Six stages. One pipeline. Zero data stored.
Human and AI identities coexist in one protocol.
These are the concrete outputs of the MyShape pipeline. This is what your application receives.
// Motion Vector hash — the geometric signature
{
"mv_hash": "0x7a3f1b2c8d4e9f01...3a6b5c7d2e8f4a9b",
"dimensions": 128,
"feature_groups": ["kinematics", "acceleration", "jerk", "spectrum"],
"sst_topology": "18-point",
"poseidon_params": { "t": 5, "rounds": 80, "field": "BN254" },
"non_invertible": true
}// ZK-Presence composite proof
{
"proof_type": "ZK-Presence",
"pes_score": 0.94,
"pes_components": { "timing": 0.91, "noise": 0.96, "frequency": 0.89, "biological": 0.97 },
"timestamp": 1719000000,
"session_id": "sess_7b3f2a1c...",
"zkp_hash": "0x9f2a8b1c3d4e...",
"proof_size": 486,
"verification_ms": 0.8
}Zero overlap between distributions. AI cannot cross the entropy threshold — mathematically provable via §10.
Human motion carries irreducible biological entropy. AI-generated motion is mathematically clean — and that cleanliness is detectable.
PES Score threshold for human verification: ≥ 85%
Four independent dimensions. Each one alone can flag synthetic motion. Together, the false-positive rate approaches zero.
| Dimension | Human | AI |
|---|---|---|
| D1 — Micro-Timing | Chaotic — breathing, fatigue, neural jitter | Uniform — generated frames have consistent timing |
| D2 — Noise Residual | Organic — CMOS noise interacts with real light | Absent — perfect coordinates are a red flag |
| D3 — Frequency Entropy | Broad spectrum — multiple overlapping frequencies | Narrow peaks — limited frequency range |
| D4 — Biological Perturbation | Present — humans cannot suppress these | None — AI does not model involuntary biology |
7 known attack vectors. 7 verified defenses. Each independently tested against real-world threat models.
| Attack | Defense | Detection |
|---|---|---|
| Generative AI | 4D entropy gap — AI cannot reproduce biological micro-timing | 99.2% |
| Replay Attack | Temporal nonce + frame freshness check | 99.8% |
| Imitation | Motion signature is individual — imitation ≠ original | 98.5% |
| Motion Capture | Sensor noise profile verification | 99.0% |
| Sensor Spoof | Device attestation + hardware fingerprint | 99.5% |
| Adversarial Pose | Kinematic consistency check — joints must obey biomechanics | 98.7% |
| Statistical | Frequency-domain analysis exposes synthetic patterns | 99.1% |