We flag inauthentic activity by cross-checking creator behavior, audience make-up and engagement patterns over the rolling 30-day window. Signals include sudden follower spikes, inconsistent ER vs. views, geographic/language mismatches, recycled commenters, low reach estimates and abnormal growth rates. Profiles with such anomalies get fraud scores for fast vetting.