EthCrossVenuePositiveFundingBasisCarryNeutral4H

ETH BINANCE_SPOT + ETH HYPERLIQUID Perp Cross-Venue Positive-Funding + Basis Carry (Market-Neutral Pair, 4H Monitoring, 3-Parameter, No Directional Stop)

A DOLLAR-NEUTRAL CROSS-VENUE PAIR strategy that captures TWO stacked structural edges on ETH: (a) Hyperliquid perp funding rate income when funding is persistently positive, and (b) basis convergence between Hyperliquid perp and Binance spot when HL perp trades at a premium. Mechanism: on each 4H bar close, read the latest HL 8h funding rate and compute the HL_perp_close / BINANCE_SPOT_close - 1 basis. When BOTH conditions are met simultaneously (HL funding >= +0.02% per 8h, AND HL/spot basis >= +0.15%), open a market-neutral pair: LONG ETH on BINANCE_SPOT and SHORT ETH on HYPERLIQUID perp, dollar-matched at 20% of equity per leg (40% total committed). While the pair is open, the short HL leg RECEIVES the positive funding payments every 8h (cash income), and we wait for the HL/spot basis to converge (basis P&L). Exit the pair (close BOTH legs simultaneously at market) when ANY of: HL funding drops to < 0 (carry source exhausted), basis converges to <= 0 (basis profit realized), or 14 calendar days elapse (time stop). No directional stop because the position is dollar-neutral; instead a portfolio-PnL kill-switch closes both legs if combined unrealized loss exceeds -5% of equity (catches execution skew / basis blowout). This explicitly avoids the failed mechanism classes from this session: NOT a momentum/breakout strategy (regime-decayed), NOT a single-leg short on funding (wrong-side-of-bull-regime), NOT cross-sectional on correlated majors (correlation-collapsed), NOT a pair MR on perps (4-leg fee gap). The edge is MECHANICAL CASH FLOW + BASIS CONVERGENCE on a SINGLE asset across TWO venues, not statistical mean-reversion or directional bet. Hits 4 under-represented portfolio quotas at once: HYPERLIQUID (4.9% vs 20% target), BINANCE_SPOT (10.5% vs 15% target), CROSS-VENUE (6.3% vs 15% target), LONG+SHORT direction (12.6% vs 45% headroom).

The feedback confirmed the strategy is correctly implemented but far too selective (4 entries / 28 months -> ~1 trade per walk-forward window, which would fit noise), and identified the 0.15% basis AND-gate as the binding constraint because an ETH HL-perp/Binance-spot basis rarely reaches 15 bps. The smallest fix is the prescribed parameter loosening plus the requested density diagnostic, leaving all mechanism logic intact: (1) entry_basis_pct default 0.0015 -> 0.0, so the basis is a soft non-discount filter rather than a hard co-requirement, letting the funding carry (the edge the engine reliably credits to the short leg via funding_accrual) be the primary driver; (2) entry_funding_8h default 0.0002 -> 0.00005, toward the persistent positive-funding floor so entries fire across the positive-funding regime; (3) a one-time-style density log (emitted every 1000 bars with running fractions) reports, over the full span, how often funding>=threshold, basis>=threshold, and BOTH hold -- directly showing whether a tradeable carry density exists, exactly as requested. Everything else -- the funding-series parsing (module-level helpers to survive the Layer-2 proxy), contemporaneous-basis computation, dual-leg entry (HL SELL + spot BUY dollar-matched), the funding-exhaustion / basis-convergence / time-stop / kill-switch exits, the two-leg flatten, and dollar-neutral sizing -- is byte-identical, so the passing earlier layers and the engine-credited funding mechanism are preserved, and the no-price-fallback discipline (unknown funding => no entry) is retained. leverage stays 1.0 (dollar-neutral; leverage>1 would reintroduce ETH beta) so the unused-leverage gate does not apply. Per the feedback's decision rule: re-run and confirm >=50 pair trades with a non-flat positive net return after fees; if the loosened gates still show the funding+basis carry is intrinsically too thin on ETH (the density log will quantify this), abandon or reframe onto a venue/asset with a persistently richer funding/basis rather than thrash.

