BtcHyperliquidImpulseBarContinuationTrendLS1H

BTC Hyperliquid Large-Impulse-Bar Continuation, Long-Short (Single-Instrument Perp, 1H, Trend-Aligned Momentum, Tight ATR Stop, Trailing-Stop Winners, Pure OHLCV)

A LONG-SHORT, SINGLE-INSTRUMENT, pure-OHLCV momentum-continuation strategy on BTCUSD.HYPERLIQUID perpetual futures (1-HOUR bars). It deploys the factory's single strongest VALIDATED edge — the magnitude-outlier impulse-bar continuation (proven on ETH-1H promoted, ETH-4H/SOL-1H/SOL-4H in pipeline/paper) — onto the highest-priority UNDER-represented venue (HYPERLIQUID is 8.1% of the book vs a >=20% target, and live trading is migrating there), with the most under-represented DIRECTION (long-short; the book is 86.5% long-only vs a <=55% target) at a non-over-represented HORIZON (1H). This is deliberately NOT a basket and NOT a port of a whipsaw-axis name: the analyst meta-notes establish that the impulse-continuation edge lives in SINGLE-ASSET expressions on the clean-continuation names (BTC/ETH/SOL) and is diluted to break-even by equal-weight majors baskets (1H and 4H baskets both abandoned). BTC is one of the three validated clean-continuation names and currently has NO single-asset impulse expression anywhere in the portfolio (only a BTC time-series-momentum daily sibling exists), and it has never been run on Hyperliquid. This fills the venue gap and the direction gap simultaneously while staying inside the one mechanism the factory has repeatedly confirmed carries a real, fee-clearing edge. It explicitly avoids every recent failure mode: not funding-carry (dispersion decayed, regime-dormant), not options (engine/structure limits), not opening-range (equity effect that doesn't transfer to 24h crypto), not a diversified basket.

Implements the hypothesis by porting the factory's single strongest VALIDATED edge — the magnitude-outlier impulse-bar continuation (promoted on ETH-1H, in pipeline/paper on ETH-4H/SOL-1H/SOL-4H/DOT-1H) — verbatim onto BTCUSD.HYPERLIQUID, swapping only instrument, precision (BTC 0.001 step), and the Hyperliquid $10 min notional. It fills two gaps at once: the under-represented HYPERLIQUID venue (live trading is migrating there) and the under-represented long-short direction, at the non-over-represented 1H horizon. It is deliberately a SINGLE-ASSET expression on a validated clean-continuation name (BTC), not a majors basket (which the analyst meta-notes show dilutes this edge to break-even) and not a whipsaw-axis name. The impulse threshold (>= ~1.5 ATR body with a strong-close filter) ensures each thrust is large enough to clear Hyperliquid's ~0.09% round-trip taker cost, and the ATR trailing stop lets continuation winners run while the tight initial stop bounds per-trade risk to ~1% of equity. Long+short symmetry trades both up and down impulses. Overfit containment mirrors the SOL/DOT siblings: impulse_mult hard-clamped to [1.0,1.8] and strong_frac<=0.70 keep every walk-forward window populated, and risk/sizing constants are frozen to shrink the optimization surface. leverage=1.0 and consumed in the notional cap, so it is not an unused leverage flag. Note: Hyperliquid serves ~5000 1H candles (~7 months), which is sufficient for this 1H strategy's signal density while respecting the venue's hard history cap.

