Hyperliquid ATOM–ALGO–ETC Supertrend with a Causal Shadow-Equity Gate

Hypothesis

Derived by the Research Lead from ChaoZhang's article, https://www.fmz.com/strategy/438771: operationalize its equity-curve gating concept on three independent Hyperliquid perpetuals using 4-HOUR bars. The supplied Pine code trades every Supertrend flip and only plots filtered performance; this adaptation explicitly implements a causal gate rather than claiming to reproduce its orders. Each instrument maintains an ungated virtual reference strategy while actual trading may pause. Market orders execute at signal-bar closes, the registry's equivalent of next-bar-open execution on continuous markets. Use available 4H history without assuming more than approximately two years. Tier-0 taker fees are 0.045% per fill, verified in [Hyperliquid's official fee schedule](https://hyperliquid.gitbook.io/hyperliquid-docs/trading/fees).

Addresses long-only concentration of 76.9%, Hyperliquid's underweight 12.3% allocation and excessive single-symbol research. Options have the largest venue deficit, but this mechanism does not justify an options wrapper. corpus_concentration is empty, so recent instrument shares cannot be verified. Snapshot searches found generic Supertrend strategies and an XLM–ETC daily Turtle shadow-loss filter, but no indexed instance of source 438771 or this equity-EMA contract. The distinguishing premise is persistence of reference-strategy profitability, without Turtle channels, pyramiding or previous-trade-loss selection.

Reject this Research Lead operationalization on ATOM–ALGO–ETC, not the source author's strategy or equity gating generally. The registered search cannot reliably produce candidates within the fixed 15% drawdown limit, and only 1 of 3 exposed calendar years is profitable versus the enforced 60% requirement. The sole selected candidate loses out of sample. Average trade return 0.3452% clears the supplied 0.18% economics floor, so this is not a fee-floor rejection. No demonstrated implementation defect or validated risk-admissible parameter region supports another developer iteration. Inconclusive holdout significance is not the abandonment reason. No accounting or data defect is asserted.

Implementation

Independent ATOM, ALGO and ETC Hyperliquid 4H Supertrend flip strategies with causal completed-shadow-trade equity EMA entry gates. Market orders at signal-bar closes implement the registered next-open equivalent for continuous markets.

Minimal Layer-4 implementation fix: sizing no longer requires processing time to equal the decision bar close. The observation timestamp guard rejects a leg state newer than the decision bar while allowing delayed processing of valid flips and reversals. Signal, shadow accounting, sizing caps and configuration are preserved. Smoke drawdown, impact and absent funding remain advisory concerns for full-history evaluation; no additional trading rules were introduced. Verification remains with the pipeline.

Verification Results

The smoke report shows 37.5151% drawdown versus the registered 15% constraint, impact consuming 91.0036% of gross profit, and no funding events. Evaluate these on full history and reconcile funding coverage before trusting perpetual net economics; they do not establish an implementation defect.

Backtest Review

Sharpe

0.14

Total return

1.08%

Max drawdown

38.09%

Trades

210

Win rate

32.9%

Profit factor

1.01

210 closed trades over 746 days provide a functioning baseline for the registered search. Both directions trade on every leg; the code causally gates entries while retaining opposite-flip exits.

Average net trade return 0.3452% exceeds the 0.15% Hyperliquid review reference. Closed-trade PnL reconciles to $1,078.62 and +1.0786% account return, with no terminal inventory.

QA's missing-funding concern is not borne out on full history: all three legs receive funding cash flows, totaling approximately -$1,379.13.

Portfolio economics are marginal: PF 1.0083 and Sharpe 0.1373. ATOM earns $4,627.22 and ALGO $16,881.49, offset by ETC losing $20,430.09.

QA's risk and impact concerns are confirmed: drawdown is 38.0872% versus the registered 15% constraint, and impact consumes 95.4067% of gross profit. Optimization must assess whether the permitted parameters can address these weaknesses.

Funding coverage remains unverified despite credited payments. Buy-and-hold is explicitly labelled an unsuitable benchmark for this two-sided basket.

