One pass over the pre-aggregated per-wallet rollup — never the raw fills table — keeping every wallet that clears all of the following. Malformed rollup keys are rejected before the floor is applied.
≥ 20 resolved bets
positive PnL
bots and market makers excluded
first seen within 90d
Ranking metric
Members are ordered by realized PnL (our ledger), descending. Ranking is profit-first by design: win rate is a sample-quality reading here, not a gate, and no cohort on this site is selected on win rate alone.
Why win rate is shown shrunk
The column shows a sample-size-shrunk win rate with its confidence bar underneath, over the resolved-bet count on its left. The raw per-fill rate is deliberately not displayed: on small samples it reads far too high, and on this rollup it is stale. A shrunk rate over a stated denominator is the honest form of the same number.
What this is not
Membership is recomputed on every read. A wallet that stops clearing the floor leaves, and nothing records that it was ever here — this is a live filter, not a historical record or a hall of fame. The figures come from a rollup that lags the chain; the panel above states how stale it is.
Top 32 members by PnL, square-root scaled — the top wallet clears the median member by orders of magnitude, and a linear axis would hide every bar but one.
