Wallet attribution · Sui · with published label accuracy

Who actually traded.

Price and volume tell you a coin moved. They do not tell you who moved it. Every DeepBook fill is read from Sui and traced back to the wallet behind it, then those wallets are grouped by behaviour — so a candle can be split by the kind of trader who filled it.

Method

How the numbers are made

  1. Step 1

    Read every fill from the chain

    DeepBook fills are read straight from Sui, and each balance manager is resolved back to the wallet that owns it. Nothing is inferred from price.

  2. Step 2

    Label as of each trade

    A wallet is labelled only from the trades it made before that moment — a count-based rule from its own history, a ratio-based one from the trailing 30 days. History never peeks ahead.

  3. Step 3

    Group the volume

    Each trade is attributed to its wallet’s label at that instant, so a candle can be split by who actually filled it. Market makers are shown separately because they trade both sides.

  4. Step 4

    Check the labels against people

    Rules can be wrong, so wallets are reviewed by hand and scored: precision and recall per class, plus agreement where two reviewers judge the same wallet. Those scores are published below.

Attribution, not prediction. We show who traded and how sure we are of each label; we do not claim to know what any group will do next. Refreshed hourly.