SolHyperliquidDailyDonchianBreakoutLongFlat

Hyperliquid SOL Daily Donchian Breakout, LONG/FLAT (Single-Instrument HYPERLIQUID Perp, Pure OHLCV, Simple Fixed Single-Entry Sizing, No Flips, Clean-Sizing Venue While BINANCE USD-M Perp Is Engine-Corrupted, Tail-Safe, Low-Parameter)

A LONG-ONLY (LONG/FLAT, never short), SINGLE-INSTRUMENT, DAILY Donchian-channel breakout trend strategy on SOLUSD.HYPERLIQUID. It completes the clean-sizing HYPERLIQUID trend sleeve across the three core majors (BTC + ETH siblings pending), chosen because the factory backtest engine has a CONFIRMED over-exposure defect SPECIFIC to BINANCE USD-M perp — proven by the BTC long/flat Donchian 'definitive reproducer' that, despite being purpose-built (no shorts, no flips, fixed single-entry size, dual flat guard) and pre-registering an acceptance test, still over-deployed to avg_position_pct 429.7% with impossible single-day returns; that run isolates the defect to the base USD-M perp position_size->order-qty->equity pipeline itself. BINANCE spot-CASH (MTM defect) and multi-instrument (aggregation defect) are likewise broken, and census signals (liq/OI/account-ratio/taker) are data-walled. The ONLY clean evidence point is HYPERLIQUID: the prior HL run sized correctly (~20-29% capital, metrics_reliable=true), failing only on short data — localizing the over-exposure bug away from HL. This applies the factory's VALIDATED daily trend edge to the highest-beta core trender (SOL) on the one clean-sizing venue, in the simplest evaluable form: single-instrument, OHLCV-ONLY, LONG/FLAT (no flip-stacking), ONE fixed-fractional size at entry with NO dynamic resizing, leverage 1.0, close-to-flat before re-entry. SOL is the highest-beta HL major (largest trend amplitude) and HL is the most UNDER-REPRESENTED venue (6.4% vs 20% target). PRE-REGISTERED ACCEPTANCE / SIZING CHECK: avg_position_pct ~<=100% and single-day returns within ~+/-30% (SOL at 1x) — prior HL sizing (~20-29%) suggests it passes; if it over-exposes, the defect is engine-wide (escalate). Low parameter count.

Implements the hypothesis by applying the factory-validated daily trend edge to the highest-beta core major (SOL) on HYPERLIQUID, the one venue with confirmed-correct position sizing (prior HL run deployed ~20-29% capital, metrics_reliable=true) while BINANCE USD-M perp carries a proven over-exposure defect. The design deliberately minimizes exposure to known engine defects: (1) single-instrument so the multi-instrument aggregation defect cannot apply; (2) pure OHLCV with no funding subscription; (3) LONG/FLAT with no flips or resizing, mirroring the clean reproducer structure so any over-exposure is attributable to the venue's position_size->order-qty->equity pipeline rather than flip-stacking; (4) ONE fixed-fractional size (notional = 0.95 * equity, qty = notional/price) computed once at entry and frozen, leverage left at 1.0 so the engine margin model and in-strategy notional agree (no leverage referenced in sizing, avoiding the leverage_set_but_unused gate). The Donchian breakout + SMA regime filter is the validated daily trend edge; ATR-normalized score keeps the signal continuous and conviction-bearing. Daily bars match the long-term timeframe and keep fee drag negligible on HL (~0.09% RT, few trades/year). Pre-registered sizing check: avg_position_pct should be <=~100% and single-day returns bounded within ~+/-30% (SOL at 1x); a violation would indicate the over-exposure defect is engine-wide and should be escalated.

