Human Continuity Dataset
The primary research asset of The Continuity Lab. Growing monthly.
Current Snapshot
Composition
576 experimental runs across 4 evidence engines (EE-001–EE-003 + VS-001), as of 2026-07-19. Point-in-time total; newer runs are not yet reflected in this number.
Human motion. 30-second unstructured motion sequences. Standard webcam at 30 fps. MediaPipe Pose (33 landmarks). No choreography, no controlled environment. Ecological variability is intentional.
Synthetic motion (200 samples, 50 per strategy).
1. Random walk — Gaussian per-joint walk calibrated to human joint range.
2. Spline interpolation — Human pose keyframes connected via Catmull-Rom splines.
3. GAN-generated — Conditional GAN trained on human pose sequences.
4. Near-static — Minimal displacement with escalating Gaussian noise.
All samples are anonymized. Joint coordinates only. No face data, no identity-linked templates, no PII. The dataset collects geometry, not identity.
Growth Timeline
Access
The dataset is currently used internally for benchmark development. We plan to release a public subset for peer review and independent verification. If you are a researcher interested in access before the public release, contact us via GitHub.