SolHyperliquidDonchianBreakoutTrendLS1D

SOL Hyperliquid Perp Donchian Channel Breakout Trend, Single-Instrument Long-Short (Turtle-Style, Daily Bars, Pure OHLCV, ATR Stop, Channel-Exit, Low-Parameter)

A LONG-SHORT, SINGLE-INSTRUMENT, pure-OHLCV Donchian channel breakout trend-following strategy on the SOL perpetual on HYPERLIQUID (daily bars). This is a deliberate VENUE + DIRECTION port of the factory's analyst-validated Donchian-breakout trend family (currently represented only as LONG-ONLY strategies on Binance USD-M: AdaDonchianBreakoutTurtleLong, BnbDonchianBreakoutTrendFilteredLong1D, AdaDonchianRisingTrendFilterLong). It is distinct from the already-taken time-series-momentum and impulse-continuation families because the signal primitive is a PRICE-CHANNEL BREAKOUT (close penetrating the extreme of a trailing window), not a trailing-return sign or a single impulse bar. It targets the two most under-represented portfolio buckets simultaneously: HYPERLIQUID venue (5.3% vs 20% target) and long_short direction (13.3% vs push-toward-neutral target). Single-symbol by design to avoid the cross-sectional-basket beta artifacts that broke the funding-carry baskets, and deliberately low-parameter (entry channel, exit channel, ATR stop multiple) to resist the overfitting that killed the single-asset impulse siblings. Hyperliquid's ~0.09% round-trip taker cost is below Binance USD-M, and daily Donchian trends run for many days at multi-percent magnitude, so per-trade moves clear the fee floor with wide margin. No funding, no options, no sub-1H grain, no cross-venue legs — sidesteps every recently-abandoned failure family.

Implements the hypothesis as a venue+direction port of the analyst-validated Donchian-breakout trend family (previously only long-only on Binance USD-M) onto the under-represented HYPERLIQUID venue and long_short direction buckets. The signal primitive is a PRICE-CHANNEL BREAKOUT — close penetrating a trailing extreme — which is genuinely distinct from the already-taken time-series-momentum (trailing-return sign) and impulse-continuation (single outlier bar) families, giving a lower-correlation trade population. calculate_signal returns a CONTINUOUS signed penetration depth in ATR units every bar (positive above the upper channel, negative below the lower channel, 0 inside), so the signal varies and exposes the breakout logic to the verifier; entry direction is thresholded in should_enter and the two-channel/ATR exit lives in should_exit. The wide entry / tight exit asymmetry is the classic Turtle expectancy shape. It is single-symbol by design (avoids the cross-sectional-basket beta artifacts that broke the funding-carry baskets) and deliberately low-parameter (3 core knobs) to resist the overfitting that killed the single-asset impulse siblings. Hyperliquid's ~0.09% round-trip taker cost plus multi-day, multi-percent daily Donchian trends clear the fee floor with wide margin. leverage=1.0 with notional capped at 1x equity keeps sizing honest and clear of the inert-leverage gate while remaining a MARGIN long-short futures strategy. It avoids every recently-abandoned failure family: no funding, no options, no cross-venue legs, no sub-1H grain, no confluence overlay.

Outlier-driven single-day artifact, not a validated edge: ~60% of gross profit comes from one +43.1% day (2024-01-22, $43k win); stripping it collapses profit factor from 1.57 to ~0.64, making the strategy a net loser. 3 of 4 calendar years are negative (only the outlier year 2024 is positive), Sharpe CI straddles zero (-2.89 to 1.83), and skew 12.85 / kurtosis 179.9 / tail_ratio 0.0064 all confirm the equity curve rests on a single tail event. Optimization maximizes in-sample Sharpe and would simply overfit to that Jan-2024 spike, then fail DSR/holdout downstream — not worth 2 hours of compute. This is a premise-level problem (no edge on the SOL/Hyperliquid long-short Donchian instance), not a tunable-parameter or code problem, and it matches this session's breakout-LS family pattern (SOL Donchian LS sibling PF 0.99, DOGE Keltner, BTC inverse Donchian — all abandoned as regime-dependent bull beta with no out-of-sample edge).