Failed deflated Sharpe: DSR=0.0, expected-max=11.71 over 225 trials — the optimized Sharpe (2.91) is far below the best-of-N luck bar and statistically indistinguishable from selection noise (is_significant=false, PBO=0.5926>0.5, deflated_sharpe=0.0). Every decisive generalization gate fails in concert even though SENSITIVITY PASSED (0 cliffs): walk-forward out-of-sample Sharpes are [0.0, 0.0, 0.0] against in-sample [12.24, 10.05, 10.70] with is_overfitted=TRUE (the entire ~11 avg IS Sharpe vanishes OOS), and the holdout FAILED with a NEGATIVE holdout_sharpe -2.60 (ratio 0, passed=false) — the once-tested recent window loses money. Sensitivity passing only confirms full-period in-sample smoothness; it does not measure generalization, and the generalization gates (OOS, holdout, DSR) all fail. Root cause (symptom, not independently verified): the SOL-HL trend edge is concentrated in the 2023-2024 bull (annual 2024 +338%) and decays/reverses thereafter (2025 -2.8%, 2026 -19.9%), on only ~3 years of Hyperliquid history with 15 trades whose +247% headline rests on a single 2024-01 bull-recovery trade (skew 8.43, kurtosis 75.68) — the OOS walk-forward windows are empty/edgeless and the recent holdout is negative, so the sample is too short and too regime-concentrated to survive forward-testing or multiple-testing deflation. Not iterate (attempt 1 of 2): the failure is OOS non-generalization + deflation + a negative holdout, not a tunable parameter region — sensitivity already passed, so a second sweep would re-find the same IS-positive / OOS-zero / negative-holdout pattern on the same thin data; there is no robust region to tune toward when all three OOS windows are zero and DSR is 0.0. Not revise_hypothesis: the daily Donchian breakout is being validated per-instrument and SOL was the highest-beta HL major (best-case candidate); it fails purely because HL's ~3-year history is too short to validate a low-frequency daily-trend edge — an engine/data limitation, not a premise a different instrument reframes. This is the THIRD HL daily-Donchian instance this session (BTC-HL, ETH-HL) to die on the identical signature: all-zero walk-forward OOS, a negative holdout, DSR 0.0, PBO >0.5. FAILURE PATTERN: single-instrument daily Donchian breakout on Hyperliquid majors cannot be validated on HL's ~3-year history — even the highest-beta sibling (SOL, sensitivity-passing, 0 cliffs) collapses to all-zero walk-forward OOS, a negative holdout (-2.60), DSR 0.0 and PBO 0.59 because the trend edge concentrates in the 2023-2024 bull and a fat-tail single-trade headline defeats forward-testing and multiple-testing deflation. PRESERVED FINDING: HL sizes cleanly (avg_position_pct 95.8%, metrics_reliable=true, 0 dropped entries) — the hypothesis's sizing acceptance check passes and the BINANCE USD-M over-exposure defect is HL-absent; only the trend edge is unvalidatable on this venue's short data.

Long/flat daily Donchian-channel breakout trend strategy on SOLUSD.HYPERLIQUID (Hyperliquid perp), pure OHLCV. Enters long when the bar closes above the prior 20-bar Donchian high while above the 50-bar SMA trend filter. Exits on a trailing 10-bar Donchian low break, a 3x-ATR hard stop frozen at entry, or an 18% percentage stop (widened for SOL's higher beta). Never shorts, never flips; must be fully flat before re-entering. One fixed-fractional position (95% of equity notional) set once at entry with no dynamic resizing, leverage 1.0. Continuous signal = (close - Donchian high)/ATR in ATR units.

Pre-registered SIZING PROBE PASSED: avg_position_pct 95.76% (<=100%), exposure_pct 47.3%, end_unrealized_pct 7.6%, metrics_reliable=true — clean sizing confirmed, completing the HL clean-sleeve across BTC/ETH/SOL.

The +222% single-period return is a legitimately captured SOL Q4-2023 trend leg (~$25->$85 over the hold), corroborated by clean avg_position_pct and low end-unrealized — NOT an over-exposure artifact.

Strong trend behavior: profit_factor 2.0, avg_win/avg_loss ~4:1, tail_ratio 1.78, omega 4.65, Sharpe 1.77, Sortino 13.3, Calmar 10.8.

Sound, bug-avoiding implementation identical to the clean HL siblings (long/flat, no look-ahead, one fixed-fractional entry, leverage 1.0).

Only 15 trades over ~3 years — borderline too few for the optimizer to fit without curve-fitting to noise.

SINGLE-TRADE DEPENDENCY: the entire return is carried by one Q4-2023->Jan-2024 trade (+222%); excluding it leaves the strategy roughly flat/negative (2023 -0.4%, 2025 -2.8%, 2026 -19.9%).

Sharpe CI -2.78 to +3.24 straddles 0; max_consecutive_wins=1, return_skew 8.4, kurtosis 75.7 — one fat-tail outlier drives everything.

Recent years negative (2025, 2026) — the edge is concentrated in the 2024 bull and has not recurred since.

3-window walk-forward will isolate the monster trade in one window with ~5 trades/window elsewhere; deflated Sharpe over ~225 trials on this sample is very likely to fail.

