Matching should be explainable, or it should not exist.
Placing an animal is one of the most consequential decisions a shelter makes, and a person makes it every time. The ranking in WhiskerMatch discovery exists to do the honest part of that work: compare what is known about each animal against what a household is looking for, show its reasoning in plain words, and stay out of the decision itself.
A transparent preference ranking.
Each animal's facts, species, breed, age, size, sex, energy, compatibility with dogs, cats and children, special-care needs, transport, location, availability, are read as known or unknown. Each of your preferences is a must-have or a nice-to-have. Known conflicts with a must-have exclude; everything else produces reasons and warnings you can read. The full step-by-step lives on the mechanism page.
Five choices the ranking is built on.
None of these are the obvious engineering choice. Each one costs the system something, speed, simplicity, or engagement, in exchange for treating animals and people fairly.
01
Unknown is not no
A missing fact never disqualifies an animal. If a source did not state a breed, an age, or whether the cat likes dogs, the animal stays in results with a plain warning that the fact is unconfirmed. Missing data should never punish the animal it describes. Only a known conflict with a must-have excludes.
02
Being open to special needs never means a pet must have them
The special-needs preference is deliberately asymmetric. Saying you are open to special-care context means those animals stay visible to you; it never filters out animals without needs. The alternative, treating openness as a requirement, would be a quiet way of hiding healthy animals from willing people. The system refuses that reading.
03
Distance is honest about being approximate
Public listings use deliberately approximate positions, never exact addresses. Distance math carries an uncertainty radius: an animal is only excluded when it is definitely outside the chosen range, and every result warns that the distance is estimated. Precision about an animal's location would be a privacy failure; the ranking admits its own blur.
04
No score is ever shown, because a score becomes a verdict
Internally, nice-to-have preferences carry small ordering weights, gentle, normal, strong. They are ordering mechanics, and they are never returned as a number, a percentage, or a rank to the visitor. The moment a system shows '94% match', people treat it as a recommendation. It is not one. It is a starting point, and the page says so in words, with reasons and warnings, not digits.
05
Equal animals rotate, because exposure is a resource
When candidates are otherwise equal, a stable seed hash rotates their order so no provider's ordering can monopolize attention. Paid placement, photo count, clicks and popularity are not accepted inputs at all. Nobody can buy the top of the list, and the most-clicked animal does not get a boost for being clicked.
What the ranking is not.
It does not score, rank, or approve adopters
It does not score, rank, or approve adopters
There is no adopter score anywhere in the system. Applications are reviewed by people against context the organization holds. A discovery preference is about the animal's known facts matching a household's stated preferences; it says nothing about the household's worthiness, and it is never used that way.
It does not place animals
It does not place animals
No automated matching or placement exists in WhiskerMatch. The ranking orders a browsing list. The placement decision, the application review, the home check, the judgment call, belongs to named people at the organization, every time.
Swiping is interest, not an application
Swiping is interest, not an application
In the swipe view, saving interest is a bookmark. It does not apply, notify the shelter, or start any process. Unknowns stay unknown through the whole interaction: the system never fills in what it does not know to make a card more appealing.
Lost and found matching is rule-based, and a person decides
Lost and found matching is rule-based, and a person decides
Possible matches on lost and found reports come from explicit rules, and a person marks them. There is no image matching and no AI matching of animals to reports. When the stakes are a family's lost pet, the system suggests and the person decides.
What the ranking cannot know.
We do not know which signals predict lasting placements.
The attributes the ranking compares, energy, compatibility, special-care context, are the ones shelters already record and households already ask about. Whether matching on them produces better long-term outcomes is genuinely unsettled, and we will not claim otherwise. The ranking is a way to browse with your preferences stated, not a prediction about your future.
Facts on a profile are only as good as their sources.
A "good with dogs" fact is an attributed claim from intake or a foster, reviewed by a person, not a measured trait. The ranking treats it as reported, and the profile shows the uncertainty where it exists. Anyone who wants certainty about an animal should meet the animal, through the organization.
Researchers: the data model is built to be studied.
Known and unknown are stored separately. Reasons and warnings are generated, not hand-written. Exclusions carry their conflict codes. That structure exists so the ranking can be audited and, one day, so its relationship to real outcomes can be examined honestly. If you study adoption outcomes, we want to talk.
Studying adoption outcomes? Bring your questions.
The ranking is built to be audited. If you research placement, compatibility, or adopter behavior, the mechanism is open and the data model is honest about its limits.