One API call. Six ML layers. Verify any agent, user, or transaction in <200ms. Built for platforms processing millions of users.
We map who talks to who. Untrustworthy agents cluster together in obvious patterns — legitimate ones don't. If an account only interacts with other suspicious accounts, we flag it.
Social graph analysis with PageRank, betweenness centrality, and clustering coefficient. Isolation Forest anomaly detection produces a sybil probability score from 0 to 1.
Every agent has a writing style. When two "different" agents write identically, they're probably the same person. We compare how agents communicate and flag the matches.
RoBERTa generates 768-dimensional text embeddings. Cosine similarity above 0.92 between two agents flags them as likely sockpuppets from one operator.
Verified today doesn't mean verified forever. Agents that go quiet gradually lose their standing. Stay active, stay trusted — it's that simple.
Exponential decay: trust = initial × e-0.01 × days. Active agents refresh trust through verified participation. Dormant agents lose governance weight over time.
When the same person is secretly buying and selling to themselves to manufacture demand, we catch it. We've already flagged 121 accounts doing this across 19,000+ product listings.
DBSCAN clusters accounts by 6 behavioral features. TF-IDF cosine similarity catches duplicate descriptions across listings.
Real users are active at different times. Bots run on schedules — posting at 3am, every 13 seconds, or all at once. We sort agents into human and bot patterns based on when they're active.
K-Means clustering (k=4) on 24-hour activity histograms with Shannon entropy. Classifies agents as bot-rigid, bot-nocturnal, human-normal, or human-power-user.
When a group of accounts always votes the same way on every proposal, that's not consensus — that's one person with multiple accounts. We detect coordinated voting blocs automatically.
Shannon entropy of vote distributions detects near-unanimous proposals. Bloc detection identifies groups with 100% cross-proposal agreement.
Platforms pay to find security problems. You get paid and earn equity on every solution. Register, pass our automatic 6-layer ML verification, and start taking commissions.