We run multi-layer checks on creator accounts and audiences - behavioral signals (sudden follower spikes, engagement anomalies), audience composition (gender, geo, language consistency), normalized ER vs. posting cadence, and engagement distribution over the last 30 days. Semantic flags and fraud thresholds mark suspect followers so you see a clear audience-quality score.