Scoring methodology

provably good, not lucky

Every number below is computed deterministically from public Hyperliquid data — fills, account state, and candles. No AI judgement, no manual labels, no photoshopped equity curves. Anyone can recompute it.

The smart score (0–100)

The score is a weighted composite of the criteria below, computed over closed trades reconstructed from the wallet's public fill history (a trade runs from a flat position to the next flat position; direction flips split into a close plus a re-open). Open positions and unrealised PnL never enter the score.

When a criterion cannot be computed honestly — for example candle data is missing for a coin — it is dropped and the remaining weights are re-normalised. We never fill unknowns with zero, and we never fake a sub-score.

CriterionWeightWhat it measures
Realised edge28%Average return-on-notional per closed trade (realised PnL ÷ entry notional). Size-normalised, so a $100 win on $1,000 counts the same as a $10,000 win on $100,000.
Hit rate (Wilson)16%Share of positive trades with a Wilson 95% lower bound, which penalises small samples: 7/10 wins scores far lower than 70/100.
Consistency14%Sharpe-like mean ÷ standard deviation of per-trade returns. Steady edge beats one lucky spike.
Market timing12%Entry/exit versus the local candle range: longs buying low and selling high (shorts the reverse). Separates timing skill from being leveraged-long in a bull market.
Drawdown control12%Maximum drawdown of the sampled account value (or a per-trade proxy when no equity history exists).
Size-adjusted edge10%Notional-weighted return: does the edge survive on the big trades, or only the small ones?
Liquidation avoidance10%Liquidation events observed for this wallet, penalised relative to trade count.
Asset specialisation6%Concentration of trades by coin (HHI). Focus is evidence of a repeatable playbook.
Captured move / exit quality / adverse control / trade efficiency6% eachCandle-based trade quality: how much of the favorable move was captured, what happened after exits, average adverse excursion, and reward-to-heat. Only computed where candles exist.
Out-of-sample stability4%Edge on the held-out most-recent trades (see the gate below).

The gate: a score is hidden until it is proven

Scan enough wallets and pure randomness produces "stable winners". To avoid labelling luck as skill, a wallet's score stays hidden (lane: tracking) until one of two gates passes:

Strict gate — at least 30 closed trades, andthe most recent 30% of trades (a forward hold-out the wallet "hasn't seen") show a positive edge on their own, and the older 70% do too. A hot streak that stops working never earns the label.

Patient gate — for low-frequency swing traders: at least 12 closed trades with market timing ≥ 65/100 on well-covered candles, Wilson hit-rate ≥ 50%, a positive median trade (the edge cannot ride on one outlier), max drawdown ≤ 35%, and a non-negative forward edge.

Every gated wallet shows its exact gate reason on its page — which threshold it has not met yet.

Smart ≠ hot ≠ whale

Smart is the full-history, gate-passed score above. Hot is recent-window performance only (published only for a genuinely positive window: positive average and median and realised PnL, with enough trades) — a hot wallet never inherits the smart label. Whale is size, not skill. Market-makers (two-sided, low-net-exposure flow) are lane-separated and never ranked as directional smart money.

Why our ranking differs from PnL leaderboards

Most trackers rank by raw PnL or ROI, so three lucky trades put a wallet on top. We deliberately do not: a wallet with a huge PnL but too few closed trades, a negative forward hold-out, or an outlier-driven edge stays in the tracking lane with its score hidden. If a wallet tops a PnL leaderboard elsewhere and is unranked here, that is the gate working — not missing data. The trade-off is honesty over flattery, and it is the entire point of this product.

Data sources and limits

Inputs: the public Hyperliquid info API (fills, account state, funding, portfolio) and public candles. Fill history is backfilled in bounded windows, so very deep histories may be partially covered at first — coverage is always shown next to the score, and scores refine as backfill deepens. Nothing on the ranking path fetches at request time; you always see a reproducible snapshot.

Verify any wallet yourself: open its page at /w/<address> — every sub-score shows the basis it was computed from. Back to Smart Money →