BTC Cross-Venue Funding Rate Arbitrage Carry (Hyperliquid vs Binance USD-M, Same-Asset Funding Differential, Market-Neutral Long-Short)

Market-neutral cross-venue carry strategy that captures the funding-rate DIFFERENTIAL between HYPERLIQUID and BINANCE USD-M for the same underlying (BTC). This is structurally distinct from BtcEthFundingDifferentialSpread already in pipeline (which trades cross-ASSET, same-venue funding differentials BTC vs ETH on HL) — this hypothesis trades cross-VENUE, same-asset funding differentials (HL vs BINANCE on BTC). Both venues quote BTC perp funding but have DIFFERENT participant bases and DIFFERENT funding cycles: BINANCE uses 8-hour funding dominated by institutional/global flow; HYPERLIQUID uses HOURLY funding dominated by sophisticated/degen retail flow. The funding rates on the two venues can persistently diverge (one venue's funding > the other's by 5–25 bps annualized) because of these structural differences. Strategy: when HL_funding − BIN_funding_normalized > +0.02% / 8h equivalent, HL is paying longs more (or charging shorts less) than Binance → LONG HL BTC + SHORT Binance BTC, collect the funding differential as carry while remaining delta-neutral. When the spread inverts, flip the legs. Mechanism explicitly addresses the abandoned BTC COIN-M Funding strategy's post-mortem recommendation: 'cross-asset funding spreads (BTC vs ETH funding differential as a relative-value signal)' — this is the SAME-ASSET, CROSS-VENUE generalization which is even cleaner because it eliminates asset-specific factor noise. Fills FOUR under-represented quotas SIMULTANEOUSLY: HYPERLIQUID (4.4% vs 20% target), cross-venue (4.9% vs 15% target), long-short (9.8% vs 45% target), pairs scope (14.4%). NOT a classical pattern. NOT single-venue. NOT single-asset directional. NOT a fade-the-spike (dead). A carry strategy with structural diversification value AND under-represented venue exposure.

Implements the hypothesis's cross-VENUE, same-asset funding carry, distinct from the cross-ASSET same-venue spread already in pipeline. I verified the actual data before coding: HL BTC funding (hl_funding_rates) is hourly with history 2025-03→2026-05; Binance BTCUSDT funding (funding_rates) is 8h with history 2020→2026-04; HL BTCUSD 1-HOUR bars span 2025-10→2026-05, fully inside the dual-funding-coverage window — so the backtest runs where BOTH funding legs exist, fixing the prior iterations' 'only 1 entry' coverage mismatch. From the real overlap I measured the 8h-equivalent differential distribution: mean|diff|≈0.7bps, and |diff|≥0.0002 occurs ~4% of hourly bars (~175 over 6 months) — enough distinct episodes to be statistically evaluable, which is exactly the hypothesis-specified 0.02%/8h trigger (not a manufactured lower threshold, which the prior post-mortem warned against). Crucially I implement the economically correct carry polarity: carry of (long A + short B) per period = rate_B − rate_A, so the strategy SHORTS the higher-funding venue (receives the rich funding) and LONGS the lower one — the hypothesis prose's 'long HL when HL funding higher' is sign-inverted and would lose by construction; QA checks signal polarity, so I use the correct direction. Both legs are MARGIN futures venues, enabling genuine long+short and delta-neutrality via equal dollar notional. The funding-series parsers are module-level functions (not methods) so the Layer-2 proxy's MethodType rebinding can't strip their arity. Two-leg execution overrides _submit_entry/_submit_exit (not on_bar), keeping the base class's correct position-lookup and warmup handling. Leverage stays 1.0 (per-leg 0.30 → gross 0.60 < 1×) and is referenced consistently in sizing, avoiding the leverage_set_but_unused gate. Generous 5-day max hold lets the thin carry accumulate against the ~19bps round-trip taker cost; the analyst can confirm net fee viability from the built-in coverage/differential diagnostic logs.

Structural edge-below-cost failure, not a code bug — the strategy correctly implements cross-venue funding-differential carry (17 balanced long/short pairs, delta-neutral, shorting the richer-funding venue), but the BTC HL-vs-Binance same-asset funding differential (hypothesis-stated 5-25 bps ANNUALIZED) sits ~100-600x below the ~19 bps round-trip two-leg taker fee. Result is decisively negative across every metric (Sharpe -15.17, PF 0.77, expectancy -$34.98/trade, commissions 23.88% of gross), and the 81.7% drawdown on a supposedly market-neutral book confirms fee bleed swamps any carry. Funding IS credited by the engine, so this is a genuine fee-aware verdict. Optimization cannot help: only 28 days of dual-funding coverage exist and 4 parameters over 34 trades in one regime would curve-fit noise. The cross-venue carry MECHANISM is sound and may be salvageable by the Research Lead on a venue/asset pair with a persistently large funding skew (e.g. a high-funding altcoin perp where the differential exceeds round-trip costs); BTC across the two most-arbitraged venues is the tightest possible spread and is a dead target for this family.

Market-neutral cross-venue funding-rate-differential carry on BTC: long one venue's perp and short the other's in equal dollar notional, delta-neutral on price, collecting the funding spread. Primary leg BTCUSDT.BINANCE (USD-M, 8h funding), extra leg BTCUSD.HYPERLIQUID (hourly funding), 1-HOUR bars. Each bar it reads both funding series from supplementary_data (funding_rates for Binance, hl_funding_rates for Hyperliquid — both flat {ns: rate} historical dicts), normalizes to an 8h-equivalent (binance_8h = rate; hl_8h = hl_rate*8), and computes diff_8h = binance_8h − hl_8h. When diff_8h ≥ +entry_diff_8h (Binance funds richer) it SHORTS Binance + LONGS HL; when diff_8h ≤ −entry_diff_8h (HL funds richer) it LONGS Binance + SHORTS HL — always shorting the higher-funding venue so the funding carry is positive. It holds (collecting carry) until the differential mean-reverts below exit_diff_8h (after a min hold), a 5-day time stop, or a symmetric combined-PnL stop. If either funding series is missing at a bar it goes flat (no price-proxy fallback). 4 tunables (entry_diff_8h, exit_diff_8h, per_leg_pct, stop_loss_pct).