The ETH cross-venue funding+basis carry is correctly built and market-neutral (beta 0.0002, vol 0.48%), funding IS credited (per the engine's funding-accrual ledger), and the HL funding history loaded fine (496 days, 268 entries) -- so this is NOT a creditability or data-wall issue. Judged on its CREDITED result, the strategy is a consistent net loser: Sharpe -6.48, profit_factor 0.94, expectancy -$15.66/trade, CAGR -1.77%, with a relentless -0.02% to -0.05%/day bleed and negative returns in nearly every month. ROOT CAUSE: the credited ETH-HL funding+basis carry is too thin to cover the cross-venue execution cost -- commission is 10.51% of gross on a 4-leg round-trip (enter+exit on two venues), and avg_win ($499) is essentially equal to avg_loss ($507), meaning the funding income does not systematically tilt the book positive. The iteration history confirms a structural bind, not a tuning miss: at the selective 0.15% basis gate (iteration 1) it fired only 4x in 28 months (too rare to optimize), and relaxing to basis>=0 / funding>=0.00005 (iteration 2) fires 268x but each trade is net-negative -- there is no threshold middle where ETH HL carry is both frequent enough to matter and fat enough to clear fees. Not optimize: the 3 tunables cannot escape this bind -- raising the funding threshold walks back toward the 4-trade untestable regime, and the binding cost (fees vs a small credited carry on an efficient major) does not move with the parameters. Not iterate: the mechanism is correct (delta-neutral, funding credited, basis computed) and already iterated once; there is no bug, the edge simply isn't there net of fees. FAILURE PATTERN: a single-asset cross-venue funding+basis carry on a major (ETH, Binance-spot vs HL-perp) does not clear its own cross-venue fees -- the credited carry is structurally too thin (commission 10.51% of gross, avg_win~avg_loss, steady negative drip, Sharpe -6.48), and it is caught in a threshold bind (selective => 4 trades/28mo; relaxed => 268 net-negative trades). This is a measured thin-edge-vs-fees result on the CREDITED carry, distinct from the earlier (now-reversed) creditability concern; the carry on an efficient-funding major is real but smaller than the 4-leg execution cost.

Dollar-neutral cross-venue ETH carry pair: short ETHUSD.HYPERLIQUID perp + long ETHUSDT.BINANCE_SPOT, dollar-matched at 20% equity per leg, monitored on 4H bars. The short HL leg receives positive funding (the engine-credited cash-flow edge) and the position earns HL/spot basis convergence. Iteration 2 loosens the over-selective entry gate that fired only 4 times in 28 months: entry_basis_pct is dropped to 0.0 (require only that HL is not at a discount, making basis a soft filter) and entry_funding_8h is lowered toward the persistent-funding floor (0.005%/8h), so entries are driven primarily by the positive-funding carry. It adds a full-span density log reporting the fraction of bars where funding / basis / both gates clear, so the proceed-vs-abandon decision is data-driven. Exits on funding turning negative, basis convergence, a 14-day time stop, or a -5% combined-PnL kill switch. Three tunables: entry_funding_8h, entry_basis_pct, per_leg_pct.

Correctly market-neutral (beta 0.0002, vol 0.48%, max_drawdown 7.4%, no liquidation); funding IS credited and HL funding history loaded (496d) — mechanism built as designed

Dense firing after the iter-2 threshold relaxation (268 pairs), so the negative result is a real measurement, not sparsity

Consistent net loss: Sharpe -6.48, profit_factor 0.94, expectancy -$15.66/trade, CAGR -1.77%, negative in nearly every month — a steady -0.02% to -0.05%/day bleed

Credited carry too thin to beat cross-venue fees: commission_pct_of_gross 10.51% on a 4-leg round-trip; avg_win $499 ~ avg_loss $507 (funding income not tilting the book positive)

Structural threshold bind: selective gate (iter1, 0.15% basis) fired 4x/28mo (untestable); relaxed gate (iter2) fires 268x but net-negative — no profitable middle

