SOL Perp 4H Funding-Extreme Momentum Continuation (Single-Instrument USD-M, Long-Short, Follow-the-Crowd Flipped Premise)

A long-SHORT, SINGLE-INSTRUMENT, single-venue directional strategy on SOLUSDT.BINANCE USD-M perpetual futures using 4-HOUR bars, where the funding rate is a SIGNAL only (never credited as a cash flow). This is the FLIPPED PREMISE explicitly recommended by the analyst who abandoned the SOL funding-extreme CONTRARIAN fade: that strategy lost -58.6% over 693 clean trades (PF 0.83, avg_win $882 < avg_loss $968, 47.8% win rate) by FADING crowded funding positioning and catching falling knives — and its consistent negative edge is direct evidence that the INVERSE carries positive expectancy (mirror profile: avg_win > avg_loss, ~52% win rate, gross PF ~1.2). The thesis: extreme funding marks STRONG, PERSISTENT directional demand (longs paying up in an uptrend / shorts paying up in a downtrend), and in crypto that demand tends to CONTINUE rather than immediately reverse — so the edge is to FOLLOW the funding extreme as momentum, not fade it. It runs on the exact same proven-clean path as the failed fade (single-instrument USD-M, avg_position_pct ~22%, no exposure bug, funding data dense throughout, metrics_reliable), so the only thing changed is the directional premise. It avoids every recurring failure this session: NOT BINANCE_SPOT/CASH (metric & fill bugs), NOT BINANCE_CM (not backtestable), NOT options (no Deribit data), NOT cross-venue/Hyperliquid (basis artifact), NOT a multi-instrument pair (extra-leg feed bug), NOT funding-carry (P&L is price-based and fully creditable), and NOT a price-breakout (a different, overfit-prone family). It fills the long-short direction quota (only 14.4% vs heavy 85.6% long-only) and is kept to ~4 parameters to resist the overfitting that killed the breakout/trend strategies.

Iteration 2 fixes the Layer-3 'No trades / should_enter returned a side 0 times over 12427 bars' failure. The entry is gated solely on funding (self._cur_funding_valid plus |rate| >= threshold), and the diagnosis is a funding-parser bug: _build_funding filtered keys with 'if not isinstance(k, int): continue', but supplementary_data['funding_rates'] is dual-keyed ({nanosecond_int: rate, "nanosecond_str": rate}) and the backtest subprocess JSON round-trip stringifies the integer keys, so when the keys arrive as strings the isinstance filter discards EVERY entry, leaving _fund_map empty, _cur_funding_valid permanently False, and should_enter standing aside on every bar -- which is exactly why the contrarian sibling (with correct parsing) produced 693 trades while this one produced zero. The minimal fix replaces the isinstance filter with int(k) parsing inside a try/except so both int and string keys are accepted (and the int/string pair naturally dedups to one map entry). No threshold or logic change was needed: SOL 4H funding regularly exceeds the 0.02% entry threshold, and funding_tol_ns=8h matches the funding cadence so each 4H bar resolves a nearby funding event once the map is populated. All other code -- the nearest-funding lookup, the continuous funding+ROC signal, the follow-the-extreme entry, the normalize/time-stop/stop-loss exits, and the precision-aware sizing -- is byte-identical, so Layers 1-2 stay green. Funding remains a signal only (P&L is price-based) and the strategy still stands aside when funding is genuinely absent (no price-only fallback). leverage stays 1.0 used as a real notional context; single-instrument USD-M avoids the spot/CM/cross-venue/multi-leg failure modes.

The flipped-premise thesis is falsified by its own dense backtest: following the funding extreme as momentum loses just as fading it did. Predicted ~52% win rate and gross PF ~1.2; actual win_rate 39.5% (WORSE than the abandoned fade's 47.8%) and PF 0.933 (<1.0), with expectancy -$52/trade and Sharpe -0.30 (CI -1.64 to +0.94). The +18.95% headline is a mark-to-market mirage — end_unrealized_pct 37.24 means a +37% open un-exited position inflates it; realized PnL is ~ -18%. Because BOTH the contrarian fade (-58.6%) AND this momentum follow lose, the SOL funding-extreme signal carries NO directional edge in either direction — the 'inverse of a loser must be a winner' premise is a fallacy: commission is paid on both sides (3.44% of gross) and the exit structure (funding-normalize / time-stop / stop-loss) is not a clean sign-flip of the fade, so the payoff geometry does not mirror. This is a dense, clean 349-trade sample (not a sparsity artifact), and the loss is concentrated: negative in 5 of 7 years (2020 -16%, 2022 -8%, 2023 -28.6%, 2025 -6.7%, 2026 -3.6%) with the only positive year being 2021's +63.8% mania. Optimization cannot help — no funding_threshold/hold_bars/stop_loss setting manufactures a directional edge that is absent in both directions, and a best-of-225 sweep would at most fit the 2021 window. FAILURE PATTERN: a directional funding-extreme strategy on a single crypto perp has no exploitable sign — fading catches falling knives and following buys exhausted demand, both losing after fees; the 'flip the failed contrarian into momentum' reasoning is invalid because trading costs and an asymmetric exit structure mean the inverse of a fee-bleeding strategy is another fee-bleeding strategy, not its mirror-image winner. The MTM-inflated headline (realized -18% behind a +19% mark) is the diagnostic tell that masks the realized loss.