WhaleWatch detects unusually large public positions relative to the exact market in which they occurred. It does not label every large dollar trade a signal, infer a trader's identity or claim that an outlier predicts the winning outcome.
Market metadata comes from Polymarket's public Gamma API. Public taker-side fills and wallet activity come from Polymarket's public Data API and public blockchain sources. Source availability, API coverage and indexing delays can create gaps, so every result must be read with its displayed coverage context.
Related fills from the same public wallet, condition market, outcome and BUY or SELL direction are joined into a position episode. Grouping prevents one economic action split across multiple fills from being mistaken for many independent signals. BUY and SELL remain separate because a sale can reduce an older position rather than express a fresh directional view.
Each episode is compared only with the normal size distribution of its exact condition market. The threshold combines robust distribution statistics—the median, 95th percentile and log median absolute deviation—with a liquidity-scaled dollar floor between approximately USD 1,000 and USD 10,000. This avoids comparing a thin niche market with a highly liquid national event.
LOW, MID and HIGH describe how unusual position size is relative to the market baseline; they are not probabilities of success. Strict alerts also require freshness, adequate data coverage and meaningful wallet dominance, including the wallet's share of observed BUY flow and comparison with the next-largest observed wallet.
The reported whale entry is the volume-weighted average execution price of the grouped episode. The current outcome price is shown separately. A large move after detection may make an observation stale or less actionable even when the original anomaly remains historically valid.
Eligible signals receive a detection-time snapshot containing the model version, threshold, position value, outcome, direction, entry price and available coverage evidence. Later evaluation is appended without replacing that snapshot. Open outcomes stay open; misses are retained alongside hits.
Questions or correction requests: contact the methodology team.