Analysis

210 closed trades reconcile to $1,078.62 net profit with no terminal unrealized gain.

Average net trade return 0.3452% clears the supplied 0.18% economics floor. Buy-and-hold is not a meaningful benchmark for this book.

Only 1 of 5 optimization windows produced a risk-admissible candidate, with none in the latest window. All 60 training candidates in each of the last three windows breached the registered 15% drawdown constraint.

The sole selected candidate deteriorated from training Sharpe 2.044 to outer OOS Sharpe -1.107 across 45 trades.

Final parameters remain the baseline: no improvement in Sharpe 0.137, return 1.079%, PF 1.008 or drawdown 38.087%. Only 1 of 3 exposed calendar years is profitable, failing the supplied enforced persistence floor.

Impact consumes 95.41% of gross PnL. ETC loses $20,430.09; one ALGO winner contributes $33,668.48, exceeding the entire portfolio net gain.

The 168-day holdout has 43 trades and Sharpe -0.289, but significance is inconclusive because registered net-MTM and lineage evidence are missing. It is not a statistically rejected holdout.

Full-history trade count and average trade return exceed predictions; return misses the predicted 8%, and drawdown breaches the declared risk limit. Return predictions are not eligibility floors.

Optimization demonstrably ran despite the supplied gate text saying otherwise. That contradictory gate entry is not used as a rejection reason; sensitivity cliffs are advisory.

inconclusive

hard

Iteration History

Before every entry/reversal, calculate the projected aggregate absolute notional across all three legs using causal current prices and pending orders. For a reversal, replace the old leg's exposure with the proposed target rather than counting the closing quantity as new exposure. Limit the target to both 15% of equity and the remaining 45%-gross budget, accounting for closure PnL/fees as appropriate. If the remaining target is below minimum notional, skip a flat entry or close the opposite position without reopening. Verify that every exposure-increasing fill respects the aggregate cap.

_leg_size() implements max_gross_fraction=0.45 only by limiting each newly sized leg to 0.15 of settled equity. It never measures the current gross notional of the other open legs or reserves pending exposure. Existing quantities remain unchanged as prices and account equity change, so a new entry or reversal can exceed the contract's 45% gross limit at entry. For example, two existing legs worth 18% of current equity each plus a new 15% leg produce 51% gross. The supplied sandbox corroborates this with peak_entry_leverage=0.487273, above 0.45. The framework's venue-margin cap does not enforce this strategy-specific limit.

Remove the strict clock-equality requirement. Guard causal market observations using their own timestamps relative to the decision bar; if stale-bar protection is needed, use an explicit age tolerance consistent with the framework. Verify that a valid flip still produces positive size when processing time is slightly later than bar.ts_init.

_leg_size returns zero whenever self.clock.timestamp_ns() > ts_ns. ts_ns is the completed bar's close timestamp; the framework uses wall-clock time in paper/live, where processing necessarily occurs after that close. Consequently valid flat entries are suppressed on all three legs, and allowed reversals close the existing position without opening the required new target. The smoke backtest masks this defect because its simulated clock equals the bar timestamp. This is an entry-path defect, not restart-persistence behavior.

Outcome Summary

Passing the per-trade economics floor is insufficient when drawdown, persistence and out-of-sample performance remain inadequate.

The analyst abandoned it because optimization could not reliably satisfy the registered 15% drawdown limit, and only 1 of 3 exposed calendar years was profitable against the enforced 60% requirement. Average trade return cleared the economics floor; inconclusive holdout significance was not the abandonment reason.

Trade 4-hour Supertrend flips on Hyperliquid ATOM, ALGO and ETC perpetuals, using a causal shadow-equity EMA to gate each leg's entries.

The baseline returned 1.0786% ($1,078.62 net) across 210 closed trades over 746 days, with Sharpe 0.1373 and maximum drawdown 38.0872%. The sole selected optimization candidate deteriorated from training Sharpe 2.044 to out-of-sample Sharpe -1.107 across 45 trades.