labs-barkley/synthetic-doggraph-sample

Synthetic DogGraph Sample

Longitudinal Canine Behavioral Trajectories — Barkley AI

A synthetic research dataset for individual behavioral baseline modeling, temporal drift detection, and silence-as-signal classification in companion animals.

DOI: 10.5281/zenodo.20356188

---

Dataset Summary

This dataset contains synthetic longitudinal behavioral records generated to support research in individual-level behavioral intelligence for dogs. It is not a population dataset. It is not a breed-comparative dataset. It is a trajectory dataset — designed to enable the study of how a single individual's behavioral signal evolves over time relative to its own established baseline.

400 rows × 14 columns — 50 synthetic subjects × 8 quarterly observations each.

Each record reflects variation in:

• Activity trajectories — temporal movement and activity proxies
• Recovery latency — time-to-baseline following stimulation or disruption events
• Distributional drift — progressive divergence from the individual's own historical mean
• Contextual annotations — environmental and routine metadata

The underlying generative model is deterministic and seeded, ensuring full reproducibility across all derivative analyses.

⚠️ Synthetic data only. No real animal behavioral data was collected, used, or approximated in the construction of this dataset. Not for veterinary or clinical use.

---

Dataset Structure

Column	Type	Description
dog_id	string	Unique individual identifier (synthetic)
breed	string	Breed label (10 breeds)
age_years	float	Age at enrollment (2–14 years)
observation_quarter	int	Quarter index (1–8)
observation_date	date	Observation date (synthetic)
bin_a_nocturnal_variance	float	Circadian nocturnal activity variance
bin_b_weekly_baseline_normalized	float	Weekly noise-normalized activity baseline
bin_c_quarterly_ks_statistic	float	Quarterly distributional drift (KS statistic)
cognitive_performance_index	float	Composite cognitive proxy score

Files:

synthetic_doggraph_sample.csv   # Full dataset (400 rows × 14 columns)
README.md                       # This file

---

Why This Matters

Most behavioral monitoring systems in the animal health domain are calibrated against population norms — breed averages, age cohort baselines, or generalized thresholds. This framing has a structural limitation: a dog can remain within breed-typical ranges while sustaining a meaningful, sustained decline relative to its own behavioral history.

This dataset operationalizes that distinction. It provides:

1. A controlled environment for developing individual-baseline modeling approaches
2. A reproducible trajectory corpus for evaluating temporal drift detection algorithms
3. A structured framework for studying behavioral signal as a longitudinal phenomenon

The concept is formally described in the companion framework paper:

From Surveillance to Cognition: A Unified Framework for Precision Behavioral and Metabolic Intelligence in Companion Animals DOI: 10.5281/zenodo.20060327

---

Companion Repository

This dataset can be explored through the Barkley Reference Architecture — the source-available computational backbone demonstrating the public schema, multi-resolution temporal binning, individual baseline mechanics, Rate of Drift, and missingness classification used to interpret synthetic DogGraph-compatible events.

git clone https://github.com/labs-barkley/barkley-reference-architecture
cd barkley-reference-architecture
python -m pip install -r requirements.txt
python examples/run_reference_pipeline.py

This dataset is part of the Barkley Canine Cognition Lab public research ecosystem.

---

Limitations

• No real-world validation. Trajectories are generated from a parameterized model and have not been validated against empirical canine behavioral data.
• Single-axis behavioral proxy. Real behavioral states are multivariate. This dataset uses composite scores as proxies, a deliberate simplification for modeling clarity.
• Controlled noise structure. Generative noise does not fully replicate non-stationary structures found in real sensor data.
• No ground-truth health outcomes. Drift indicators are structural, not clinically grounded. No causal linkage to health events is implied.

These limitations are features of a controlled research instrument.

---

Citation

@dataset{remoissenet2026doggraph,
  author    = {Remoissenet, Elodie Aishwarya P.},
  title     = {Synthetic DogGraph Sample —
               Longitudinal Canine Behavioral Trajectories},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.20356188},
  url       = {https://doi.org/10.5281/zenodo.20356188}
}

Framework paper:

@article{remoissenet2026framework,
  author = {Remoissenet, Elodie Aishwarya P.},
  title  = {From Surveillance to Cognition: A Unified Framework for
            Precision Behavioral and Metabolic Intelligence
            in Companion Animals},
  year   = {2026},
  doi    = {10.5281/zenodo.20060327},
  url    = {https://doi.org/10.5281/zenodo.20060327}
}

---

Links

• 🌐 Barkley AI — https://getbarkley.com
• 📦 Barkley Canine Cognition Lab — https://github.com/labs-barkley/barkley-canine-cognition-lab
• 🧠 Reference Architecture — https://github.com/labs-barkley/barkley-reference-architecture
• 🤗 Dataset — https://huggingface.co/datasets/labs-barkley/synthetic-doggraph-sample
• 🎯 Drift Explorer — https://drift-explorer.getbarkley.com
• 📄 Framework Paper — https://doi.org/10.5281/zenodo.20060327
• 🆔 ORCID — https://orcid.org/0009-0004-6031-659X

---

© 2026 Barkley AI. All rights reserved. Synthetic data only. Not a veterinary or medical diagnostic instrument.