Long-short, single-instrument, pure-OHLCV Turtle-style Donchian channel breakout trend strategy on the SOL perpetual on HYPERLIQUID, daily bars. It goes LONG when the daily close penetrates above the highest high of the prior entry_channel (20) completed bars, and SHORT when the close penetrates below the lowest low of that window. It exits on the first of: an opposite-side exit_channel (10-bar) break (a long closes when close drops below the prior 10-bar low; a short closes when close rises above the prior 10-bar high), or an ATR stop at atr_stop_mult*ATR adverse from entry. Channels are computed over bars preceding the current bar (no lookahead). Single position, risk-based sizing (2% of equity at the ATR stop) capped at 1x equity notional, no leverage. Three core parameters (entry_channel, exit_channel, atr_stop_mult).

Trade pattern matches hypothesis: 32 trades, balanced 16 long / 16 short, correct LS implementation of the Donchian-breakout primitive

Reasonable trade cadence for daily bars (~10/yr); no degenerate trade count or code bug

Per-trade edge clears the Hyperliquid fee floor in nominal terms (commission_pct_of_gross 0.95%, capacity $387M)

Entire profitability hinges on a single +43.1% day (2024-01-22, $43k = ~60% of gross profit); removing it drops PF from 1.57 to ~0.64 (net loser)

3 of 4 calendar years negative (2023 -0.3%, 2025 -11.0%, 2026 -3.5%); only the outlier year 2024 is positive

Sharpe CI straddles 0 hard (-2.89 to 1.83); return_skew 12.85 / kurtosis 179.9 / tail_ratio 0.0064 confirm single-outlier dependence

Short side (the new un-validated direction) is the weak leg at 0.25 win rate; max_consecutive_losses 7, max_drawdown 25.5%

Optimization would maximize in-sample Sharpe by fitting to the Jan-2024 spike — guaranteed overfit

SolHyperliquidDonchianBreakoutTrendLS1D ported the validated long-only Donchian breakout family to a long-short SOL instance on Hyperliquid, a low-parameter Turtle-style trend-follower aimed at the under-represented venue and direction buckets. The code was clean (32 balanced trades, fees a non-issue at 0.95% of gross) and the +27.5% / PF 1.57 headline looked plausible, but ~60% of gross profit came from a single +43.1% day — removing it collapses PF to ~0.64 — with 3 of 4 years negative, a Sharpe CI straddling zero, and extreme skew/kurtosis confirming single-outlier dependence. The analyst ruled it a premise-level no-edge artifact that optimization would only overfit, consistent with the session's broader breakout-LS failures. It ended after one iteration as abandoned, never advancing to optimization or risk review.

A flattering daily-Donchian headline that hangs on a single tail day is not an edge: stripping one +43% day flips it to a net loser with 3 of 4 years negative, and the long-only Donchian family did not transfer to a SOL/Hyperliquid long-short instance — matching this session's breakout-LS family pattern (SOL Donchian LS PF 0.99, DOGE Keltner, BTC inverse Donchian) of regime-dependent bull beta with no out-of-sample edge.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) as an outlier-driven single-day artifact rather than a validated edge — a premise-level no-edge problem where optimization would simply overfit the Jan-2024 spike and fail downstream — so optimization and all later stages were never reached.

A single-instrument, pure-OHLCV long/short Turtle-style Donchian channel breakout trend-follower on the SOL perpetual on Hyperliquid (daily bars) — going long on a close above the prior-N-bar high and short on a close below the prior-N-bar low, exiting on a tighter opposite-channel break or an ATR stop — a deliberate venue+direction port of the factory's analyst-validated long-only Donchian family to fill the under-represented Hyperliquid and long_short buckets.

Over ~3 years and 32 trades (16 long / 16 short) the headline looked acceptable — total return +27.5%, Sharpe 0.75 (CI [-2.89, 1.83]), profit factor 1.57, expectancy $826/trade, max drawdown 25.5% — but it rested entirely on one +43.1% day (2024-01-22, ~$43k ≈ 60% of gross profit) whose removal drops profit factor to ~0.64 (net loser), with 3 of 4 calendar years negative, return skew 12.85 / kurtosis 179.9 / tail_ratio 0.006, and the new short side the weak leg at 25% win rate.