ETH is an efficient-funding major; the persistent-positive-carry premium is small relative to the cross-venue execution cost

The strategy is correctly implemented but far too selective to optimize: the combined entry gate fired only 4 times in ~28 months (8 leg-trades, data_days=26). With ~1 trade per walk-forward window, optimization would fit pure noise and the holdout would be degenerate. Before re-running optimization, raise the trade count to a statistically meaningful level (target >= 50 pair trades over the span) and confirm density via the FUNDING COVERAGE / basis distribution that the strategy already logs. SPECIFIC CHANGES: (1) The 0.15% basis AND-gate is the binding constraint — an ETH Hyperliquid-perp/Binance-spot basis rarely reaches 0.15%, so it suppresses nearly all entries. Lower entry_basis_pct substantially (e.g. test 0.0003-0.0005, or 0.0) so the strategy is driven primarily by the positive-funding carry (the edge that is now reliably credited by the engine), with basis as a secondary filter rather than a hard co-requirement. (2) Lower entry_funding_8h toward the persistent-funding floor and verify how often HL funding alone clears it. (3) Add a one-time log that reports, over the full span, the fraction of bars where funding>=threshold, where basis>=threshold, and where BOTH hold — this directly shows whether a tradeable density exists. (4) Re-backtest and confirm >= 50 trades and a non-flat net return before promoting to optimization. IMPORTANT: if the diagnostic shows the funding+basis combination is intrinsically rare on ETH even after loosening, or the looser gates produce no positive net edge after fees, this should be abandoned (or reframed onto a venue/asset with a persistently richer funding/basis) rather than thrashed — report the distribution so the next decision is data-driven.

EthCrossVenuePositiveFundingBasisCarryNeutral4H tried to harvest two stacked structural edges on ETH — Hyperliquid positive funding income and HL/spot basis convergence — by holding a dollar-matched short-perp/long-spot pair, explicitly avoiding the directional and statistical mechanisms that had failed earlier in the session. The build was sound: it was genuinely market-neutral (beta 0.0002), funding was credited by the engine, and 496 days of HL funding history loaded cleanly, so the negative result was a real measurement rather than a data or creditability artifact. But judged on its credited result it bled steadily — Sharpe -6.48, profit_factor 0.94, CAGR -1.77% — because commissions ran 10.51% of gross on the four-leg round-trip and the thin carry left avg_win essentially equal to avg_loss. After two iterations exposed a structural threshold bind (selective gate fired only 4x in 28 months; relaxed gate fired 268x but net-negative), the analyst abandoned it at the backtest-review gate, concluding the carry on an efficient major is real but smaller than its own cross-venue execution cost, so it never reached optimization.

A single-asset cross-venue funding+basis carry on an efficient-funding major like ETH is caught in a threshold bind with no profitable middle — a selective gate fires too rarely to test (4x/28mo) while a relaxed gate fires often but net-negative — because the persistent-positive-carry premium is smaller than the 4-leg cross-venue execution cost.

The analyst abandoned it at the backtest-review gate (verdict: abandon) because the credited carry was structurally too thin to cover cross-venue fees — funding income did not tilt the book positive, producing a steady -0.02% to -0.05%/day bleed negative in nearly every month; optimization was never run.

A dollar-neutral cross-venue ETH carry pair that shorted the Hyperliquid perp and went long Binance spot when HL funding was positive and the HL/spot basis was non-negative, aiming to stack mechanical funding-income cash flow plus basis-convergence P&L while staying delta-neutral, monitored on 4H bars with three tunables.

It was correctly market-neutral (beta 0.0002, vol 0.48%, max_drawdown 7.4%, no liquidation) and fired densely after the iteration-2 threshold relaxation (268 pairs / 536 legs over 496 days), but was a consistent net loser: Sharpe -6.48, profit_factor 0.94, expectancy -$15.66/trade, CAGR -1.77%, with avg_win ($499) ≈ avg_loss ($507) and commissions at 10.51% of gross on the 4-leg round-trip.