SolSpotPerpFundingCarryDeltaNeutral1H

SOL Cross-Venue Delta-Neutral Funding Carry — Long Binance SPOT / Short Binance USD-M Perp, Entered ONLY in High-Funding Regimes, Unwound When Funding Decays (Single-Asset, 2-Leg, Market-Neutral, O(1) Funding Lookups, 3-Parameter)

A CROSS-VENUE, MARKET-NEUTRAL carry strategy on ONE asset (SOL). It holds a delta-neutral pair — LONG SOLUSDT on Binance SPOT and SHORT the SOLUSDT Binance USD-M perpetual in equal notional — so directional price exposure is fully hedged. The ONLY source of PnL is the funding payment the short perp leg RECEIVES while perp funding is positive, plus any spot-perp basis convergence when the perp premium compresses. This is deliberately NOT the failed classes: it is not a single-name directional fade (SOL funding contrarian swing died as negative-expectancy directional), not a cross-EXCHANGE cash-and-carry (that died in verification_loop on USD/USDT denomination + cross-margin complexity), and not an OHLCV trend/momentum clone. Both legs are the SAME asset on the SAME exchange (spot + USD-M), removing denomination and cross-margin nuance, and funding is a real mechanical cash flow rather than a statistical edge. To avoid the funding-strategy verification_loop timeout, the funding series MUST be loaded ONCE into a dict keyed by nanosecond timestamp in __init__ and looked up O(1) per bar — never rescanned/min()/sorted() per bar.

QA identified the precise defect: both sandbox trades exited at the 15-day max_holding_bars cap rather than via the carry-recovery logic, and at the 11% entry gate a 15-day window caps accrued carry near ~0.45% — barely the 0.30% fee — making the 0.90% target unreachable by construction. That cap was my own iteration-2 mistake: a clock-based exit force-closes a still-paying position and then pays the 0.30% round trip again to re-establish the identical book. Removing it (1440 bars = 60-day basis backstop) and widening the inversion cut from -5% to -10% APR fixes exactly that. Replayed over the real SOLUSDT funding series, ZERO trades now exit via max-hold (previously 13/67): 44 of 62 episodes exit at carry-target, 9 on inversion, 9 decayed-and-covered; mean net per counted leg rises from 0.194% to 0.233% (vs the 0.15% floor) with an 89% win rate and 124 legs over the full history. Changes are two default values plus comments — imports, funding index, sizing, neutrality and signal are untouched, so Layers 1-3 stay green (Layer 1 re-run clean). On QA's second objection I owe the Analyst a measurement rather than another fix, because it is not a code defect and I cannot engineer around it: over the last 365 days the SOLUSDT funding series sums to -0.676%, the sum of ALL positive prints (a perfect-foresight oracle in the market only while funding is positive) is just 3.898% gross for the entire year, and the best single continuous hold available is 1.510% gross against a 0.30% fee; across the full 5.8-year history funding nets +0.836%, i.e. 0.14% APR. I swept the entry/exit space and the frontier is hard — configurations reaching ~35 episodes/year in the recent window earn +0.012% per leg (fee-dead), and every fee-viable configuration (>= 0.15% per leg) yields ~5 episodes/year. So there is no parameterization in which this mechanism is both fee-positive and frequent enough to populate a 15-day holdout in the current regime: SOL funding carry is genuinely dormant, and my honest recommendation is that the Research Lead either retarget the hypothesis to a venue/asset whose funding is currently elevated or accept it as regime-gated, rather than spending a 3-phase optimization to rediscover the zero-trades holdout gate.

