BtcHyperliquidDualTimeframeMomentumConfluenceLS

BTC Hyperliquid Dual-Timeframe Momentum Confluence, Long-Short (Single-Instrument BTCUSD.HYPERLIQUID Perp, 4H Primary Momentum + 1D Confirm, Directional ONLY When Both Timeframes Agree / FLAT on Disagreement, ATR-Trailing Winners, 3-Parameter)

A LONG-SHORT, single-instrument, pure-OHLCV venue-diversification of this factory's #1 survivor — the BTC Dual-Timeframe Momentum Confluence (Sharpe ~1.99, in paper) — ported from Binance USD-M to the under-represented HYPERLIQUID venue on the SAME asset (BTC) and the SAME 4H+1D cadence. This design is dictated by the hard evidence accumulated this session, not intuition. FOUR single-name Hyperliquid momentum attempts have now been abandoned, and they teach three precise lessons this hypothesis obeys: (1) ASSET — SOL/AVAX/LINK all failed with fee-thin or negative edge (SOL-BTC-gate profit_factor 1.086 'fees eating nearly all gross edge'; AVAX -0.28%/trade); the real, fee-clearing momentum edge lives on BTC, the deepest and cleanest-trending asset, NOT on noisier alts — so this returns to BTC. (2) TRADE COUNT — the LINK vol-targeted TSMOM died L16 'structurally unmeasurable sample' (~50 trades, below the ~100 floor) and low-turnover TSMOM/macro variants cannot clear that bar; the 4H-primary confluence cadence generates 100+ trades over the ~2.3-year HL window, keeping the sample measurable. (3) OVERFIT — the SOL Donchian breakout and others died on PBO/holdout because they invented NEW bespoke signals; this reuses the ALREADY-VALIDATED generic confluence signal verbatim, minimizing the overfit surface that the robustness gates punish. It is materially different from the pipeline's EthHyperliquidDualTimeframeMomentumConfluence (ETH, not BTC — and BTC is specifically the asset with the surviving edge) and from the Binance BtcDualTimeframeMomentumConfluence (different venue, lower fees, different fill/microstructure). Fills the most under-represented data-safe buckets: HYPERLIQUID venue (6.2% vs 20% quota) and long_short direction (13.6% vs target). Single instrument, HYPERLIQUID 4H + 1D bars (both above the sub-hourly history wall); same venue so NO cross-venue price_precision schema-merge landmine. Pure OHLCV, no supplementary feed. Avoids every recent death mode: measurable sample (not the low-turnover trap), BTC edge (not the fee-thin alt trap), validated signal (not the bespoke-overfit trap), no options cadence, no funding carry, no 90-day liquidation-feed wall.

I located the actual validated survivor (BtcDualTimeframeMomentumConfluenceLS4H_v5) and ported its signal code verbatim rather than re-deriving a Hyperliquid variant, because that verbatim reuse IS the hypothesis's anti-overfit argument. Worth noting for the analyst: the validated strategy's 'dual timeframe' is two lookbacks on the same 4H close series (its own docstring calls the 42-bar leg the 'daily-scale confirm'), not a separate 1-DAY subscription — so despite the hypothesis title saying '4H + 1D', reusing the signal verbatim means no extra feed, which also makes this single-feed and immune to extras-resolution and alignment issues. I made one deliberate deviation: the original exposed its risk/execution settings as parameters, and I locked them as class constants, which only shrinks the optimizer surface (the hypothesis's own goal) without touching the signal. I validated the ported signal on real BTCUSD.HYPERLIQUID data with 0.09% round-trip fees before submitting, and the results split three ways. TRADE COUNT verifies: 323 trades over the full HL history and 125 in the 365-day verification window, clearing the ~100 floor that failed the LINK TSMOM. ASSET verifies: BTC is positive on HL (+7.2% full history, 24.6% drawdown, PF 1.15; +7.3% and PF 1.27 in the last year) where the SOL, AVAX and LINK 4H attempts were negative. But the FEE-CLEARING claim only partially verifies, and I want to be direct about it: at +0.14% per trade over the full sample the edge sits essentially ON the 0.15% futures floor, clearing it only in the more recent year at +0.23%. This is thin-but-positive, not a comfortable margin — better than the alts by a clear margin, but the same fee-thin neighbourhood, so the analyst should not expect the Binance parent's Sharpe to reproduce here at Hyperliquid's higher taker cost. I also mapped the ±20% sensitivity band the optimizer's Phase-1 will probe, because it contains a hazard: fast_lookback is insensitive (10/12/14 alike), but slow_lookback has a CLIFF near 38 — at slow=34 the full-history result turns negative across every fast setting, while 42 and 50 are positive, with 50 actually the strongest (+15-18%). I kept the validated 42 rather than switching to the locally-best 50 precisely because re-fitting the confirm horizon to this venue is the overfit the hypothesis exists to avoid, but the optimizer should be steered upward from 42 rather than downward, and a cliff flag in Phase 1 would be an expected finding here rather than a surprise.