Failed deflated Sharpe: DSR=0.0392, expected-max=3.28 over 225 trials — the optimized Sharpe of 0.97 is far below the best-of-N luck bar and statistically indistinguishable from selection noise (is_significant=false, PBO=0.7566>0.5, sharpe_ci_low -2.24 straddles 0). Every decisive gate fails in concert: walk-forward is_overfitted=TRUE with avg IS +2.11 collapsing to avg OOS -1.58 and a catastrophic middle OOS window (windows [2.79, -10.86, 3.32]); the holdout failed catastrophically (holdout_sharpe -10.16, ratio 0); and sensitivity failed with 2 cliffs (atr_period and impulse_mult). As with the AVAX sibling, the optimizer drove impulse_mult to 2.12 (clamped to the code's 1.8 ceiling), parking the entire edge on the cliff's high-threshold edge — the pre-emptive [1.0,1.8] clamp did NOT prevent the optimizer from concentrating the edge on its high-threshold cliff. The favorable-looking initial backtest (Sharpe 0.85, +6.3% return, 3.7% DD, balanced 58L/58S) measures consistency on a short 171-day single-regime sample, not multiple-testing significance, and is swamped by the all-failing post-selection gates. Not iterate: this is attempt 3 of 2 (over the limit), there is no robust parameter region to tune toward (the 225-trial sweep already produced negative avg OOS, a deeply negative holdout, and a cliff on the already-clamped core edge param), and narrowing the sweep cannot lift a sub-luck-bar Sharpe above DSR 0.95. Not revise_hypothesis: the failure is multiple-testing/cliff overfit with OOS collapse (is_overfitted=TRUE), not structural decay of a proven mechanism on a dead target — the impulse-continuation edge is already captured on ETH (promoted) and SOL, and BTC was the hypothesis's own chosen validated-name target, so re-pointing the same mechanism elsewhere would just re-overfit the same cliff. FAILURE PATTERN: single-asset 1H impulse-bar continuation ported to a new major/venue (BTC on Hyperliquid) produces a modest full-period Sharpe that fails DSR (0.04, far below the expected-max luck bar), with PBO 0.76, is_overfitted=TRUE (OOS avg -1.58, one OOS window -10.86), a catastrophic holdout (-10.16), and a 2-cliff sensitivity centered on the clamped impulse_mult edge param — the impulse-continuation family now fails on BTC and AVAX while working only on ETH/SOL, and clamping impulse_mult does not stop the optimizer from concentrating the edge on its high-threshold cliff.

Single-instrument long/short impulse-bar continuation on BTCUSD.HYPERLIQUID perp, 1-HOUR bars, pure OHLCV. Detects a magnitude-outlier directional thrust (body >= impulse_mult * Wilder-ATR AND a strong close location in the bar range), then enters WITH the impulse to capture short-horizon momentum continuation. Risk is cut fast and winners run: a tight init_stop_mult*ATR initial stop (which also sizes the position via fixed fractional risk), a trail_mult*ATR ratcheting trailing stop, and a 48-bar time cap. ATR-fractional-risk sizing off realized equity, hard-capped at 25% notional, one-position-at-a-time (flat between signals, no reversals/resizing). Only four params are optimizable (impulse_mult clamped to [1.0,1.8], trail_mult, init_stop_mult, atr_period); strong_frac/risk_pct/notional caps are frozen to contain overfit.

116 trades over 171 days — sufficient sample for meaningful parameter optimization, not degenerate

Genuinely balanced long-short (58/58), implementing the hypothesis's under-represented direction on the under-represented Hyperliquid venue

Trades match the stated mechanism: impulse-bar continuation entries, all 116 signals submitted with 0 dropped

Healthy risk profile: max_drawdown 3.75%, Sortino 1.84, recovery_factor 1.58, profit_factor 1.28, positive expectancy ($44.87/trade)

Win rate 32.8% with 2.6:1 win/loss and tail_ratio 2.42 — coherent with the trailing-stop 'let winners run' design

Base Sharpe only 0.85 (CI low -1.66 straddles 0) — right in the zone where best-of-225-trial selection commonly fails deflated Sharpe post-optimization

impact_cost_pct 18.3% and capacity_usd only ~$2.98M — edge is fragile to size/liquidity

Same impulse-continuation mechanism already FAILED DSR on its Binance BTC sibling and on AVAX; family is proven on ETH/SOL but weak/regime-dependent on BTC

Monthly returns are lumpy (Mar +3.99%, Apr -2.03%, Dec -1.26%) — gains concentrated in a few impulse-rich months, a multiple-testing risk

Mechanism is implemented cleanly and is genuinely market-neutral (beta -0.034, alpha +0.059, benchmark_correlation -0.21), filling the stated venue/direction gap by design

Initial backtest looked superficially balanced (58 long / 58 short, avg_trade_return well above fee floor, max_drawdown 3.7%, capacity ~$3M)

No degenerate execution: 116 entries submitted, 0 dropped on size/min-notional, metrics_reliable=true