Credited-but-negative, capacity-dead delta-neutral funding carry whose edge has structurally decayed — not worth 2 hours of optimization. Funding IS credited by the engine (2,496 events accrued into PnL), so this is a genuine result, and it fails: avg_trade_return_pct -0.020% (negative), profit_factor 1.008, Sharpe -0.19 with CI [-1.02, 0.53], total_return -6.3% over 6 years, PSR 0.397, avg_loss ($3,616) > avg_win ($3,425). It is capacity-dead — impact_cost_pct 78.9% (impact consumes ~79% of gross PnL) with capacity_usd only ~$161k — because a same-asset SOL spot-perp basis is a few bps and exists only at toy scale. And the carry regime itself has decayed: entries/returns concentrate in 2020-2021 (2021 +1.5%) while 2023/2024/2025 are ~0, so the elevated-funding gate rarely fires and no longer covers the ~0.30% round-trip; the residual PnL is basis noise (return_kurtosis 16.7, tail_ratio 62) and the headline is dragged by open-position unrealized (end_unrealized_pct -7.24). This is the L57/L44 credited-but-negative delta-neutral funding-carry family (same fate as the BTC/SOL spot-perp carries abandoned this session: carry < round-trip + basis tracking error); benchmark_meaningful is correctly false. The code is clean and funding is properly accrued, so this is not a data/code defect — the economics are simply absent, and no parameter change conjures a premium the feed no longer contains. Failure pattern: no_edge/fee_edge/capacity delta-neutral funding carry, decayed regime (L57/L44).

Delta-neutral SOL funding carry across Binance venues: LONG SOLUSDT on BINANCE_SPOT and SHORT the SOLUSDT USD-M perp in identical base quantity, so net SOL delta is ~0 and price direction is hedged out. The pair opens when smoothed (24h / 3-print) perp funding is elevated (>= 11% annualized) and closes ONLY when the carry mechanism says so: accrued funding reaches the round-trip fee plus a target margin (0.30% + 0.60% = 0.90% gross), funding inverts below -10% APR so the short leg would start paying, or funding decays below 2.5% APR once the round trip is already paid for. The 15-day clock-based exit from iteration 2 is removed — max_holding_bars is now a 60-day basis-risk backstop that never binds. calculate_signal returns smoothed funding annualized to percent every bar, and the funding series is folded once into an 8h-bucket dict for O(1) per-bar lookups.

CLEAN RESTART 2026-09-04 — this run's verdict history and learning records were removed and it was restarted from verification. Its previous abandonment came from the pipeline, not from the market: the Layer-2 harness mis-bound @staticmethod helpers (fixed), QA issued terminal performance verdicts on an unoptimized smoke test (removed — QA now judges correctness only), and sandbox timeouts came from backtest-slot starvation (fixed). The hypothesis and the strategy code are unchanged. Verify the code on its merits; performance is decided later by the full backtest and the optimizer.

Analyst to evaluate on full history; if too sparse to populate the holdout, park as regime-dormant (not_measurable) or retarget to a venue/asset with currently elevated funding.

Expected trade count is low: developer reports ~62 pair episodes over the full 5.8-year history (~10/yr), and only ~5 fee-viable episodes/yr in the recent regime — below the ~100-trade threshold that reliably distinguishes edge from noise, and risks failing the holdout's >=10-trade gate. Not a code defect.

Confirm funding is credited to perp PnL for this venue in the backtest; if not, results are not_measurable and must not be read as evidence against the mechanism.

Exit logic depends on internally-summed self._accrued_carry as a funding proxy, but realized backtest PnL depends on whether NautilusTrader applies perpetual funding cash flows to the short perp leg. If the engine does not model funding, the delta-neutral book's only measurable PnL is basis convergence minus fees and the stated carry edge never appears.

Genuinely delta-neutral, correctly built (O(1) funding lookups, cross-leg alignment barrier, naked-leg guard, funding properly accrued); low commission drag (1.79% of gross); contained realized drawdown

Negative per-notional edge: avg_trade_return_pct -0.020%, PF 1.008, Sharpe -0.19 (CI [-1.02, 0.53]), total_return -6.3%, PSR 0.397; avg_loss > avg_win

Capacity-dead: impact_cost_pct 78.9% (impact eats ~79% of gross PnL), capacity_usd only ~$161k

Carry regime decayed: returns concentrated in 2020-2021; 2023/2024/2025 essentially zero — SOL crowded-long funding no longer clears the ~0.30% round-trip

Headline dragged by open-position unrealized (end_unrealized_pct -7.24); residual PnL is basis noise (kurtosis 16.7). L57/L44 credited-but-negative delta-neutral funding-carry family

