How to evaluate a Hyperliquid copy trading system: a practical checklist

Why performance numbers alone are insufficient

Every copy trading product leads with a return figure. It is the easiest thing to display and the least informative thing to evaluate, because the number describes a path that has already happened under conditions that will not repeat.

This is not a claim that performance is irrelevant. It is a claim about ordering. Performance tells you what a system did; structure tells you what it can do to you. Structure is stable, checkable, and knowable in advance. Performance is none of those things.

Survivorship, sample length and unrepeatable regimes

A displayed track record is almost always a surviving track record. Strategies that blew up are not on the page, and the accounts that would have shown the failure mode most clearly are precisely the ones that stopped reporting.

Sample length compounds the problem. On perpetual futures, a few weeks of directional beta in a trending market produces curves that are visually indistinguishable from skill. Distinguishing the two requires a sample that spans at least one regime change, and most published windows do not.

Ask what market conditions produced the record. A number earned entirely in a one-directional market tells you almost nothing about what happens in a chop or a violent reversal.

• Short windows over-represent luck and beta.
• Failed accounts are silently excluded from most displays.
• A record earned in a single regime does not generalise to another.

The number you can see is not the number you would have earned

Even a genuine leader record is not your record. You enter later, at a different price, with a different account size, in a different margin state, and you pay costs the leader did not.

The gap between the two — tracking error — is structural rather than occasional. Any system that presents leader performance as follower performance, without stating that gap, is either not measuring it or not disclosing it.

So the first checklist question is not "what did it return?" It is "what would I actually have received, after divergence and costs, and can that be shown?"

Custody and permission model

Start here, because nothing further matters if you get this wrong. There are only two states: either your capital sits in an account you control, or it does not.

If capital moves to a pooled account, a company wallet or an exchange sub-ledger you cannot withdraw from unilaterally, the system is custodial. That may still be an acceptable trade — it simplifies operations — but you have added counterparty risk on top of market risk, and you should price it as such.

If capital stays in your own Hyperliquid account, the question becomes what the granted permission can sign. On Hyperliquid the relevant object is an agent approval. A correctly scoped one is trade-only agent approval; no withdrawal or transfer rights — it can place and manage orders and nothing else.

A good answer is specific and verifiable. A warning sign is vagueness: "secure", "bank-grade", "your keys stay safe" are not permission descriptions. Ask exactly which actions the granted key can sign, and how you revoke it without asking anyone.

• Where does the USDC sit while the system is running?
• Which actions can the granted permission sign — orders only, or transfers too?
• Can you revoke unilaterally, on-chain, without a support ticket?
• What happens to open positions the moment you revoke?

Position isolation and netting risk

A perpetual-futures account holds one net position per market. If two leaders take opposite sides of the same market in one account, the exchange does not record two trades — it records the difference. Both leaders can be right and you can end up flat, having paid fees to get there.

So ask where each mirrored leader's positions live. One account for all leaders means netting is not a risk, it is a certainty whenever leaders disagree. Separate accounts — on Hyperliquid, sub-accounts — keep each leader's book intact and give you per-leader attribution.

The isolation unit worth asking about is one Hyperliquid sub-account per mirrored leader. Anything coarser than that means positions can silently cancel, and it also means a single margin pool backs every leader, so one leader's liquidation consumes margin the others were relying on.

Isolation costs something: margin is fragmented and cannot be shared. A system that isolates should say so plainly rather than presenting it as free.

Trader selection and replacement process

Selection is where most systems are weakest, because it is where marketing language substitutes most easily for method. "Top traders", "verified pros" and "elite performers" are labels, not criteria.

A serious answer names the inputs and the floors. Which metrics enter the score, how are they combined, what disqualifies a candidate regardless of returns, and over what window is each measured? Realized PnL consistency, win rate, profit factor, position discipline and account survivability are the kind of inputs that can be computed from public history and checked.

Replacement matters at least as much as selection, because edge decays. Ask what triggers a leader being dropped, how quickly, and what happens to their open positions when it happens. A well-designed basket is sticky by default — a higher score elsewhere should never on its own be a reason to drop someone — and should name the specific emergency and soft-issue conditions that do force a removal.

