Intelligence begins
where averages end.
The first behavioral intelligence research platform built around individual baselines rather than population averages. Longitudinal models designed to detect the slow behavioral drift population statistics are built to miss. Built for dogs. Designed to change how intelligence is measured.
Every model compares. The question is — to what?
The reference class is the hidden variable of machine learning. Compare a dog to its breed and you get a statistic. Compare a dog to itself and you get a signal. Barkley makes that variable explicit — and rebuilds the entire behavioral stack around the individual.
The individual baseline
Each dog's longitudinal norm, learned from its own history — not sampled from a population it never belonged to.
Temporal identity
Behavior is a trajectory, not a snapshot. Velocity and rate of change carry the diagnosis long before the state does.
Silence as signal
Missing data is classified, never discarded. A dog that stops doing something is saying something.
Longitudinal baselines are accumulated, not scraped: every tracked day deepens a per-individual dataset no competitor can reconstruct retroactively.
Same detector, two reference frames: +0.053 AUC and +19 points of recall. The gain lives in the comparison, not in the parameters.
Eight open, testable layers with a reproducible 30-seed benchmark — DOI-archived, synthetic-data validated, ready to extend.
Same dog. Same data. Opposite conclusion.
One year of a single dog's behavior. Switch the reference frame and watch the meaning invert: what a breed model reads as a return to normal, an individual baseline reads as a clear behavioral drift.
A dog can be normal for its breed — and abnormal for itself.
Decline is a slope, not an event.
Most behavioral change emerges gradually — through drift, reduced recovery, altered exploration, and informative silences. The same pattern can mean stability for one dog and deterioration for another. Population norms can't tell the two apart. An individual baseline can.
Not a model. An architecture.
Eight composable layers — each one published, testable, and open. This is the reference stack for individual-first behavioral intelligence.
The benchmark is the pitch.
No claims without artifacts. The head-to-head validation runs the same detector against two reference frames — the individual's own baseline versus the breed average — and archives every result with a DOI.
Same detector, individual vs population reference. Synthetic, 30 seeds — a reproducible proof of the reference-class effect, not validation on real dogs. Reproduce on GitHub ↗
Open by method, not by marketing.
Papers, datasets, code, and live demos — a research house working in public, with every artifact archived and citable.
Built from the inside out.
Barkley did not begin as a market thesis. It began with four Jack Russell Terriers, years of behavioral work, and one realization: pet technology was learning to measure dogs while still failing to understand them.
"A dog can look perfectly normal for its breed while quietly drifting away from itself. Barkley is the missing layer."
Cross-disciplinary systems architect and certified canine behavior specialist. Her work is shaped by a long-standing concern with ethics, living systems, and the role technology plays in society — in particular, with how AI systems classify, average, simplify, and sometimes erase the individuality of living beings.
Beyond monitoring.
From behavior to physiology to prediction — one continuous, individual-first pipeline. Edge processing, privacy by architecture, and a memory that belongs to the dog.
Execution, in public.
Investors don't only buy today's state — they read the speed of execution. Papers, patents, an open benchmark, and a reference architecture: all shipped, all public, all in one year.
Where this goes.
Current stage: pre-commercial research · 2026. The science is built; the next phase takes it to the field.