Research
How we tell
humans from bots.
CertiLayer's research spans behavioral machine learning, cryptographic verification, and low-latency systems engineering — published here as it happens, with real numbers, not projections.
Behavioral ML
The four-stage detection pipeline — rule-based gates, a Siamese LSTM for identity, an XGBoost classifier, and contrastive autoencoders for anomaly detection.
Cryptographic Trust
BLAKE3 key hashing and a Merkle-tree audit ledger that makes every scored session independently verifiable and tamper-evident.
Systems Performance
A Rust ingestion pipeline built to score behavioral signal in single-digit milliseconds, measured under real concurrent load — not simulated.
Latest
The Human Confidence Score: a passive approach to bot detection
The full methodology behind CertiLayer's four-model pipeline, the 34-signal feature schema, and how the HCS formula was derived and validated.
Ingestion latency under load: 11ms p95 at 100 concurrent users
Real k6 load-test results against the production ingestion endpoint, plus validation AUC/AUROC scores for every model in the pipeline.