Inter-Exchange Funding-Spread Carry, BTC Hyperliquid vs Binance USD-M (Delta-Neutral Two-Leg Perp, Long the More-Negative-Funding Venue / Short the More-Positive, Funding-Driven Cross-Venue Mispricing, Low-Parameter)

A MARKET-NEUTRAL, CROSS-VENUE funding-rate arbitrage on BTC perpetuals across two exchanges: BTCUSD.HYPERLIQUID (primary) and BTCUSDT.BINANCE (USD-M perp, hedge leg). Both legs are BTC perps, so equal-and-opposite notional makes the pair delta-neutral — directional BTC price risk cancels and the strategy harvests ONLY the funding-rate DIFFERENTIAL between the two venues. Hyperliquid (hourly funding, retail/degens-heavy flow) and Binance USD-M (8h funding, institution-heavy) persistently price funding differently because their participant bases and liquidity differ; when the spread is wide the carry is structurally positive. This is NOT a single-instrument directional perp trade (which hits the documented over-exposure engine bug) — it is a hedged two-leg carry, the same structural class as the factory's already-successful IntraBinanceSolSpotPerpFundingCarryDeltaNeutral1H, but EXTENDED CROSS-VENUE to fill the cross-venue (6% vs 15% target) and Hyperliquid (8.2% vs 20% target) gaps simultaneously. Deliberately low-parameter (one entry threshold, one exit threshold, one max-hold) to resist overfitting. Sizing is fixed equal-notional per leg at ~40% of equity per leg (delta-neutral net), never scaled by signal strength, never re-multiplied by leverage — explicitly bounded to avoid the over-exposure failure family.

Implements the hypothesis exactly as the same structural class as the validated IntraBinanceSolSpotPerpFundingCarryDeltaNeutral1H (hedged two-leg carry), extended cross-venue to fill the cross-venue and Hyperliquid quota gaps. Two key engineering lessons from the sibling iterations are already baked in: (1) VENUE WIRING — Binance is the PRIMARY leg, not Hyperliquid as the brief suggested, because the backtest's supplementary loader scopes the top-level funding_rates series to the primary venue's exchange filter; with Binance primary, funding_rates resolves to the Binance USD-M series and hl_funding_rates resolves venue-independently by coin, so BOTH legs' funding read correctly (delta-neutrality and the edge are identical regardless of which perp is labelled primary). (2) ATOMIC HEDGE — _submit_entry is overridden to validate that BOTH legs are market-ready and to precompute the equal-notional HL hedge size BEFORE placing either order, so the primary leg is never submitted naked; _submit_exit closes both legs in lockstep. The funding differential is annualized with venue-correct period counts (Binance 8h => 1095/yr, Hyperliquid hourly => 8760/yr) and nearest-timestamp lag tolerances matched to each venue's funding cadence (8h / hourly). There is deliberately NO price-only fallback edge: when either leg's funding is unavailable the signal is 0.0 (flat). Sizing is fixed equal-notional (~40% equity/leg), never scaled by signal or leverage, explicitly bounding exposure to avoid the single-instrument over-exposure failure family. Low-parameter by design — one entry threshold, one exit threshold, one max-hold — to resist overfitting. Routed to futures venues because the strategy holds short perps and the edge is funding-based; leverage stays 1.0 and is not used in sizing, so there is no leverage_set_but_unused mismatch.

No edge after two-venue costs, and not fixable by optimization or further code iteration. The strategy loses money in every month of the sample (Oct-2025 to Jun-2026, -0.55% to -3.72%): total_return -14.0%, Sharpe -4.65, Sortino -3.87, profit_factor 0.76, expectancy -$60/trade, PSR 0.0001. The iteration-2 atomic-hedge fix succeeded — the book is now genuinely delta-neutral (beta 0.0048, end_unrealized -0.44%, no artifact spike), which is precisely why the true carry is visible and it is structurally negative. Fees dominate: commission_pct_of_gross 17.81% ($15.5k commission + $6.9k impact on 450 trades), so the BTC HL-vs-Binance funding differential is smaller than the ~0.19% two-venue round-trip cost it pays each turnover. This is the honest, artifact-free version of the z-scored cross-venue BTC sibling reviewed earlier (same venue pair, same no-edge). NOT iterate: the leg-netting bug is already fixed (iteration 2) and the residual is a structural cost problem, not a code defect a developer can change; tuning entry_apr/exit_apr just trades less of the same negative-after-cost spread. This is NOT a funding-uncreditable false negative — funding IS credited and accrued by the engine; the carry is simply too thin on BTC. Possible Research-Lead reframe (revise_hypothesis): retry the same delta-neutral cross-venue mechanism on a higher-funding-skew instrument (e.g. a high-funding altcoin perp) where the venue differential exceeds two-venue fees. FAILURE PATTERN: cross-venue funding-differential carry on BTC majors (HL vs Binance USD-M) is genuinely market-neutral but structurally unprofitable — the major-coin funding differential is smaller than the ~0.19% two-venue round-trip cost, so the book bleeds every month (PF 0.76, Sharpe -4.65) with fees ~18% of gross; once the leg-netting artifact is fixed the result is a clean, consistent loss, not an optimizable edge.