How to allocate capital across multiple Hyperliquid traders

Allocation is the decision, not selection

Assume you have done the selection work: several leaders clear your floors on consistency, payoff, position discipline and survivability. You now face a harder question, and it is one that has a mechanical answer rather than a judgemental one.

How much capital does each leader get, how small can a sleeve be before it stops working, and what do you do when the weights drift away from where you set them?

These questions have concrete answers on Hyperliquid because the constraints are concrete: minimum order sizes, per-market margin requirements, and the fact that isolated sub-accounts cannot borrow from each other.

Why equal weighting is usually suboptimal

Equal weighting feels neutral. It is not: it is an active claim that every leader in your basket has the same expected quality, which is a claim your own selection process contradicts. If you scored them, you already believe they differ.

It also has a mechanical cost. The weakest leader in an equally weighted basket receives the same capital as the strongest, so the basket's result is dragged toward the median of the set rather than toward the top of it.

The deeper argument for weighting by evidence — and the caps that keep it from becoming concentration — is set out in the note on score-weighting versus equal-weighting. This note takes that argument as given and deals with what happens next: the sizing, fragmentation and drift problems that appear once weights exist.

Score-weighted allocation in practice

In practice, weighting means each leader's share of capital is proportional to their composite score: capital weighted in proportion to each leader's composite score. A leader that clears the floors comfortably on every input receives more than one that scrapes through on two of them.

Two implementation details matter more than the weighting formula.

The first is a cap. Without a maximum weight, a scoring system that rates one leader highly will reproduce single-trader concentration under a different name. A cap makes the basket's diversification a property of the structure rather than an accident of the score distribution.

The second is a floor. Below some weight, a sleeve cannot hold a position large enough to matter, and it should not exist. When a leader's proportional weight falls below what a functioning sleeve requires, the correct action is fewer sleeves at larger size, not a token allocation.

The concentration–diversification trade-off

Adding leaders reduces the impact of any one of them being wrong, and it dilutes the impact of any one of them being right. Both effects are real and both scale with the number of sleeves.

The benefit is also not linear. Going from one leader to three removes most of the single-point-of-failure risk. Going from seven to ten changes the risk profile far less while adding turnover, fee drag and margin fragmentation.

Correlation determines where the useful ceiling sits. If three of your leaders trade the same two perps in the same direction, you have three sleeves and roughly one bet. Breadth of markets and difference in holding period are what make an additional sleeve additive rather than duplicative.

Minimum viable size per leader

This is the constraint that most allocation plans ignore, and it is the one that actually binds for small and medium accounts.

Minimum order size and rounding

Hyperliquid perps have minimum order sizes and price and size increments. A mirrored order is derived proportionally from the sleeve's capital, then rounded to a valid size.

When the sleeve is small, rounding is no longer a detail. A position that should be 1.4 units becomes 1 or 2 — a 30% or 40% sizing error on that trade. The sleeve is no longer tracking the leader's risk profile; it is tracking a coarse approximation of it, and the approximation error can exceed the edge being copied.

Below a certain sleeve size, some of the leader's trades cannot be mirrored at all because the proportional size rounds to zero. You then hold a filtered subset of their strategy chosen by rounding rather than by intent.

Margin fragmentation across isolated sub-accounts

Isolation is what makes multi-leader copying coherent — one Hyperliquid sub-account per mirrored leader means opposing positions do not cancel. It also means margin cannot be shared. Each sleeve must independently hold enough collateral to open its positions and survive normal adverse movement.

The aggregate effect is that a fragmented account holds more idle margin than a single cross-margined account with the same exposure. That idle margin is the price of containment, and it should be budgeted for rather than discovered.

It follows that sleeve count is capped by capital, not by preference. Ten sleeves funded below their minimum viable size is worse than three funded properly: you pay ten sets of fees for three sleeves' worth of functioning exposure.

Why sleeve count should follow capital

This is the reasoning behind gating full diversification behind volume rather than offering it to everyone immediately. Below $100,000 of mirrored volume, a single well-funded sleeve tracks its leader far more faithfully than several underfunded ones track theirs. Above it, up to 10 sleeves can each carry meaningful size.

The threshold is not a paywall dressed as a feature. It is the point at which the arithmetic of minimum sizes and per-sleeve margin stops working against you.

Drift and rebalancing

Weights do not stay where you set them. A sleeve that performs well grows as a share of the account; one that loses shrinks. Left alone, the basket drifts toward whichever leader has been recently lucky — the opposite of what a weighting scheme is for.

Two things drive rebalancing: drift from target weights, and changes in the underlying scores. They are different triggers and deserve different treatment.

Rebalancing on a fixed calendar is the worst option: it trades when nothing has changed and waits when something has. Threshold-based rebalancing — act when a sleeve is more than some tolerance from its target — trades less and acts sooner where it matters.

Rebalancing also has a cost. Moving capital between sub-accounts and resizing positions means turnover, spread and fees. A tolerance band exists precisely so the account does not pay that cost for cosmetic precision.

• Drift from target weight is a rebalancing trigger; the calendar is not.
• Score changes are handled by re-weighting, not by waiting for drift.
• Every rebalance costs spread, fees and possibly funding.
• A tolerance band is a cost control, not sloppiness.

Practical guidance by account size

What follows is reasoning about constraints, not advice about what you should do with your capital. The right numbers depend on which markets your leaders trade — a basket of BTC and ETH traders can operate at smaller sleeve sizes than one trading small-cap perps with wider ticks and thinner books.

At small size, the binding constraint is rounding. One sleeve, properly funded, will track its leader far better than three that cannot express the leader's smaller positions. Accept the concentration risk consciously, and size leverage down to compensate for it.

At medium size, several sleeves become viable, and the useful move is breadth rather than count: prefer leaders whose markets and holding periods differ, because two correlated sleeves cost twice as much as one and diversify almost nothing.

At larger size, the constraint inverts. Minimum sizes stop mattering and market impact begins to: a sleeve large enough to move the book on entry pays slippage that grows with the square of nothing convenient, and the leaders worth copying are the ones whose own size proves they can operate at that scale.

Common allocation mistakes

Most allocation errors fall into a small set, and all of them are avoidable by checking the constraint before the intention.

• Adding leaders the account cannot fund, so every sleeve rounds badly.
• Equal-weighting a set you have already ranked as unequal.
• Weighting without a cap, reproducing single-trader concentration.
• Treating correlated leaders as independent bets.
• Rebalancing on a calendar rather than on drift.
• Sizing to maximum leverage because the leader does.
• Reallocating to whichever sleeve performed best last month.

Limitations

Allocation cannot manufacture edge. If none of your leaders has one, weighting them cleverly changes only the shape of the loss.

Scores are estimates built from public history, and history is a small sample of a leader's possible behaviour. Weighting by score is weighting by the best available evidence, not by knowledge.

Past performance is not indicative of future results. Perpetual futures are leveraged instruments and carry a substantial risk of loss, including the loss of your entire position.

Conclusion

Allocate against evidence, cap the concentration that evidence would otherwise produce, fund each sleeve above the size at which rounding destroys it, and rebalance on drift rather than on the calendar. Those four rules do more for a multi-leader basket than any refinement of the shortlist.

How the live basket applies them — current weights, sleeve count and the volume threshold that governs it — is visible on the performance page and described in the mode note.