BTC Cross-Venue Realized-Funding-Spread Mean-Reversion Pairs (Binance USD-M Long/Short Hedged by Hyperliquid Perp, Daily Decision)

A market-neutral, delta-hedged, multi-instrument cross-venue pairs strategy that EXPLICITLY USES BINANCE + HYPERLIQUID REALIZED FUNDING RATES (NOT predicted, NOT single-venue) as the divergence signal. THIS IS A FRESH HYPOTHESIS DIRECTLY ADDRESSING THE ANALYST'S EXPLICIT FEEDBACK on the prior BtcFundingCarryLongShort: 'generate a NEW hypothesis that specifies hl_funding_rates (REALIZED, not predicted) as the supp data dependency, and have the developer build the cross-venue spread signal from the start.' The mechanism: when 24h-summed realized funding rates diverge significantly between Binance USD-M (which pays every 8h, 3 payments/day) and Hyperliquid (which pays every 1h, 24 payments/day), the venue with HIGHER funding is paying its long side more — meaning longs there are crowded and getting taxed. Strategy goes SHORT on the higher-funding venue + LONG on the lower-funding venue, capturing both: (a) the funding payment differential (taxing the crowded longs pays our short side, while we pay less to be long on the cheaper venue), AND (b) inevitable mean-reversion of the funding-spread back toward zero as cross-venue arbitrage forces convergence. Position is delta-neutral in $ notional — direction-of-BTC doesn't matter; we profit from the FUNDING SPREAD moving toward zero. ADDRESSES MULTIPLE QUOTA GAPS: (1) cross-venue (currently 4.0% vs ≥15% target, gap 11%); (2) HYPERLIQUID (6.2% vs ≥20%, gap 13.8%); (3) long_short direction (11.7% vs ≥45% implicit, gap ~33%); (4) pairs/multi scope. AVOIDS EVERY KNOWN FAILURE PATTERN: (a) NOT ADA 4H (different instruments, different venues); (b) NOT microstructure / tick-data (uses funding rate supp data, not order book); (c) NOT a regime-fragile breakout (works in any direction because delta-neutral); (d) DAILY decision cadence = ~2300 daily bars over 6.3 yr → smaller Optuna search space addressing 7-for-7 4H optimization collapse; (e) ONLY 3 PARAMETERS total (entry_spread_threshold, exit_spread_threshold, max_hold_days) — radically simple. EXPLICIT DATA REQUIREMENTS (analyst's prescribed spec): BTCUSDT.BINANCE 1D bars (primary, most reliable bar data), BTCUSD.HYPERLIQUID 1D bars (hedge leg), funding_rates (Binance USD-M realized), hl_funding_rates (Hyperliquid REALIZED, NOT predicted — REQUIRED supp data).

Directly implements the analyst-prescribed hypothesis: the decision variable is the cross-venue REALIZED funding spread built from funding_rates (Binance) + hl_funding_rates (Hyperliquid REALIZED, NOT hl_predicted_fundings) from the start. Daily cadence (~2300 daily bars over 6.3yr) keeps the Optuna search space small and addresses the 4H optimization-collapse pattern; only 3 edge parameters (entry_spread_threshold, exit_spread_threshold, max_hold_days). 24h-window summation correctly normalizes Binance's 3 payments/day against Hyperliquid's 24 payments/day so the spread compares like-for-like realized cash flows. Both legs are MARGIN/futures venues (BINANCE + HYPERLIQUID), enabling a genuine long+short delta-neutral book — addressing the cross-venue, Hyperliquid, and long_short quota gaps. The same 1-DAY timeframe on both legs engages the base class cross-leg alignment barrier so the spread and fills always use contemporaneous prices. Equal dollar notional per leg makes BTC price direction irrelevant; profit comes purely from funding capture plus spread convergence. Coverage gating (require >=2 Binance and >=12 HL funding points in the trailing window) prevents partial-window phantom spreads and forces a flat book rather than a degenerate price proxy when funding data is missing. Leverage left at 1.0 (0.30 per-leg = 0.60 gross) and not referenced in sizing to avoid the leverage_set_but_unused gate.

No edge — the cross-venue realized funding spread on BTC is structurally too small to harvest after costs. Code and data are correct (hl_funding_rates loaded, 50 pairs traded, funding credited +503, hedge near-neutral, trades match the hypothesis), so this is neither the zero-trade HL-key bug nor a funding-crediting issue. The problem is economic: over 50 pairs the captured Binance-vs-Hyperliquid funding differential was only +503 (~$10/pair) because cross-venue arbitrage keeps the two venues' BTC funding tightly converged, while the 'delta-neutral' book lost -2,978 on basis drift between the BTCUSDT-perp and BTCUSD-perp contracts and paid -2,684 commission — a ~10:1 cost-to-edge ratio, net -2,476. Every aggregate is consistently negative (PF 0.93, omega 0.23, expectancy -24.76, 14 of 15 months red). The three parameters (entry/exit spread threshold, max_hold) can only select which spreads to trade; they cannot enlarge a structurally tiny funding differential or remove the basis-drift + commission hurdle, so no parameter region clears breakeven. Compounding this, Hyperliquid realized-funding history only starts 2025-03, so all trades sit in one 14-month regime — too thin for reliable optimization. FAILURE PATTERN: cross-venue (Binance/Hyperliquid) BTC realized-funding-SPREAD mean-reversion has no harvestable edge — the spread is arbitraged near-flat (~$10/pair) and is dwarfed ~10:1 by cross-contract basis drift and fees. NOTE for Research Lead: this directly tested the prescribed realized-cross-venue-funding spec and the spread itself is too tight on a major like BTC; a funding-DIFFERENTIAL trade would need a venue pair / instrument with persistently DIVERGENT funding (less-arbitraged alt), not BTC, but the basis-drift cost of hedging two different perp contracts remains a structural drag on any such pair.