Cavalier King Charles Spaniel suits families" without knowing whether the author factored in the MVD heart predisposition, at 100% beyond age 10 (a considerable veterinary cost), or simply copied the FCI standard.

DogMatch starts from the opposite principle: every breed is scored on 14 personality axes from 1 to 5, your answers build an ideal profile on those same 14 axes, and we compute the distance between the two. No magic — just transparent arithmetic, with verifiable data and a publicly documented method.

In practice, the algorithm runs in four steps:

(1) Hard filters. Before any score is computed, some combinations are disqualifying. If you say you live in an apartment under 50 m², breeds over 35 kg with a space requirement of 5/5 are set aside — not because they are "incompatible" in the absolute (a Saint Bernard can live in an apartment given enough exercise), but because the odds of a daily imbalance are too high. The same goes for category 1 or 2 breeds if you have no prior experience: French law requires an ownership permit that first-time owners generally do not have.

(2) Ideal vector. Your quiz answers feed a 14-dimensional vector: energy, exercise, tolerance of children, space needed, and so on. Each option of each question pushes every one of those 14 axes up or down, following modifiers defined in questions.json. Ticking "activity 5/5" pushes ideal energy toward 5 and exercise toward 5. Ticking "first dog" lowers ideal trainability (you need a cooperative breed) while keeping the need for attention.

(3) Weighted Euclidean scoring. For every breed that survives the hard filters, we compute the weighted Euclidean distance between its native profile (the 14 scores of 1-5 taken from the literature) and your ideal vector. The weighting (see section 5) reflects how much each axis bears on real quality of life. The final score is normalized to 100, where 100 would be a theoretical perfect match (never reached in practice: no breed is exactly you).

(4) Sorting and tie-break. Breeds are sorted by descending score. Ties are broken by popularity in France (SCC data) — not because popularity is a quality criterion, but because it is a reliable proxy for the availability of serious breeders and of general veterinary support for that breed.

Every breed in the catalog goes through a structured extraction protocol based on four primary sources:

• FCI (Fédération Cynologique Internationale) — the official standard for each breed: size, weight, life expectancy, temperament as defined by the country of origin. FCI nomenclature →
• SCC (Société Centrale Canine) — French-specific data: LOF registration statistics, legal classifications (categories 1 and 2), the breeder network. SCC →
• AKC (American Kennel Club) — the American breed standards, more detailed on behavioral traits and health tendencies. Used as a complement when the FCI entry stays terse. AKC Breeds →
• French Wikipedia — for verifiable figures (average weight, life expectancy, genetic prevalence), always cross-checked against the primary sources. Never used on its own for a subjective trait.

For behavioral traits that do not appear in official standards ("easy to train", "tolerant of children"), we rely on published behavioral studies and veterinary consensus. When sources disagree, we record the median and flag the discrepancy on the breed page.

The quiz is not an individual predictive test — it is a classifier of breeds relevant to your profile. The distinction is crucial.

A Border Collie is not a guaranteed match just because it comes out first. Every puppy is singular, breeding temperament counts, and the training you provide counts for more still. The quiz points you toward a breed that is statistically coherent with your life; choosing a specific individual is done at a kennel or with a breeder, spending time with the puppy and its parents.

The quiz does not capture the "love at first sight" dimension. If you grew up with a Bernese Mountain Dog and that is the one calling you, no algorithm should talk you out of it — unless there is an objective dealbreaker (allergy, a budget incompatible with the veterinary costs, unsuitable housing).

The quiz does not replace a veterinary assessment or a conversation with a dog trainer. If your situation is particular (a very young child, an adult dog already at home, a respiratory allergy), go and see a professional who will meet your family before confirming the choice.

We also have structural biases. The catalog breeds are mostly FCI/AKC-recognized ones — so European and American, with a few Asian breeds. Primitive non-standardized breeds (village dogs, landraces) are not covered. If you adopt from a shelter, the quiz will give you a sense of the traits compatible with your profile, but the on-site assessment by the volunteers always comes first.

The 14 axes do not carry equal weight in the final score. A poor match on daily exercise (high-energy breed plus sedentary life) is devastating; a poor match on cold tolerance barely registers in mainland France. The weighting below reflects those differences.

How we calibrated these weights. The first version was based on veterinary consensus and behavioral literature (high weights for exercise/energy/space, low weights for climate tolerances in a temperate climate). Validation used synthetic personas: 18 hand-built user profiles (young urban professional, rural family of three children, senior in an apartment, and so on) for each of which we know the expected answer. We iterate on the weights until at least 12 personas produce a top 3 judged plausible on manual review.

Three operating commitments, for as long as this site exists:

(1) Every calibration change is public. If we change a weight, a hard filter, or the modifiers of a question, it is announced at the top of this page with the date and the reason. Your results may legitimately shift between two visits — you will know why.

(2) No breeder promotion influences the ranking. We accept neither money nor breed placement. The only monetizations we consider (never switched on without notice) are affiliate links on generic utility products (litter, transport, insurance) that create no bias in the breed ranking.

(3) Your data stays local. Your quiz answers are stored in localStorage — on your device, never sent to a server. You can erase them at any time from your browser settings. We run no intrusive analytics and there is no user account.

Does a ranking look wrong to you? Challenge it. Every reasoned report is checked against the sources, and the entry is corrected if you are right — it is our best calibration loop. A question, a breed missing from the catalog? Same door: the contact page.