What changed
Every change to a page a reader sees, newest first. Corrections are listed beside new work, because a figure is only worth trusting if the last wrong one was fixed in the open. To report something wrong, use the corrections page.
- Correction
A breed's genetic diversity rank now shows how sure it is
We rank breeds by genetic diversity. Checked against an independent study of different dogs on a different genotyping array, the ranking holds up well. But when we resampled each breed's dogs, many exact positions moved by more than 10 places, mostly in breeds with few dogs. A rank now shows with its range (the Golden Retriever is 28th least diverse of 107, between 25th and 31st), and a breed whose range is too wide shows its measured diversity and dog count with no position. Four breed pages also described diversity wrongly, and are fixed: two in the wrong direction, one crediting our number to another study, and one garbled.
- Correction
A breed relative now needs at least 20 dogs of each breed behind it
Some named relatives rested on as few as 8 dogs of one breed, under the 20 we require before reporting any figure. A relative is now named only when both breeds have at least 20 dogs in the atlas. The Golden Retriever lost the German Shorthaired Pointer (12 dogs), the Labrador lost the Curly-Coated Retriever (8), and the Akita answer now names the Chinese Shar Pei and says the Chow Chow, which the independent study also places beside it, is not confirmed yet at 17 dogs.
- Correction
Breed relatives are named only where two independent lines of evidence agree
The atlas, the question box, the research lens pages, the breed compare pages and one answer page named a breed's closest relatives from a single genetic distance. Tested against an independent study of 161 breeds, that distance did not recover known relationships well enough to be a family claim, partly because dogs genotyped by the same study look alike. A relative is now named only when it is near in Sniff's genetics and the independent study places it in the same group, and the study is credited. The Akita answer changed: its first-named relative had been a village population.
- Correction
Breed pages no longer say a share of dogs carries a variant when the figure counts gene copies
The "In plain terms" cards on breed pages gave the share of tested gene copies as if it were the share of dogs carrying a variant. Each dog has two copies, so the share of dogs carrying one is higher than the figure shown. The cards now say it in copies, and the six condition cards are rewritten in plain words, each sentence still tied to its source.
- Correction
The Sniff Atlas dataset is at version 1.0.2, and every page cites it
Version 1.0.2 of the open dataset is published. It carries no model pathogenicity score, describes its scope as genome-wide, and drops ten knowledge-graph entries whose genome positions had been mapped wrongly; every frequency is unchanged. The citation blocks on the dataset, cite, gene, disease, variant and breed pages still named version 1.0.1 and its old title. They now name 1.0.2, and the concept DOI they give always opens the newest version.
- Change
The science page states the classification rule in one checkable sentence
The science page said Sniff does not score variants with a model of its own. It now says Sniff assigns no pathogenicity to any variant, and that every classification on the site belongs to a cited curated source. The dataset, the agent package and the pages already worked this way.
- Correction
Gene and atlas dog pages no longer show trait figures that do not track the trait
The atlas reads fifteen trait genes (size, legs, coat, ears, face shape) through one marker each. Breed pages already withheld these, because none of the markers tells breeds that visibly have a trait from breeds that lack it. Gene pages still listed their breed frequencies (the long-coat page put Pugs at 100%), and research-dog pages turned them into predictions such as a Golden Retriever "predicted small". Those figures and predictions are gone from every page, each gene page says why, and a build check now fails any page that shows one.
- Change
The question box answers from the same figures the pages show
Asked for a breed and a condition together, the question box now answers with the figure the breed page shows, with the number of dogs tested and the source, and leads with the curated standing where a test is disputed. Asked to put one number on a dog's health, or to say how long a dog has left, it says why it will not and offers the cited life expectancy instead. A weekly check reads the live site and asks the question box the questions it must refuse; the first run found these four gaps.
- Correction
One breed page was built at an address the site redirects
The screening panel files the Papillon under its FCI name, Continental Toy Spaniel, and a page was built under that name while the site sends that address to the Papillon page. The two showed different figures. The panel's dogs under both names are now pooled on the Papillon page, with the sample sizes added, and a build check fails if any page is ever built at a redirected address again.
- Correction
Breed pages give one life expectancy, and say where it comes from
The written sections of fourteen breed pages quoted lifespans that did not match the cited life expectancy at the top. On four, the cited figure had been pasted into a breed-club range, so the Golden Retriever page read "10 to 13.2 years". On the rest, the cited figure was called the atlas median, but the atlas does not measure lifespans. The Havanese and Belgian Shepherd pages quoted a figure with no source at all; they now say no published life table covers them. The Golden Retriever page also attributed a cancer mortality figure to the lifetime study that did not come from it, and now cites the necropsy series it came from. A build check now reads every written breed section and fails if a lifespan disagrees with the cited figure or names no source.
