6 min read · 1290 words
On most exchanges, “smart money” is a black box: you can guess what large traders are doing from funding rates and aggregate ratios, but you can never see the actual positions. Hyperliquid breaks that wall. As an on-chain perpetuals exchange with a public leaderboard and fully transparent wallet positions, it turns the question “what are the best perp traders actually holding right now?” into something you can literally look up. This post explains how to read that smart-money positioning signal, how it fits a confluence model, and how Huginai aggregates it.
Hyperliquid is an on-chain perpetual-futures DEX. Trades, positions, and liquidations settle on its own chain rather than inside a centralized exchange’s private database. That architectural choice has a side effect that matters enormously for signal hunters: positions are public by default.
Two things follow from that:
On a centralized venue, this data simply doesn’t exist for outsiders. On Hyperliquid it’s free and falsifiable: anyone can verify the same wallet you’re looking at. That makes it one of the cleaner public smart-money signals available, in the same family of evidence as the wallet-level on-chain flows covered in our guide to whale signals.
One profitable trader being long ETH tells you very little. The signal lives in consensus across many top traders. The job is to turn a list of individual positions into a readable aggregate, which means being careful about how you count.
Gross notional is the total size of all positions in a coin regardless of direction — it tells you how much attention the coin is getting. Net notional is longs minus shorts — it tells you which way the crowd of top traders actually leans. A coin can have huge gross interest but a net near zero, which is a genuinely different (and more ambiguous) state than a coin where net is heavily skewed long.
Beyond notional, count how many distinct top traders sit on each side. Ten profitable accounts independently long is a stronger consensus than one very large account long against nine small shorts. Two dimensions are worth separating:
When breadth and skew point the same way, that’s the high-conviction case. When net notional is heavily long but only because of a single outsized account, you should treat it as one trader’s bet, not a consensus.
Position size is not the same as conviction. A large notional can simply reflect a large account, leverage, or a hedge leg against a position held elsewhere. Always normalize against breadth and against the trader’s own account size before reading a big number as a strong opinion.
Leaderboard positioning is one input, not a complete thesis. Its real value shows up when it agrees or disagrees with the other signals you’re already watching:
The point of confluence is not to add signals until one says “buy.” It’s to look for cases where several independent lenses describe the same picture — and to notice, just as importantly, when they contradict each other.
Huginai’s chain-watcher does this mechanically:
What that fed into is the part worth stating plainly. Positioning consensus was one input to a directional conviction score, and that book is closed. Three cohorts, a measured 40% win rate against the 57.4% the exit structure needed to break even, and no positive slice at n≥5 on any axis — including conviction bucket, which turned out to be anti-predictive. Hyperliquid positioning did not cause that result and it did not rescue it either. Read the aggregation above as a way to read the venue, not as evidence that reading the venue makes money.
The Hyperliquid workstream still running is a different one: a paper funding-carry basket, forward-tested against a kill rule. Carry, not direction, and paper only.
Transparent does not mean infallible. Keep these limits in view:
Used as a transparent, falsifiable input inside a broader confluence model, leaderboard positioning is a genuinely useful free signal. Used as a copy-trade button, it’s a fast way to get squeezed. None of this is financial advice — it’s a description of how to read the data.
A research project, not a product you can buy. There is no signup and no account. The leaderboard reader and per-coin aggregation described above run in the repo — they are not a service you can subscribe to. The dashboard at /app is a demo on synthetic sample data, there to show the interface and nothing more.