BLOODHOUND is a multi-layer behavioral detection system that identifies coordinated inauthentic behavior (CIB) on social media platforms. It analyzes when and how accounts post — not what they say — to find clusters of accounts operating with statistically similar behavioral patterns.
The system does not determine intent, origin, or affiliation. It surfaces behavioral patterns that warrant human investigation. All detections are probabilistic, not definitive.
Measures how predictable each account's posting schedule is using Sample Entropy (SampEn). Values below 0.2 indicate highly regular, clock-like posting consistent with automation. Requires ≥200 events per account.
Builds a 480-dimensional behavioral distribution for each account (5 action types × 96 time bins), computes pairwise Jensen-Shannon Divergence, and uses the Leiden algorithm to identify communities of behaviorally similar accounts. Includes expansion analysis for accounts with 50–199 events.
Identifies accounts that share identical content (via hash matching) with accounts flagged by L1+L2. Recovers low-volume operators that evade behavioral detection by linking them to the detected network through content co-sharing. Includes Transfer Entropy analysis for temporal influence patterns.
Uses Claude API to classify narrative themes, fingerprint coordination patterns, and assess remaining undetected accounts. Provides human-interpretable explanations of what coordinated accounts are doing, not just that they're coordinated. Requires API key.
Submit a new analysis to get started.