HyperliquidBtcImpulseBarContinuationLS1H

Hyperliquid BTC Impulse-Bar Continuation, Long-Short (Single-Instrument Perp, 1H, Pure OHLCV, ATR-Relative Directional Impulse Entry, Tight ATR Stop, Chandelier-Trailed Winners, Time-Stop, Low-Parameter)

A LONG-SHORT, SINGLE-INSTRUMENT, pure-OHLCV intraday MOMENTUM-CONTINUATION strategy on BTCUSD.HYPERLIQUID perpetual futures (1-HOUR bars). It transplants the factory's SINGLE repeatedly-VALIDATED edge — directional impulse-bar continuation (the SOL sibling reached paper_stage at Sharpe ~0.69; the ETH/SOL/BTC time-series-momentum siblings passed) — onto the most under-represented venue in the portfolio: HYPERLIQUID (5.7% vs the >=20% live-trading quota, and where the factory is shifting live execution). It is a FRESH framing within that validated family, not a duplicate: the impulse-continuation mechanism has been proven on SOL and ETH but NOT on BTC (BTC currently only carries a 1D time-series-momentum strategy, not a 1H impulse-continuation one), and never on the HYPERLIQUID venue. This deliberately AVOIDS the premise that just failed on HL: the abandoned ETH opening-range-breakout died because equity-market session-handoff structure (Asia/EU range -> US-session resolution) does not transfer to crypto's 24h tape. Impulse continuation makes no time-of-day assumption — it reacts to information/liquidity shocks whenever they occur. The impulse threshold is ATR-RELATIVE so the identical logic self-calibrates to BTC's lower absolute volatility versus SOL. Pure OHLCV means zero supplementary-data dependency (sidestepping the funding/option-settlement data failures), BTC/HL has clean multi-year 1H history (dodging the earlier HL-alt data-availability failures), and the design is held to 4 core parameters to stay inside the factory's validated, non-overfit envelope.

Transplants the factory single repeatedly-validated edge -- directional impulse-bar continuation (the SOL sibling reached paper_stage at Sharpe ~0.69; ETH/SOL/BTC time-series-momentum siblings passed) -- onto the most under-represented venue (HYPERLIQUID, 5.7% vs >=20% live quota, where live execution is shifting). It is a fresh framing within that validated family, not a duplicate: impulse-continuation is proven on SOL/ETH but NOT on BTC at 1H (BTC only carries a 1D TSM strategy) and never on HYPERLIQUID. It deliberately avoids the premise that just failed on HL: the abandoned ETH opening-range-breakout died because equity-market session-handoff structure does not transfer to crypto 24h tape -- impulse continuation makes NO time-of-day assumption. The ATR-relative impulse threshold self-calibrates to BTC volatility; pure OHLCV means zero supplementary-data dependency (sidestepping funding/option-settlement data failures), and BTC/HL has clean multi-year 1H history (dodging HL-alt data-availability failures). Non-inverted geometry (chandelier-trailed winners with a tight initial stop) lets winners run while capping losers; the time-stop cuts dead trades. HYPERLIQUID futures genuinely uses long+short. leverage=1.0 and sizing uses risk_frac/notional cap not leverage, so no unused-leverage gate triggers. 4 core parameters keep it inside the non-overfit envelope.

Not worth the 2 hours: a structurally data-limited, thin, regime-dependent edge that matches a failure pattern already confirmed this session. First, I verified via the equity curve ($100k->$105.5k, trough $99.1k) that the metrics are PERCENT-scaled and sane (+5.5%/~7mo, max_dd 7.2%, vol 7.3%) and sizing is correct (30%-notional cap holds) — so there is NO code bug to iterate on. The disqualifiers are structural and optimization-unfixable: (1) DATA SPAN ~7 months (5,316 1H bars, Oct 2025-May 2026), contradicting the hypothesis's 'multi-year 1H history' premise — a 3-window walk-forward + 15-day holdout on 7 months of intraday data validates one micro-regime, not robustness, and the optimizer cannot manufacture history. (2) THIN, UNSTABLE EDGE: Sharpe 0.45, PF 1.16, rolling Sharpe swinging +5.5 to -8.1 — the impulse-continuation family's documented BTC regime-dependence. (3) IMPACT 32.6% of gross at $938k capacity — the net-of-impact edge is marginal and real only at toy scale. (4) KNOWN-FAILURE-PATTERN MATCH: the identical BTC 1H impulse-bar continuation mechanism was abandoned earlier this session (DSR 0.0552, negative holdout -1.877, 4 cliffs); porting the same mechanism/asset onto a thinner-data venue does not change the edge, only weakens the validation. If the impulse-continuation family is to be pursued on Hyperliquid, it needs a venue/instrument with genuine multi-year 1H history AND where the mechanism is not already a proven DSR failure on that asset — that is a Research Lead reframe, not an optimization of this BTC instance.

