SOL Perp Funding-Confirmed Trend (Single-Instrument USD-M, Long-Short, Daily, Carry-Momentum Confluence, Flat on Disagreement)

A long-SHORT, SINGLE-INSTRUMENT, single-venue directional strategy on SOLUSDT.BINANCE USD-M perpetual futures using DAILY bars, that takes a position ONLY when the price trend and the funding rate AGREE in direction, and stays FLAT when they disagree. This is a deliberate, conservative extension of the factory's ONE promoted survivor — the ETH Perp Daily Funding-Confirmed Trend (Sharpe 2.74, at PAPER) — onto a higher-beta major, to test whether the validated confluence mechanism transfers across the beta spectrum. CRITICAL DESIGN CHOICES grounded in this session's accumulated failures: (1) SIGNAL = FUNDING, not OI or taker-flow. The crypto-native-signal-gated family repeatedly hit a data wall because open_interest and taker_ratio (from the metrics files) have multi-year coverage HOLES that fragment the backtest and make walk-forward degenerate; funding_rates is the cleanly-covered series that actually produced the promoted ETH winner. Choosing funding is a direct, evidence-based decision to avoid that data wall. (2) NON-naked-trend. Pure breakout / dual-EMA / Donchian directional momentum is now confirmed dead across 14+ abandons in the 2024-26 chop — but the funding-CONFIRMED variant survived, because the funding-confluence gate filters out the unsupported whipsaws that destroy naked trend systems. The differentiator is the AND-gate with funding plus the flat-on-disagreement discipline, NOT the trend leg alone. (3) NOT a closed family: it is NOT a pairs-ratio MR (crypto ratios are non-stationary — closed), NOT delta-neutral funding CARRY (funding is a directional SIGNAL on a single price-creditable leg, never booked as cash flow), NOT cross-venue / HYPERLIQUID, NOT COIN-M (engine-blocked), NOT drawdown-accumulation, NOT a rare funding-EXTREME gate (which is event-sparse post-2021); it uses the continuous funding SIGN/level so it fires across all regimes and populates OOS windows densely. Intentionally minimal: 3 tunables (trend_lookback, funding_lookback, funding_deadband).

The Layer-3 failure was zero trades (should_enter returned a side 0 times over 2051 bars) for a funding-gated strategy, which is the signature of the funding series never loading. The bug was in _build_funding: it filtered keys with `if not isinstance(k, int): continue`, but supplementary_data['funding_rates'] is delivered with dual keys ({ns_int: rate, "ns_str": rate}) and the backtest subprocess's JSON serialization stringifies the integer keys, so every key failed the isinstance check, _fund_ts/_fund_rate ended up empty, _daily_funding_value always returned None, _cur_funding_avg was always None, and the confluence gate (and the no-funding stand-aside) made should_enter return None on every bar. This is invisible at Layers 1-2 because the synthetic harness passes live (non-serialized) dicts, and only surfaces as zero trades at Layer 3. The minimal fix parses every key via int(k) (tolerant of str) and dedups the int+str dual keys into a dict before sorting -- exactly the pattern used by the working funding strategies -- so the SOL funding series loads, the daily funding mean populates, and the trend+funding AND-gate fires across regimes. Only _build_funding changed; the trend computation, daily-funding windowing, confluence gate, continuous blended signal, exit logic (agreement-break + 10% stop), and capital-relative sizing are byte-identical, preserving the passing Layer-1/2 checks and the no-price-only-fallback discipline. funding_deadband 0.00002 is small relative to SOL's typical ~1e-4 daily funding magnitude, so the funding vote is active without being event-sparse, keeping OOS windows densely populated. leverage stays 1.0 (single position at 15% notional, no amplification) so the unused-leverage gate does not apply; SOLUSDT.BINANCE has clean multi-year daily OHLCV and funding coverage, avoiding the OI/taker data holes the hypothesis was designed around.

The funding-confirmed-trend confluence mechanism transfers to higher-beta SOL only as extreme-tail 2021-mania capture, not a robust edge. return_kurtosis is 356 with skew 17.4 and the daily returns contain implausibly-large single days (+49.1% on 2021-09-24, +41.8% on 2021-03-15, +29.5% on 2021-05-15) that cannot be reconciled with the strategy's own sizing (position_pct 0.15, avg_position_pct 16.8%, leverage 1) — a 15% position cannot return +49% in a day from SOL price alone, so the headline rests on a suspected artifact or multi-day PnL-attribution quirk (symptom flagged; mechanism not independently verified) plus a handful of 2021 pump days. The entire +142.9% is 2021 (+148.5%, larger than the total); the strategy is negative in 4 of 7 years including a -28.6% 2022, with Sharpe 0.54 whose CI straddles 0 (-0.36 to +1.06), a 37.6% max drawdown, and end_unrealized_pct 32.8% (a third of the headline is open mark-to-market). PSR is 0.96 but the kurtosis-356 profile guarantees the deflated-Sharpe gate rejects it (the classic PSR/DSR split). This is materially worse than the BNB sibling cleared to optimize (kurtosis 165, max DD 10.3%, 2022 -2.6%, a distributed 2025 window): SOL has roughly double the kurtosis, 3.6x the drawdown, a catastrophic 2022, and implausible single-day returns — over the line into the documented SOL outlier-luck pattern (SOL Keltner kurtosis 124, NEAR Supertrend, XRP Chandelier) that fails forward. The higher-beta transfer amplified tail-dependence rather than producing a robust edge, so optimizing would at best fit the 2021 spikes and fail DSR. FAILURE PATTERN: transferring a confluence-gated trend mechanism onto a higher-beta major (SOL) does not preserve the edge — it concentrates the return into a few 2021 mania days (kurtosis >300, implausible single-day prints), producing a CI-straddles-0 Sharpe with a large drawdown and a losing 2022; the validated-on-ETH gate is asset-selective and does not robustly survive the beta spectrum's tail-heavy members.