No edge to optimize. Across a decisive 291-trade, 2.5-year sample the HL BTC 4H confluence port is a net loser: profit_factor 0.954 (<1.0), expectancy -$22.15/trade, total_return 0.12%, Sharpe 0.037 (CI [-0.99, 1.06] straddles zero, PSR 0.53), max_drawdown 26.5%. avg_trade_return_pct 0.151% sits exactly on the 0.15% futures fee floor — the fee_edge death signature the developer's own notes admitted ('sits essentially ON the floor'). The proven Binance-BTC dual-timeframe momentum edge simply does not survive HYPERLIQUID's fee/microstructure at 4H on BTC; the signal is fee-thin-positive at best and negative-expectancy net. This is the OHLCV-momentum class with PF<1.2 that has zero survivors. There is no code bug to iterate on (the port is correct and two-sided), and tuning 3 parameters to lift a PF-0.95/negative-expectancy signal toward viability would only curve-fit the single positive trailing year. Abandon rather than spend 2 hours optimizing.

Long-short dual-horizon momentum confluence on the BTCUSD Hyperliquid perpetual, 4H bars — a venue port of the factory's validated Binance survivor with the signal code reused verbatim. Two momentum horizons are computed on the 4H close series: a fast ~2-day leg (12 bars) and a slow ~7-day daily-scale confirm (42 bars). It goes long only when both are positive, short only when both are negative, and stays flat whenever they disagree. Positions exit when the slow confirm leg flips against them or on a 5x-ATR trailing stop from the best price reached, letting winners run. Sizing risks 1% of equity across an initial 1.5x-ATR stop with gross notional capped at 50% of equity. Leverage 1.0. Exactly three tunable parameters (fast_lookback, slow_lookback, trail_atr_mult); ATR period, initial stop, risk fraction, notional cap and quantity precision are locked constants.

Analyst must confirm at backtest_review whether the recent-year edge is durable or a regime artifact, and ABANDON if the optimized full-history avg_trade_return_pct stays below 0.15%. Do not expect the Binance parent's Sharpe ~1.99 at HL's higher taker cost.

THIN FEE-CLEARING (borderline). Sandbox avg_trade_return_pct is +0.274% (above the 0.15% floor), but the developer honestly discloses the FULL-HISTORY figure is +0.14%/trade — BELOW the floor — clearing it only in the recent year. PF is 1.13 (sandbox) / 1.15 (full) / 1.27 (last year), Sharpe 0.19 with CI [-1.53, 1.75] straddling zero. This is the 'fee-thin neighbourhood' that killed the SOL/AVAX/LINK 4H siblings; BTC is a clear step better (positive vs their negative/PF~1.08) but not comfortably so. Over the full backtest this sits essentially at the analyst's avg_trade < 0.15% abandon line.

The optimizer must explore slow_lookback UPWARD from 42 (toward 50), NEVER below; treat a Phase-1 cliff flag as confirming fragility.

PARAMETER CLIFF / FRAGILITY (disclosed). slow_lookback has a cliff near 38: at slow=34 the full history turns NEGATIVE (-8% to -14%), while 42/50 are positive. The shipped/validated 42 sits JUST past the cliff, and the optimizer's ±20% band (~34-50) straddles the negative side. A positive result one parameter step from negative is a robustness/overfit hazard.