Long-short intraday momentum-CONTINUATION on BTCUSD.HYPERLIQUID perpetual (1-HOUR bars), pure OHLCV. When a bar prints an abnormally large directional thrust relative to recent volatility -- |close-open|/ATR(14) >= impulse_atr_mult AND volume >= vol_mult x 20-bar average (and below an exhaustion_atr_mult blow-off cap) -- the strategy enters in the thrust direction to ride the short-horizon continuation. calculate_signal returns a continuous signed impulse score = (close-open)/ATR that varies every bar; should_enter applies the ATR-relative + volume thresholds. Exit uses a tight initial ATR stop (stop_atr_mult x ATR) that ratchets into a CHANDELIER trailing stop (chandelier_atr_mult x ATR off the favorable extreme via max/min of the two levels), plus a time-stop of max_hold_bars. Sizing is risk-based (risk_frac of equity over the initial-stop distance) capped at max_notional_frac (30%) of equity. Single position, flat between signals. The ATR-relative threshold self-calibrates to BTC lower absolute volatility; no time-of-day assumption (reacts to information/liquidity shocks whenever they occur). 4 core parameters (impulse_atr_mult, stop_atr_mult, chandelier_atr_mult, max_hold_bars); ATR/volume windows and risk fraction frozen.

Mechanically correct and functioning: 149 trades (68 long / 81 short), position sizing respects the 30%-notional cap (verified on the equity curve, $100k->$105.5k), trend payoff ratio 2.3:1 (avg_win $535 / avg_loss $232), pure-OHLCV so no supplementary-data dependency.

Modest positive result (+5.5% / ~7mo, +6.4% CAGR) and fills the under-represented Hyperliquid venue bucket.

Data span is only ~7 months (5,316 1H bars, Oct 2025-May 2026), contradicting the hypothesis's 'multi-year 1H history' claim — too short for a credible 3-window walk-forward + holdout (windows ~7 weeks each, one micro-regime).

Thin edge: Sharpe 0.45, profit_factor 1.16, win_rate 33.6%, and a rolling Sharpe that swings wildly from +5.5 to -8.1 — regime-dependent, the impulse-continuation family's documented BTC failure mode.

Impact eats 32.6% of gross PnL at a capacity of only $938k — the net-of-impact edge is marginal and exists only at toy scale; turnover 148x compounds the cost drag.

Known-failure-pattern match: the identical BTC 1H impulse-bar continuation mechanism was abandoned earlier this session (DSR 0.055, negative holdout -1.88, 4 sensitivity cliffs); this is the same mechanism on the same asset, just on a thinner-data venue.

~7 months (5316 1H bars)

multi-year for robust 1H WF+holdout

Sharpe 0.45, PF 1.16

Sharpe >0.5 / PF >1.2

impact 32.6% of gross, capacity $938k

impact <25%, deployable capacity

HyperliquidBtcImpulseBarContinuationLS1H transplanted the SOL/ETH-validated impulse-continuation edge to BTC on Hyperliquid, an ATR-relative volume-confirmed thrust trade with chandelier trailing, aiming to fill the under-represented HL venue. The code was correct and the result modestly positive (+5.5%, Sharpe 0.45, 2.3:1 payoff), but the edge was thin and regime-dependent (rolling Sharpe +5.5 to -8.1), impact ate 32.6% of gross at $938k capacity, and the data span was only ~7 months despite the multi-year premise. The analyst ruled it structurally data-limited and a direct match to the BTC 1H impulse mechanism already abandoned this session (DSR 0.055, -1.88 holdout), not worth optimizing. It ended after one iteration as abandoned, never advancing to optimization or risk review.

Porting the validated impulse-continuation family to Hyperliquid BTC changes neither the edge nor its known failure: the mechanism is already a proven DSR failure on BTC, and the HL venue adds only ~7 months of history (too thin to validate) plus a capacity/impact wall — pursuing the family on HL needs an instrument with genuine multi-year 1H history where the mechanism isn't already refuted on that asset, a research reframe, not an optimization of this instance.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) on structural, optimization-unfixable disqualifiers — only ~7 months of history (contradicting the 'multi-year' premise, too short for a credible walk-forward/holdout), a thin regime-dependent edge, impact eating a third of gross at toy capacity, and a direct match to the identical BTC 1H impulse-continuation mechanism already abandoned this session (DSR 0.055, negative holdout) — so optimization and all later stages were never reached.

A single-instrument, pure-OHLCV long/short intraday impulse-bar continuation strategy on BTCUSD.HYPERLIQUID 1H perps — entering in the direction of a volume-confirmed, ATR-relative directional thrust (|close-open|/ATR ≥ threshold, skipping exhaustion blow-offs) with a tight ATR stop ratcheting into a chandelier trailing stop plus a time stop — transplanting the validated impulse-continuation family onto the under-represented Hyperliquid venue and onto BTC (where only a 1D TSM strategy existed), with 4 core parameters.

Over only ~157 days (Oct 2025-May 2026) and 149 trades (68 long / 81 short) it was modestly positive but thin: total return +5.5%, Sharpe 0.45 (CI [-2.10, 2.89]), profit factor 1.16, expectancy $25.3/trade, win rate 33.6% with a 2.3:1 avg_win/avg_loss payoff, but a rolling Sharpe swinging wildly from +5.5 to -8.1, impact cost at 32.6% of gross, and only $938k capacity.