Guides / Quant Modeling
Anomaly detection on prediction market data — spotting a sudden depth withdrawal, an unusual trade cluster, or a spread that blows out ahead of resolution — usually starts with a baseline: what does "normal" orderbook and trade activity look like for this market, and how far does a given snapshot deviate from it. Simple z-score thresholds work for a first pass; autoencoders (reconstruction error as the anomaly signal) generalize better across markets with different liquidity profiles.
ProbSights provides the raw orderbook and trade history — bid/ask depth, mid-price, spread, and executed trades across Kalshi and Polymarket — that feeds either approach. Whale trade tracking already flags one specific anomaly type (unusually large fills) as a built-in feature; everything else runs on data pulled through the API.
Data from Kalshi and Polymarket. Search live API docs Pricing