The warning sign here is discretion without disclosure: leaders changed for unstated reasons, or a roster that only ever gains members and never loses them.

• Are the selection inputs named and computable from public data?
• Is there a floor that disqualifies a high-return account?
• What triggers a soft-issue strike, and what triggers immediate emergency removal?
• Are open positions closed, held or handed to a replacement leader?

Fee structure and the incentives it creates

Read a fee model as an incentive statement, not as a price. Each structure rewards a different operator behaviour, and that behaviour lands in your account.

A volume-based fee earns more when the system trades more, so it rewards turnover. That is honest and bounded, but it means you should check whether anything constrains churn.

A profit share earns on gains and does not refund losses, which rewards volatility: a leader paid a share of upside has an incentive to take risk that a leader carrying the downside would not. High-water marks reduce but do not eliminate this.

A flat subscription is the most neutral on turnover and the least aligned on outcome — the operator is paid identically whether you make or lose money. Spread markups are the least transparent, because the cost is embedded in your fill price and never appears as a line item.

For reference, the model used here is 0.1% of mirrored notional volume, charged through Hyperliquid's native builder-fee mechanism with a maximum rate you approve up front. The point of stating it is not that it is the only defensible model — it is that you should be able to state any system's model in one sentence, including the cap.

Risk controls and kill switches

Ask what the system refuses to do. A copy system with no constraints is a leverage amplifier attached to someone else's judgement.

Concrete controls have numbers attached: a per-leader notional ceiling, a leverage cap applied independently of what the leader uses, a limit on how much of the account any one sleeve can consume. Vague controls — "advanced risk management" — are not controls.

Then ask about the kill switch, and be precise about who holds it. There are three distinct capabilities and a system may offer only some: pausing new entries, closing existing positions, and revoking the permission entirely. Only the last one is fully in your hands if the operator is unreachable.

The test that matters: if the operator's infrastructure went dark right now, could you stop the system and manage your positions without their cooperation?

• Per-leader notional ceiling — is there a number?
• Leverage cap independent of the leader's own leverage?
• Can you pause new entries without closing existing ones?
• Can you exit entirely if the operator is offline?

Transparency and verifiability

The final property is whether you can check the previous six yourself rather than taking them on trust.

Hyperliquid is unusually good for this. Positions, fills, balances and agent approvals are visible on-chain against addresses. If you know your own address and your sub-accounts, you can confirm what was traded, when, at what price, and what permissions exist — without any reporting from the operator.

So ask which claims are verifiable and which are assertions. Leader addresses, your own fills and your own approvals are verifiable. Internal scoring, forward-looking replacement policy and aggregate model performance are assertions — reasonable ones, but assertions.

A system that distinguishes between the two in its own documentation is telling you something useful about how it will behave when something goes wrong.

The full checklist

The table below condenses the seven properties into a form you can work through in about twenty minutes for any system, this one included. The verification column matters most: a question you cannot check independently should be weighted lower than one you can.

What the checklist cannot tell you

A system can answer every question well and still lose money. Structure bounds the failure modes that come from design; it does nothing about the market.

The checklist is also blind to execution quality that only shows up under stress — how fills behave in a thin book, how quickly a replacement actually happens, whether an operator's infrastructure holds up during a volatility spike. Those are learned from operation, not from evaluation.

And it cannot price leader decay. Every leader eventually stops working. A good replacement process — sticky membership, immediate emergency exits, and soft issues that need repeated strike-days before they force a change — shortens the lag between decay and removal; nothing removes the lag entirely.

Trading perpetual futures carries a substantial risk of loss, including the loss of your entire position. No structural property changes that.

Conclusion

Performance is the output of a system under conditions that have passed. Structure is the system. If you evaluate structure first, the performance figure becomes what it should be — one input among several, interpreted in the light of how the thing is actually built.

If you want to run this checklist against the mechanics described here, the how-it-works page states the custody model, permission scope, isolation unit and fee in the same terms used above, and the risk page states the limits plainly.