Forensic Dossiers
Multi-layer cryptographic evidence of training data provenance. Each dossier combines structural probes, behavioral analysis, geometric DNA, zero-knowledge proofs, and adversarial investigation — autonomously generated, machine-verifiable.
Completed Dossiers
—Pipeline Queue
—Evidence Methodology — 6 Independent Layers
Structural Probe
Sonar v2 latent crystallography — detects domain-specific structural patterns in model embeddings without accessing training data.
Behavioral Analysis
Calibrated Differential Surprise (CDS v3.1) — measures statistical deviation from reference models on dataset-specific texts with bootstrap CI.
Geometric DNA
Cross-model weight topology comparison via Mantel test — detects shared training data lineage across model families.
Zero-Knowledge Proofs
ZK-GDNA: Fiat-Shamir challenge on secp256k1 with Merkle commitment — proves fingerprint existence without revealing weights.
Adversarial Investigation
Prosecutor vs. Defense agents, 5 rounds, challenge mechanism — verdict survives adversarial scrutiny.
Temporal Anchoring
Ed25519 signatures + Cloudflare witness timestamps — every data point is cryptographically time-locked.