Failed deflated Sharpe: DSR=0.0392 (<<0.95), is_significant=false, PBO=0.7566 (>0.5), and optimized Sharpe 0.97 is far below the best-of-N luck bar expected_max=3.28 over 225 trials — indistinguishable from selection noise

Walk-forward is_overfitted=TRUE: avg IS +2.11 collapses to avg OOS -1.58 with windows [2.79, -10.86, 3.32] — a single catastrophic OOS window

Holdout failed catastrophically: holdout_sharpe -10.16, ratio 0, passed=false

Sensitivity failed with 2 cliffs (atr_period, impulse_mult); the optimizer requested impulse_mult=2.12 (clamped to 1.8), again parking the edge on the high-threshold cliff — the exact AVAX-sibling failure the clamp was supposed to prevent

sharpe_ci_low -2.24 (initial) / -2.24 (optimized) straddle 0; high impact_cost_pct (~15-18% of gross) and commission ~9-11% of gross make the thin edge fee/impact-fragile

Over the iteration limit: attempt 3 of 2

BtcHyperliquidImpulseBarContinuationTrendLS1H ported the promoted ETH/SOL impulse-continuation trigger to BTC on Hyperliquid, filling the under-represented venue and long-short direction with the SOL/DOT clamp hardening baked in. The baseline was clean and market-neutral (116 balanced trades, Sharpe 0.85, 3.7% drawdown) and earned a full optimization, but that exposed an overfit: optimized Sharpe 0.97 below the 3.28 luck bar, DSR 0.039, PBO 0.76, a -10.86 OOS window, a -10.16 holdout, and two sensitivity cliffs with the optimizer again fleeing to the clamped impulse_mult ceiling. The analyst ruled it the same multiple-testing/cliff failure as the AVAX sibling — the family works only on ETH/SOL, BTC was the hypothesis's own chosen target, and it was over the attempt limit. It ended as abandoned, reaching optimization and analysis but never risk review.

Clamping impulse_mult to [1.0, 1.8] does not stop the optimizer from concentrating the edge on its high-threshold cliff: the impulse-continuation family now fails on BTC (Binance and Hyperliquid) and AVAX while working only on ETH/SOL, so porting it to a new BTC venue on a short single-regime sample produces a modest in-sample Sharpe that collapses out-of-sample (DSR 0.04, holdout -10.16) — BTC is not a clean-continuation name for this mechanism, and a favorable 171-day backtest measures consistency, not significance.

It passed the pre-optimization backtest-review gate (verdict: optimize) but was abandoned at the post-optimization analyst gate (verdict: abandon) as a clean multiple-testing/cliff overfit — the optimized Sharpe is indistinguishable from selection noise, every OOS gate and the holdout collapsed, and the optimizer drove impulse_mult to 2.12 (clamped to 1.8), parking the edge on the high-threshold cliff exactly as the AVAX sibling did — and it was over the 2-attempt limit, so it never advanced to risk review.

A single-instrument, pure-OHLCV long/short impulse-bar continuation strategy on BTCUSD.HYPERLIQUID 1H perps — entering with high-conviction directional impulse bars (body ≥ impulse_mult×ATR closing in the strong third of range) and riding continuation via a tight initial stop and ATR trailing stop — deploying the factory's validated impulse-continuation edge on BTC (a clean-continuation name with no single-asset impulse expression) onto the under-represented Hyperliquid venue and long-short direction, with the SOL/DOT clamp/freeze hardening (impulse_mult clamped to [1.0, 1.8]).

The baseline over ~171 days and 116 balanced trades (58 long / 58 short) looked acceptable enough to advance — Sharpe 0.85 (CI [-1.66, 2.94]), profit factor 1.28, expectancy $44.9/trade, total return +6.3%, max drawdown 3.7%, Sortino 1.84, genuinely market-neutral (beta -0.034) — but optimization failed every robustness gate: optimized Sharpe 0.97 vs a 3.28 expected-max luck bar, DSR 0.039, PBO 0.76, walk-forward is_overfitted=true (avg IS +2.11 → avg OOS -1.58, windows [2.79, -10.86, 3.32]), a catastrophic holdout (Sharpe -10.16, ratio 0), 2 sensitivity cliffs, plus 18% impact cost on only ~$3M capacity.