How the Top Traders ranking works

A tracked wallet is ranked only when it passes every rule below. The first rule it fails is shown on its wallet page.

1. Lifetime gate. Lifetime perps PnL on Hyperliquid divided by current perps equity must not be below -0.5. A wallet that has lost more than half its equity over its life can still post a good month: it stays out of the ranking, and its page says why.
2. Directional style. Market makers, wallets whose positions mostly hedge each other and wallets whose PnL on long holds comes mostly from funding are left out: copying one side of their book does not reproduce what they earn.
3. At least 6 effective episodes. An episode is a position on one coin from flat to flat. Live episodes count in full; episodes rebuilt from older public fills count less (see below). The weighted sum is the number of effective episodes.
4. A follower would have made money. Over the last 90 days, a follower copying every copyable episode 10 minutes late must end with a positive net result after the modeled cost of the delay.
5. At least 60% of episodes copyable. An episode is copyable when it stays open longer than the 10-minute gap. A wallet that closes most positions faster than a follower can react is not ranked.
6. A position change in the last 14 days. Dormant wallets leave the ranking until they trade again.
7. Perps equity of at least $250k. Small accounts are left out: their results are easier to produce by chance and harder to copy at size.

The Follow Score goes from 0 to 100. It combines five measurements of the follower curve, with fixed weights:

• 35%Follower edge. What a follower keeps per episode at the 10-minute gap, as return on equity, after the modeled delay cost. Taken at its lower bound (below).
• 20%Sortino ratio. Return of the follower curve against its downside swings only.
• 15%Max drawdown. The largest drop of the follower curve from a peak. Smaller is better.
• 15%Consistency. How many weeks end positive, with a penalty when the result rests on a few trades.
• 15%Copyability. The share of episodes a follower can enter in time, and the visible book depth on the coins the wallet trades.

Every wallet is scored with the same reference gap of 10 minutes, whatever its own detection delay, so two wallets are always compared on the same footing.

History older than the tracking is rebuilt from the public fills of each wallet over the last 90 days. For those episodes the follower result is estimated with the delay cost measured on live copies, and they count with a reduced weight (0.5). Episodes from the 30 days before a wallet entered the tracked set count even less (0.3): a wallet is often picked up because it just did well, and that stretch would flatter it.

With few episodes the score leans on a prior taken from the wallet's lifetime portfolio record, worth 8 episodes. As live and rebuilt episodes add up, the wallet's own record takes over.

The follower edge, the largest part of the score, is not the average result per episode. It is the lower bound of a one-sided 80% confidence interval around that average, given how many episodes back it. Two lucky trades give a wide interval and a low lower bound; thirty steady episodes give a narrow interval and a lower bound close to the average.

The effect is the one a follower needs: the ranking rewards an edge that holds up under uncertainty, not a short streak. The rank itself follows the score smoothed over 24 hours, so one update does not reshuffle the list.

A ranked wallet is Verified when its Follow Score is at least 70, its confidence is medium or high, and its style is known. Confidence grows with the number of effective episodes and the days they cover. Verified is a reading aid, not a guarantee.

Buildix shows data. It does not execute trades, does not hold funds and does not recommend a trade. The follower result is a model with a fixed gap and an estimated cost: your own fills, size and timing will differ.

Past performance does not predict future results. Buildix shows data, it does not execute or recommend trades.