AVAX Hyperliquid 4H Two-Endpoint String-Shape Analog Forecast

Hypothesis

Derived by the Research Lead from Erik Bartoš and Richard Pinčák's Identification of market trends with string and D2-brane maps: test whether the two-endpoint price-history map identifies recurring shapes with informative subsequent returns. This is an operationalization of the supplied full paper, not a reproduction of StringAlgo v.16, whose complete order rules are not disclosed. Use AVAXUSD.HYPERLIQUID 4-HOUR closes as a declared substitute for the paper's FX midpoint samples. Keep the two-endpoint mapping from Equation (3); add a standard nearest-neighbor forecast, explicit trading thresholds, and protective exits. All map windows end at the decision timestamp; historical neighbor outcomes must already be fully observed. Execute market orders at the signal bar's close. Protective stops trigger intrabar through platform matching. Parameters remain frozen after deployment, while the causal historical observation buffer advances.

Long-short exposure addresses the portfolio's largest directional imbalance, and Hyperliquid is underrepresented at 10.2% against its 20% target. AVAX avoids another BTC or ETH experiment; corpus_concentration is empty, so its recent instrument share cannot be verified from this brief. Four-hour sampling accommodates Hyperliquid's approximately 5000-candle history limit. The mechanism tests price-path analog forecasting rather than another indicator breakout or fast mean-reversion variant. A single map-based hypothesis is proposed because alternative map dimensions alone would be variants of the same forecasting mechanism.

negative_expectancy: decline optimization of this AVAX Hyperliquid 4H configuration because its measured net per-trade return is negative and below the supplied venue review floor across 606 closed positions. Impact materially exceeds the measured net loss, so this is not evidence that the signal is necessarily negative before execution costs; nevertheless, the current implementation is uneconomic at the tested allocation. This verdict judges the Research Lead's two-endpoint nearest-neighbor operationalization, not the paper authors' undisclosed StringAlgo rules or the entire string-map family.

Implementation

AVAX Hyperliquid 4H long-short strategy using normalized two-endpoint string maps and causal nearest-neighbor return forecasts. Entries exceed ±1.5% forecast return; exits occur on forecast reversal, after nine bars, or through an intrabar entry-anchored ATR stop. Sizing risks at most 0.5% of settled equity at the stop, capped at 30% equity notional.

Addresses the reported Layer 1.5 failure by explicitly declaring parameters.bar_interval as the fixed string 4-HOUR, matching the primary bar type and unchanged research plan. All fixed research values and the complete optimization plan are included. The only source addition records that sampling parameter; existing forecast, entry, sizing and protective-stop logic are preserved. No independent verification was run. The required outbox file could not be written because this session permits only read access; the complete task-bound JSON is returned here.

Verification Results

The sandbox reports -0.171% net return per trade across 606 closed positions. Modeled impact is $6,161 and estimated capacity is $8,431; the full backtest should assess economics at the intended capital allocation. These results do not establish a correctness defect.

Backtest Review

Sharpe

-0.47

Total return

-4.48%

Max drawdown

10.73%

Trades

606

Win rate

49.2%

Profit factor

0.93

606 closed positions provide a functioning sample; inspected trades show both directions, variable capital-relative sizing and short holding periods consistent with the implementation.

Accounting reconciles, no positions remain open, and drawdown is 10.73%.

Net average trade return is -0.1744% versus the supplied +0.15% HYPERLIQUID review floor; PF is 0.926, Sharpe -0.473 and total return -4.48%.

QA's economic concern is borne out: modeled impact is $6,161, with estimated capacity of $7,439 against the $100,000 engine account. This does not establish a correctness defect.

All reported calendar periods lose money. Forecast IC 0.0534 with t=0.98 is inconclusive, rather than proof of absent predictive information.

No parameter sweep has occurred, so these results do not establish that every parameter configuration fails.

The verdict could not be saved to the requested outbox because this session permits only filesystem reads.

Iteration History

Track the protective stop's client order ID and reconcile its quantity to the cumulative open position after every entry fill. Define a consistent entry-price anchor and update the stop trigger accordingly. Verify with an entry that fills in multiple events, followed by a stop trigger, that the entire position closes.

Protective stop quantity is not updated after partial entry fills. The first entry fill creates a stop for the quantity then held. Subsequent fills increase the position, but _ensure_protective_stop returns immediately whenever an existing reduce-only order is found. NautilusTrader supports multiple fills for a market order, including L1 residual fills. Consequently, the stop can close only the initial partial quantity, leaving the remainder exposed until a later bar recreates protection. Subsequent fills also overwrite protective_level without updating the existing stop's trigger.

Verification failed (Layer 1.5 — hypothesis/config consistency) [class=hypothesis_mismatch]: - optimization_plan fixes bar_interval at 4-HOUR

Outcome Summary

Future iterations should assess execution-cost viability at a suitable allocation; this configuration's failure does not establish that all string-map configurations fail.

Backtest review recommended abandonment because net per-trade return was negative and below the supplied +0.15% venue floor at the tested allocation. Optimization, post-optimization analysis, and risk review were not reached.

It used causal two-endpoint string maps and nearest-neighbor forecasts to trade recurring AVAX Hyperliquid 4-hour price shapes with protective stops.

The backtest produced 606 closed positions, a -4.48% total return, Sharpe of -0.473, and maximum drawdown of 10.73%. Average net trade return was -0.1744%, with a profit factor of 0.926.

Analysis

Identification of market trends with string and D2-brane maps