Hyperliquid Altcoin Surge Exhaustion: Delayed Short Basket

Hypothesis

Test whether unusually large, high-participation altcoin advances reverse over the following three days after volume contracts and price retreats. Trade independent short positions on INJ, NEAR, OP and SEI perpetuals using 4-HOUR bars. HyperliquidCollector supplies the OHLCV data. Reading actual parquet timestamps confirmed approximately 5,881–5,901 bars per instrument, with common coverage from 2024-01-25 through 2026-10-02 and one or two gaps per series; reset indicators after gaps. A development-only trigger-frequency check on the first 80% of each series found 49 candidate events combined, approximately 23 annually, without measuring subsequent returns. Inspiration: [Karbalaii's price-and-volume anomaly detection research](https://arxiv.org/abs/2503.08692). Its Poloniex detection results motivate the trigger, but do not establish profitability on these perpetuals.

Hyperliquid represents 7.4% of experiments against a 20% target, while long-only experiments account for 83.3% against a ceiling of 55%. A multi-instrument short basket addresses both imbalances without another BTC experiment. Corpus concentration was empty, so exact symbol-level novelty cannot be confirmed. This differs from the listed low-participation pair reversals and failed range-break trades: entries require a large high-volume advance followed by observed participation decay. No funding, OI, calendar or options signal is required.

verification_loop: sandbox_timeout: Verification failed (Layer 3 — sandbox backtest): smoke test exceeded the 300s wall-clock limit on a ~20000-bar window (2400 days of INJUSD.HYPERLIQUID-4-HOUR-LAST-EXTERNAL). This almost always means per-bar work that scales with history — e.g. rescanning the full funding/supplementary series, or rebuilding a list and calling min()/sorted() inside calculate_signal()/on_bar() on every bar. Precompute sorted timestamp arrays ONCE in __init__ and use bisect, or cache lookups keyed by timestamp, so per-bar cost is O(log n) not O(n).

Implementation

Trades independent shorts on INJ, NEAR, OP and SEI Hyperliquid perpetuals using four-hour bars. A high-volume six-bar advance is followed by two confirmation bars requiring a price retreat and volume contraction. Positions use entry-anchored ATR stops and targets evaluated at bar close, a maximum three-day hold, cooldowns, and portfolio exposure limits. Indicators reset after gaps.

Addresses the reported Layer 1.5 failure by explicitly supplying parameters.leverage as the float 1.0, alongside config.leverage=1.0. The consistency validator checks fixed research values inside parameters, including their types; the top-level leverage field alone does not satisfy it. Preserves the previous trading implementation and all fixed research values, with only surge_return_pct tunable. No verification or final-exam data reads were performed. The requested outbox file could not be written because this session permits only filesystem reads; the complete bound JSON is returned directly.

Verification Results

sandbox_timeout: Verification failed (Layer 3 — sandbox backtest): smoke test exceeded the 300s wall-clock limit on a ~20000-bar window (2400 days of INJUSD.HYPERLIQUID-4-HOUR-LAST-EXTERNAL). This almost always means per-bar work that scales with history — e.g. rescanning the full funding/supplementary series, or rebuilding a list and calling min()/sorted() inside calculate_signal()/on_bar() on every bar. Precompute sorted timestamp arrays ONCE in __init__ and use bisect, or cache lookups keyed by timestamp, so per-bar cost is O(log n) not O(n).

Iteration History

Verification failed (Layer 1.5 — hypothesis/config consistency) [class=hypothesis_mismatch]: - optimization_plan fixes leverage at 1.0

Outcome Summary

Multi-leg strategies with per-instrument state must keep per-bar work O(1) or O(log n), and should be profiled against the full sandbox bar budget before submission, so verification timeouts don't use up the iteration budget before the edge is ever tested.

It was abandoned in the verification loop after 3 iterations because the Layer 3 sandbox backtest went over the 300s wall-clock limit on a ~20,000-bar window (2400 days of INJUSD.HYPERLIQUID 4-hour bars); backtest review, optimization, analyst and risk stages were never reached.

Short INJ, NEAR, OP and SEI Hyperliquid perpetuals independently on 4-hour bars after an unusually large, high-volume advance is followed by volume contraction and a price pullback, betting on a reversal over the following three days.

No performance metrics exist: the backtest report is empty and no optimization was run, so return, Sharpe and trade count were never measured; the only pre-pipeline figure is a development-only trigger count of 49 candidate events (~23 per year) on the first 80% of the data.