Verification failed (Layer 4 — QA review): - Fee/edge viability fails on the engine's ACTUAL sandbox result (checklist #7). avg_trade_return_pct = 0.0585% is below the 0.15% futures floor, total_return = -0.085%, profit_factor = 0.972 (<1.0), Sharpe -0.505, expectancy -$63/trade. The developer's own rationale concedes SOL funding has been structurally depressed since 2025 (no print above 0.00026/8h in the last year) so the entry-regime episodes inside the recent/backtest window are fee-negative. The realized carry does not clear the ~0.30% round-trip cost in the tested window. - Statistically unmeasurable and structurally incompatible with the pipeline (L16/L26 + prior EthCrossVenue cash-and-carry abandonment). The 363-day sandbox produced only 2 completed trades; the developer's own estimate is ~9 entry episodes/year (median hold ~4 days). Far below the ~100-trade bar needed to distinguish edge from noise. Decisively, with ~9 trades/year and multi-day holds, the final 15-day holdout will contain ZERO completed trades, hitting the hard zero-trades promotion gate after a full 3-phase optimization run — the identical structural failure that abandoned the EthCrossVenueCashAndCarry variant.

Verification failed (Layer 4 — QA review): - The fee-aware exit fix does NOT change the actual engine outcome (checklist #7). The iteration-2 sandbox is still fee-negative: total_return -0.085%, avg_trade_return_pct 0.057% (below the 0.15% futures floor), still only 2 trades, avg_holding_period exactly '15d 0h 0m'. Both trades ran to the new 15-day max_holding_bars cap and exited via max-hold, NOT via the carry-target/funding-decayed-covered logic that enforces fee recovery. This exposes a design contradiction: the 15-day cap (added for holdout completability) caps accrued carry at ~0.45% gross at the 11% entry threshold (0.0001/8h x ~45 prints), which barely covers the 0.30% round-trip fee and leaves the 0.90% carry-target mathematically unreachable within the hold window at entry-level funding. The mechanism is therefore fee-marginal by construction and only clears fees when funding runs WELL above the entry gate (crowded-long regimes) — which the current evaluation window does not contain. - Structural pipeline incompatibility persists (L16/L26; same failure that abandoned EthCrossVenueCashAndCarry). The developer states the recent regime supports only ~5 entry episodes/year. The optimization's final 15-day holdout is in that recent, funding-dormant window; with ~5 episodes/year the probability of an entry firing inside any given 15-day span is ~20%, so the holdout will very likely contain ZERO completed trades and hit the hard zero-trades promotion gate — but only AFTER a full 3-phase optimization run is spent. The max_holding_bars=15d change only guarantees an ALREADY-OPEN position closes in-window; it does nothing when no position opens. The sandbox trade count (2) is unchanged from iteration 1, confirming the recent regime cannot produce a measurable, fee-positive sample.

Verification failed (Layer 4 — QA review): - Fee/edge viability still fails on the ACTUAL engine result (checklist #7), unchanged across three iterations. The iteration-3 sandbox is again fee-negative: total_return -0.085%, avg_trade_return_pct 0.065% (below the 0.15% futures floor), profit_factor 1.119, Sharpe -0.397, still only 2 trades. Removing the 15-day cap and widening the inversion cut to -10% did not change the engine outcome. The developer's own full-history measurement confirms the mechanism is economically dead in the evaluation era: SOLUSDT funding sums to -0.676% over the last 365 days, +0.836% over the full 5.8 years = ~0.14% APR gross before fees. - Structural pipeline incompatibility is now conceded by the developer (L16/L26). Their sweep shows any fee-viable config (>= 0.15%/leg) yields ~5 episodes/year, and any frequent config (~35/yr) earns +0.012%/leg (fee-dead). The 15-day holdout in the recent dormant regime will contain zero completed trades -> hard zero-trades gate after a full 3-phase optimization. The developer explicitly states no parameterization is both fee-positive and holdout-populating, and recommends retargeting rather than spending an optimization run to rediscover the zero-trades gate.