How a score-weighted basket behaves in different market regimes

Why regime matters more than selection

Discussions of copy trading focus almost entirely on selection: which traders, scored how, weighted by what. That is the interesting part of the design, and it is the part most within our control. It is not the part that determines most of the variance in outcomes.

Perpetual futures strategies are regime-dependent. A momentum approach and a mean-reversion approach are not two flavours of skill; they are two bets on which regime persists. A trader can execute flawlessly and lose money for a quarter because the conditions their approach monetises were absent. Selection determines who you are exposed to. Regime determines whether that exposure pays.

This note walks through the regimes one at a time and states, as plainly as possible, how the mechanism behaves in each — including the two where it behaves worst.

Sustained trend: dispersion narrows, weights concentrate

In a persistent directional move, trend-following and momentum leaders produce their strongest closed-trade records. Composite scores rise for that group, and because allocation is proportional to score, capital shifts toward them at each recomputation.

This is the regime the mechanism handles best, and it is also the regime in which the mechanism is quietly doing something you should be aware of: it is concentrating. The basket becomes progressively less diversified in strategy terms even while it still holds ten addresses, because most of those addresses are now expressing the same bet in slightly different ways.

Weight caps bound how far this can go, but they cannot make the underlying strategies less correlated. A basket that looks diversified by leader count can be highly concentrated by exposure, and a trend is exactly when that happens.

Choppy range: mean-reversion leads, momentum bleeds

In a range-bound market with frequent false breakouts, the ordering inverts. Mean-reversion and market-making style leaders produce steady closed profits, while momentum leaders take a series of small losses as each apparent breakout retraces.

Score-weighting responds to this the way it is designed to: momentum scores decay, weights fall, mean-reversion weights rise. The transition takes time — a score needs closed trades to move — so the basket carries the losing configuration for part of the range before adjusting.

The compensating property is that a chop is generally a low-magnitude regime. The bleed is real and it is slow, which is the kind of loss a rule-based system is well suited to arrest.

Volatility expansion: liquidation risk per sleeve rises

When realised volatility expands sharply, the dominant risk stops being drawdown and becomes liquidation. Positions sized for a normal distribution of moves encounter an abnormal one, margin is consumed faster than it can be topped up, and leveraged positions are closed by the protocol rather than by the trader.

Isolation matters most here and is also most clearly not protection. A liquidation in one sub-account consumes that sleeve's margin and cannot reach the other nine — that is the containment working exactly as intended. The capital in that sleeve is still gone.

Volatility expansion is also when emergency removal is most likely to fire, and when fills during an unwind are worst. The system's risk controls and the market's worst execution conditions coincide, unavoidably, because the same event causes both.

Correlated deleveraging: the worst case

The genuinely bad regime is a cascade: a large move triggers liquidations, which force selling, which extends the move, which triggers more liquidations. In these episodes, cross-leader correlation goes to something close to one. Momentum, mean-reversion and carry leaders all lose at the same time, because they are all long risk in a market where risk is being forcibly unwound.

Diversification across ten leaders does very little here, and it is important to say so rather than to describe the basket as if it were hedged. It is not hedged. It is a set of directional perpetual futures positions, and in a correlated deleveraging event a set of directional positions loses together.

What isolation still provides in this regime is bounded contagion: a liquidation in one sleeve cannot pull margin from another and cannot cascade internally. That converts a possible total-account failure into several bounded sleeve failures. It is a meaningful difference and it is not the same as being protected.

Low-volatility drift: cost dominates

The quiet regime is the one people forget to plan for. When ranges compress and moves are small, trading PnL across every strategy family gets thin, and the cost terms that were rounding errors in an active month become the dominant line.

Exchange fees and the builder fee are charged on volume regardless of whether the volume produced anything. Funding accrues on held positions. Slippage is paid on every fill. In a month where the basket's gross trading result is near flat, cost decides the sign of the outcome.

This is the regime where low-turnover leaders quietly outperform high-turnover ones with identical skill, purely because they generate less volume to be charged on.

Regime transitions: the structural weak point

Everything above describes behaviour inside a regime. The transition between them is where a score-weighted approach performs worst, and this is a property of the design rather than a bug in it.

Composite scores are computed from closed trades. Closed trades are evidence about the regime that just ended. At the moment a regime turns, the basket is maximally weighted toward the leaders who were best suited to the regime that is now over, and it will take a series of closed losses before the weights move. You are, by construction, most concentrated in the wrong strategy at the exact moment the market changes.

There is no version of a backward-looking scoring rule that avoids this. Shortening the scoring window makes the system react faster to the turn and also makes it react to noise, which produces expensive churn in the far more common case where nothing structural has changed. Lengthening it does the reverse. We have chosen a window that tolerates the transition cost rather than one that chases turns, on the view that regime turns are rare and noise is constant.

• At a turn, weights are concentrated in the strategy that just stopped working.
• Score adjustment requires closed trades, so it lags the turn by design.
• Faster scoring reduces the lag and increases churn cost in every non-turn period.

What score-weighting does mechanically across regimes

Stripped of narrative, the rule is simple: tilt toward what recently worked, bounded by a cap, revised as evidence arrives. That produces three consistent behaviours regardless of which regime you are in.

It compounds persistence. When a regime lasts, weighting toward its winners increases exposure to a bet that keeps paying. It penalises inconsistency. A leader with erratic results is weighted below a steadier one with the same average, because the scoring inputs include consistency and survivability, not just returns. And it lags reversals, which is the price of the first two.

Equal weighting inverts these properties: it lags less at turns and captures less inside regimes. The note on score-weighting versus equal-weighting works through that trade-off in detail. Neither is right in all conditions, which is why the choice should be stated rather than assumed.

What no weighting scheme can fix

Three limits apply to every allocation rule, ours included, and no amount of scoring sophistication touches them.

• Direction. A basket of directional perpetual futures positions loses when the market moves against it. Weighting changes how much, never whether.
• Correlation in stress. Weights computed in calm conditions describe relationships that do not hold in a cascade.
• Venue concentration. Every sleeve is on Hyperliquid, so a protocol, oracle or liquidity failure there is not diversified by any weighting.

Conclusion: a basket changes the distribution, not the direction

A score-weighted basket of isolated sleeves is a defensible way to allocate across Hyperliquid leaders. It compounds inside stable regimes, it degrades gracefully in chop, it contains liquidations to a single sleeve when volatility expands, and it is honest about the two regimes where it struggles: correlated deleveraging, where diversification largely stops working, and regime transitions, where backward-looking scoring is maximally wrong.

What it changes is the distribution of outcomes — fewer total failures, less dependence on one person's judgement, bounded contagion between strategies. What it does not change is the direction of the market or the fact that leveraged perpetual futures can lose you your entire position.

The current basket, its weights and each leader's recent history are published live, which is worth looking at alongside whatever regime you think we are in now.