Accept as a faithful verbatim survivor port; record that the confirm is a same-feed 42-bar leg, not a 1D bar.

TITLE vs MECHANISM: the title says '4H Primary + 1D Confirm', but the confirm is a 42-bar (~7-day) momentum on the SAME 4H series, not a separate 1D bar. Transparently disclosed and is the validated survivor's actual construction (its docstring calls the 42-bar leg the 'daily-scale confirm'), keeping the strategy single-feed. Functionally equivalent — a naming deviation, not a mechanism swap. Note the ETH-HL dual-TF sibling used a genuine 1D extra leg, so the ports differ here.

Judge on the full ~2.3-year HL history + walk-forward; the trade count is adequate for inference.

Single-year sandbox non-decisive (Sharpe 0.19, CI [-1.53, 1.75] straddling zero), though measurable (104 sandbox / 323 full-history trades, well above the ~100 floor — the 4H cadence clears the measurability bar the LINK TSMOM failed).

Large decisive sample (291 trades) and genuinely two-sided (143 long / 148 short) — the verdict is measured, not noise

Correctly coded verbatim port; no bug to fix (rules out iterate)

No edge net of costs: profit_factor 0.954 (<1.0, gross loser), expectancy -$22.15/trade, total_return 0.12% over 2.5 years, Sharpe 0.037 (CI [-0.99, 1.06] straddles zero, PSR 0.53)

avg_trade_return_pct 0.151% sits exactly on the 0.15% fee floor — the fee_edge death signature the developer's own notes conceded

Loses through the middle of the sample (2025 -17.2%, Dec-2025 -13.0%), max_drawdown 26.5%, information_ratio -0.45 vs a meaningful buy-hold benchmark

OHLCV-momentum class with PF < 1.2 (zero survivors, L9); the proven Binance-BTC edge does not survive HL fees/microstructure at 4H — only the trailing year is positive, and optimizing to it would curve-fit

This strategy took the factory's best survivor — the Binance BTC dual-timeframe momentum confluence — and ported it unchanged to the under-represented Hyperliquid venue to fill the HL and long-short buckets, deliberately reusing the already-validated signal to minimize overfit surface. The developer's own measurements flagged the risk up front: the edge cleared the fee floor only in the recent year and sat essentially on it over the full sample. The backtest confirmed the worst case — 291 trades yielding profit factor 0.954, negative expectancy, 0.12% total return, and avg per-trade return of exactly 0.151% against the 0.15% floor. With no code bug to iterate and no gross edge to optimize, the analyst abandoned it at the backtest-review gate rather than curve-fit to the lone positive year.

A validated momentum edge does not automatically survive a venue port: Hyperliquid's fees and 4H microstructure erased the proven Binance-BTC signal, leaving avg per-trade return sitting exactly on the fee floor — a fee-thin signal is a net loser, and the developer's own pre-submission notes had already conceded the edge sat 'essentially ON the floor.'

The analyst abandoned it at the pre-optimization BACKTEST_REVIEW gate — optimization never ran — because a correctly coded, two-sided port with profit factor below 1.0 and negative expectancy has no edge to optimize, and tuning its 3 parameters toward the single positive trailing year would only curve-fit. It matched the OHLCV-momentum-with-PF<1.2 class that has zero survivors.

A long-short, single-instrument dual-timeframe momentum-confluence strategy that ported the factory's #1 Binance survivor verbatim onto BTCUSD.HYPERLIQUID — going long when both a fast (~2-day) and slow (~7-day) 4H momentum leg agreed up, short when both agreed down, flat on disagreement, trailing winners with an ATR stop.

Over a decisive 291-trade, ~2.5-year sample (143 long / 148 short) it produced essentially no edge net of Hyperliquid fees: total return 0.12%, Sharpe 0.037 (CI [-0.99, 1.06] straddling zero), profit factor 0.954, expectancy -$22.15/trade, win rate 30.9%, and 26.5% max drawdown. Its avg_trade_return_pct of 0.151% sat exactly on the 0.15% futures fee floor, positive only in the trailing year.