On-chain alert feeds are noisy by default. A wallet moving $40 million between two addresses it already controls is not a market event, yet it lights up every generic tracker. The first job of a serious whale tracker is classification, not detection.
NORVENTA classifies transfers into four buckets: exchange inflow, exchange outflow, wallet-to-wallet, and liquidity events. Exchange inflows increase immediately sellable supply and are the most price-relevant category. Outflows to self-custody usually reduce sell pressure. Wallet-to-wallet movement between clustered addresses is generally noise.
Size alone is also a weak filter. A $500,000 transfer in a token with $2 million of daily volume matters far more than a $50 million BTC transfer. We normalise transfer size against the asset's real liquidity — the depth actually resting within a few percent of mid — rather than against reported volume, which is easy to inflate.
The final layer is timing. Clustered inflows from multiple unrelated wallets inside a short window are the highest-signal pattern in the dataset. They indicate coordinated distribution or a market maker preparing to work a large order, and they precede volatility far more reliably than any single transfer.
Whale data is context, not a trade. Used correctly it tells you which direction has the supply problem before the chart does.