Clean sizing on HYPERLIQUID confirmed: avg_position_pct 95.8% (initial) / 53.2% (optimized), metrics_reliable=true, leverage 1.0 honored, entry diag shows 0 dropped/size-zero entries — the hypothesis's pre-registered sizing acceptance check PASSES and the BINANCE USD-M over-exposure defect is absent on HL.

Strategy is mechanically sound and simple: LONG/FLAT only (0 short trades), single-instrument pure-OHLCV, fixed-fractional single-entry, no flips. Trades implement the stated Donchian-breakout mechanism.

Sensitivity surface is smooth (0 cliffs across all 9 parameters) and capacity is large (>$1B) — no parameter fragility and no capacity blocker.

Walk-forward out-of-sample Sharpe is [0.0, 0.0, 0.0] across ALL three windows vs in-sample [12.24, 10.05, 10.70]; is_overfitted=TRUE, avg OOS 0.0. The entire edge vanishes out-of-sample.

Holdout FAILED: holdout_sharpe -2.60, ratio 0 — the once-tested recent window loses money.

Failed deflated Sharpe: DSR=0.0, is_significant=FALSE, PBO=0.5926 (>0.5, overfitting more likely than not). Optimized Sharpe 2.91 is far below the expected-max-under-noise bar of 11.71 over 225 trials — indistinguishable from best-of-N selection luck.

Headline return is fat-tail-driven: +247% total return rests on a single 2024-01 bull trade (skew 8.43, kurtosis 75.68); annual returns are 2024 +338% but 2023 -0.4%, 2025 -2.8%, 2026 -19.9% — a single-regime bull artifact, not a persistent edge.

Thin/regime-concentrated sample: only 15 trades (initial) / 20 (optimized) over HL's ~3-year history; sharpe_ci_low -2.78 (initial) straddles 0; max_drawdown 41% with a 514-day drawdown duration.

SolHyperliquidDailyDonchianBreakoutLongFlat ported the validated daily-trend edge to the highest-beta HL major to round out the clean-sizing BTC/ETH/SOL sleeve, and its default backtest dazzled — +247%, PF 2.0, Sharpe 1.77 — earning a trip to optimization. But the reviewer already flagged single-trade dependency, and it proved decisive: the headline was one 2024 bull trade (skew 8.43, kurtosis 75.7), with 2023, 2025, and 2026 flat-to-negative. Optimization improved the in-sample numbers and even passed sensitivity (0 cliffs), yet the generalization gates collapsed in concert — walk-forward OOS [0,0,0], holdout -2.60, DSR 0.0, PBO 0.59 — exactly as its BTC and ETH siblings had. The analyst abandoned it as the third identical HL daily-Donchian failure, concluding the trend edge is real in-sample but unvalidatable on Hyperliquid's short history, while preserving the cross-cutting finding that HL sizes cleanly and the Binance USD-M over-exposure defect is absent there.

A daily-trend edge whose entire return depends on one fat-tail bull trade cannot be validated on Hyperliquid's ~3-year history — this is the third HL daily-Donchian major (after BTC and ETH) to die on the identical all-zero-OOS / negative-holdout / DSR-0 signature, confirming the limitation is the venue's short data, not the choice of instrument.

Despite passing sensitivity, the analyst abandoned it (iteration 1 of 2) because every generalization gate failed: deflated Sharpe 0.0 vs an expected-max-luck bar of 11.71, PBO 0.59, walk-forward OOS of [0,0,0] against in-sample [12.24,10.05,10.70] with is_overfitted=TRUE, and a negative holdout (-2.60).

A long/flat, single-instrument, OHLCV-only daily Donchian-channel breakout on SOLUSD.HYPERLIQUID (ATR/percentage stops, SMA regime filter, leverage 1.0, one fixed-fractional entry), applying the validated daily-trend edge to the highest-beta HL major to complete the clean-sizing HL trend sleeve across BTC/ETH/SOL.

The default backtest looked outstanding — +247% total return, Sharpe 1.77, profit factor 2.0, ~4:1 win/loss asymmetry over 15 trades at 41% max drawdown — but the entire result rested on a single Q4-2023→Jan-2024 trade (+222%, return_skew 8.43, kurtosis 75.7); optimization improved the headline to Sharpe 2.91, PF 3.73, +183% and passed sensitivity (0 cliffs), with the sizing probe clean throughout (avg_position_pct 95.8%/53.2%, metrics_reliable=true).