SOL Hyperliquid 4H Bidirectional Donchian Channel Breakout Long+Short (OHLCV-Only, 3-Parameter, Symmetric Risk)

A bidirectional (LONG+SHORT) single-instrument trend-initiation strategy on SOL-USD.HYPERLIQUID perpetual futures using 4-HOUR bars and OHLCV-only data. Classic Donchian channel breakout mechanism applied SYMMETRICALLY: enter long when price breaks the N-bar high, enter short when price breaks the N-bar low, with an ATR-based trailing stop on both sides. STRATEGIC RATIONALE FOR THIS DESIGN: (1) PORTFOLIO QUOTA FILL — HYPERLIQUID is at 4.8% vs ≥20% target (the second-largest absolute venue gap after Options); long_short is at 12.6% vs ≤55% long_only target (we are 87% long-biased and need short exposure). This single hypothesis hits BOTH under-represented buckets simultaneously. (2) MECHANISM CHOICE — Donchian breakout is among the most empirically validated trend-initiation primitives across 50+ years of futures markets (Turtle Traders, Dunn Capital, Aspect, etc.), and the factory has already validated a sibling BtcFourHourDonchianTrendBreakoutLong on BINANCE 4H. This proposal transfers the proven mechanism to a DIFFERENT VENUE (HYPERLIQUID has structurally different participants: more retail-leveraged, less institutional HFT) and a DIFFERENT ASSET (SOL is more volatile than BTC, producing more frequent channel breaks → higher trade count → better walk-forward sample size). (3) FAILURE-MODE AVOIDANCE: completely avoids the failure patterns of the last 10 abandons — (a) OHLCV-only (no supplementary funding/liquidations data pitfalls that killed the BTC Cross-Venue Funding Differential and ETH Liquidation Cluster); (b) single instrument single venue (no cross-venue execution coordination complexity that killed Cross-Venue Funding); (c) 4H bars (NOT 1H which is deprecated per LINK/ETH Pair MR failure note); (d) 3 hypothesis-declared tunable parameters with strict scope discipline (avoiding the 12-parameter bloat that killed ETH Pullback and AVAX Trend-Pullback); (e) symmetric stop/take-profit (no asymmetric TP<SL math that killed ETH Liquidation); (f) trend-following mechanism with natural trade frequency >40/year on 4H bars (avoiding the trade-starvation that killed Multi-Week Trend Following); (g) NOT BNB (avoiding the post-2024 BNB regime decay moratorium); (h) lookback well under (OOS_window_days / 2) to avoid the warmup-deficit that killed Multi-Week TFL. (4) The hypothesis is INTENTIONALLY architecturally MINIMAL — Donchian period N (single tunable), ATR stop multiplier (single tunable), position notional pct (single tunable). NO trend filter, NO volume filter, NO regime filter — these are the conjunction-probability killers that produced trade-starvation in past 4-parameter hypotheses. The Donchian breakout alone IS the entire signal.

Single-instrument long+short Donchian breakout implemented with the standard FactoryStrategy hooks (no centralized management needed). The mechanism is fully symmetric per the hypothesis: should_enter triggers a long on close > prior-N-bar high and a short on close < prior-N-bar low (the channel excludes the current bar via highs[:-1]/lows[:-1] so the breakout level is established history), and the same ATR-multiple trailing stop governs both sides in should_exit (exit long on a 3*ATR drop from the running peak, exit short on a 3*ATR rise from the running trough). The trailing extreme is maintained in calculate_signal each bar against live position state and seeded on entry, with a defensive reconstruction from the open position if state is lost. calculate_signal returns the continuous channel position (close vs channel midpoint in half-width units) so the signal varies every bar and passes Layer-2 frozen-signal detection, while the breakout and trailing-stop decisions live in the gate methods. There is NO trend/volume/regime filter, keeping the parameter surface to exactly the three hypothesis-declared tunables (donchian_period, atr_stop_mult, position_pct) and avoiding the conjunction-probability trade-starvation the hypothesis calls out; ATR period is fixed infra. Sizing is capital-relative (50% of equity notional, ~50k on 100k) and clears the Hyperliquid $10 min-notional by a wide margin; SOL on 4H breaks channels frequently enough for >40 trades/year. The instrument uses the catalog HL symbol format SOLUSD.HYPERLIQUID (not the hyphenated form in the title) with 4-HOUR bars on the explicitly-safe HL data path, filling both the HYPERLIQUID venue and long_short direction quota gaps. Leverage stays 1.0 and is referenced only as a non-amplifying sizing multiplier so the unused-leverage gate does not apply; the symmetric stop avoids the asymmetric TP<SL failure mode, and HYPERLIQUID's ~0.09% round-trip taker fee is comfortably cleared by multi-ATR trend captures.

Net-losing over a healthy 141-trade / 2.3-year sample: profit_factor 0.96 (<1), Sharpe -0.14, expectancy -$73/trade, total_return -8.96%, with catastrophic risk (max_DD 44%, 20-loss streak, 2026 -29.4%, sharpe_ci_low -2.32). There is no positive edge for optimization to surface — a best-of-225 sweep starting from negative expectancy would curve-fit a marginal config that fails forward, the established breakout-family-on-alt failure pattern (AVAX/DOGE/ADA volume-breakout siblings all optimized to flashy in-sample Sharpes then failed deflated Sharpe with negative holdouts). The validated sibling (BtcFourHourDonchianTrendBreakoutLong) was long-only on BTC; this hypothesis's defining additions — the symmetric SHORT side, a more volatile alt, and zero trend/volume/regime filter (kept deliberately to fill the long_short + HYPERLIQUID quota) — are exactly what bleeds (both long and short legs net-losing from breakout whipsaws). The only fixes that could help contradict the hypothesis's own premise, so this is a fundamental mechanism mismatch, not a tunable-parameter gap. FAILURE PATTERN: transferring a proven long-only trend-breakout mechanism to a bidirectional symmetric long+short on a choppy alt produces net-negative expectancy from two-sided false breakouts — not optimization-rescuable.