- Correction
The atlas paragraph quotes the cited life expectancy, with its cohort
On thirty-four breed pages, the paragraph headed "What the atlas says" quoted a median lifespan from an older source that disagreed with the cited life expectancy a screen above it. The two came from research files generated at different times, and nothing compared them. The paragraph is regenerated from the same cited figure, names the cohort and the number of dogs behind it, and a build check now fails if the two ever differ. The weights in kilograms are gone from it as well.
- Change
A page for every breed the screening panel tested
The Donner 2023 panel screened breeds that had no page on Sniff, so their measured variants had nowhere to land: the Lagotto Romagnolo, the Catahoula Leopard Dog, the Kooikerhondje, the American Bulldog and a hundred more. Each now has a page built from the panel alone: which tested variants turn up, with the sample size beside every figure, and the conditions recorded for the breed. Where the page has no atlas placement or lifespan, it says so and why. One rule now decides which breed pages exist, and the navigator, the sitemap and the breed index all read it.
- Change
The site is named Sniff Dog Genetics Atlas where search engines look
A bare "Sniff" was losing the search results to an unrelated app. The name search engines and link previews now see is Sniff Dog Genetics Atlas, and a shared breed page shows a card with the dog, the breed and the site name instead of a photo cropped at random. A contact page and a press page exist, so the two questions every visitor eventually asks have an answer.
- Correction
Five breed pages gain the screening data filed under another name
The Shiba Inu, Bull Terrier, Bouvier des Flandres, American Hairless Terrier and Petit Basset Griffon Vendéen pages had no variant table because the screening panel lists those breeds under a different spelling or word order. The pages now resolve those names and show the rows with their sample sizes. A second lifespan with no source, and a weight in kilograms, no longer appear in the atlas narrative on forty breed pages.
- Correction
The per-breed variant figure is an allele frequency, and every page now says so
On every breed page and every disease page, the figure from the Donner 2023 screening panel was labeled a carrier frequency. It is an allele frequency: of the gene copies tested, the share that carry the variant. Every dog carries two copies, so it is not the share of dogs, and for a recessive variant the share of dogs carrying at least one copy is higher than this figure. Every label, sentence, chart and data description now calls it an allele frequency, the glossary entry explains what it counts, and a build check refuses the old label on any page from now on.
- Change
Two lifespans on the Rottweiler page explain themselves
The page showed one figure at the top and a different one a screen below for what read as the same question. The top figure is life expectancy; the measured section gives median age at death among dogs that died, from a different study. Each now names its measure and its study, side by side.
- Change
Breed pages open with the answers an owner came for
A breed page now starts with lifespan, the tested variant with the highest allele frequency, and how many dogs stand behind the page, each with its source. Carrier variants, plain-terms condition guidance, feeding and open questions follow. Population genetics, model-organism links and the full condition list fold into one section for researchers. A navigator shows where you are on long pages.
- Change
What a page says about a condition is written once, with its source
Guidance about a condition now comes from one record that rests on a sentence quoted from the curated source or on a cataloged paper, and the same guidance renders on the breed page and the disease page. Two tested markers the curated source disputes now read "marker disputed" and carry the source's own words instead of advice built on them.
- Correction
A public dog page shows what the form promised, and no more
The add-a-dog form promised that a public page shows name, photo and breed. Pages were also showing birth year, sex, weight, diet and location. The default now matches the promise: those fields stay off a public page until the owner turns each one on. Pages created before this date changed too.
- Correction
Trait frequencies removed until a marker shows it tracks its trait
Fifteen single-position markers had been used to show coat, size and body-shape frequencies by breed. Tested against breeds that visibly have each trait and breeds that lack it, none of them tracked its trait, and two ran backwards. Nothing renders from them now. A marker returns when it passes that test.
- Correction
Heartworm prevention is safe for a dog with the MDR1 variant at the label dose
Eight breed pages said a dog with the MDR1 variant cannot take ivermectin-based heartworm prevention. The primary source says the opposite: at the label dose, every approved heartworm preventive is safe even for a dog with two copies, and the danger is high doses of ivermectin and related drugs. Twenty-four sentences were corrected and cited, and a build check now refuses a parasite-drug warning that names no dose.
- Change
No page shows a model's pathogenicity score
Sniff shows the variant classifications that curated sources publish and cites them. It does not assign pathogenicity with a model of its own, and no public page presents a model score any more. Splice predictions computed with a published third-party model are labeled as that model's output, with a note that it has not been validated in dogs.
- Change
The Rottweiler page is built from the published literature, question by question
The Rottweiler page carries a frame of nineteen questions an owner or a vet asks about a breed. Each is answered with its figure, its cohort and its paper; held, with the source named, when the source does not support a figure; or left open with the name of who holds the data nobody has read in yet. Life expectancy, body weight, hip and elbow results, leading causes of death and breed recognition are answered today.
- Correction
The atlas dataset is described as what it holds
The published catalogue's description had narrowed it to the protein-coding part of the genome. Most of what it holds lies outside those regions. The description now says genome-wide, and the page that explains the dataset says what each file contains.