Hyperliquid NFT-Sector Lead–Lag Catch-Up in SAND

Hypothesis

Test whether SAND adjusts with a delay after concordant, substantial 24-hour moves in GALA and APE. Trade SAND long or short using synchronized 4-HOUR bars; GALA and APE are signal sources only. HyperliquidCollector supplies perpetual OHLCV for all three exact instruments. Catalog inspection verified 4,975 common timestamps from 2024-06-21T04:00:00Z through 2026-10-01T20:00:00Z, with close-stamping markers present. A baseline trigger-frequency check found 93 non-overlapping signals, approximately 41/year; it did not inspect subsequent returns. Implement as a FactoryStrategy subclass with bounded rolling buffers and timestamp-matched extra-bar handling. Statistical inspiration: https://quant.stackexchange.com/questions/81139/lead-lag-statistical-test-in-python ; this source suggests testing lead–lag relationships and supplies no evidence of profitability for these tokens.

Hyperliquid is substantially below its 20% target at 6.9%, and long-only experiments exceed their 55% ceiling at 83.9%. This proposal adds bidirectional exposure and multiple signal instruments without using BTC, options, calendar triggers, funding cash flows, or same-asset cross-exchange convergence. Its falsifiable premise is delayed transmission of a shared NFT-sector shock, rather than a supplementary-data gate on single-perp momentum. Instrument-level corpus concentration was not supplied, so these tokens' historical research shares cannot be claimed. The 12-bar holding ceiling and substantial lag requirement distinguish this from sub-hour fee-sensitive reversion. Similarity to existing relative-strength families remains a novelty-review risk.

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

Implementation

Trades SAND long or short after substantial concordant 24-hour GALA and APE moves that SAND has insufficiently captured. Uses exact timestamp alignment, an entry-anchored two-ATR stop, a target capturing half the initial lag gap, a 12-bar holding ceiling, and a 12-bar cooldown. Only SAND is traded.

Preserves the previous implementation and every registered parameter permission. The supplied previous-code diagnostic reports zero evaluated bars and a 401-bar GALA gap preceding SAND history; runner inspection confirms extra-leg audits occur before engine execution. Bounded synchronized storage and incremental ATR already keep strategy work independent of history length. No strategy-only change can repair that pre-execution failure without changing scope or bypassing data integrity. The documentation now explicitly requests engineering review of the joint data window, preserving all three instruments, registered history, warmup, and the exclusive 2026-04-18 final-exam boundary. The timeout remains unresolved; no verification was run. The required outbox file could not be written because this session permits filesystem reads only.

Verification Results

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

Iteration History

sandbox_timeout: Verification failed (Layer 3 — sandbox backtest): smoke test exceeded the 300s wall-clock limit on a ~20000-bar window (2400 days of SANDUSD.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).

Outcome Summary

Before iterating, settle the infrastructure and config problems (sandbox timeouts while refilling extra-leg data, and an optimization plan that conflicts with the hypothesis) rather than resubmitting signal logic that has not changed.

After 3 iterations the strategy was abandoned in the verification loop. It failed Layer 1.5 (hypothesis/config consistency, class hypothesis_mismatch) because the optimization_plan fixes leverage at 1.0; an earlier iteration had also hit a sandbox timeout before any bars were evaluated.

Trade SAND long or short on Hyperliquid 4-hour bars, betting that it catches up after GALA and APE both make substantial moves in the same direction over 24 hours while SAND lags.

No performance figures exist: there is no backtest or optimization report. The only evidence is the hypothesis's pre-check, which found 93 non-overlapping signals (about 41 a year) across 4,975 common timestamps and did not look at what happened afterwards.