EthFourHourImpulseBarContinuationTrendLS4H

ETH 4-Hour Impulse-Bar Continuation, Long-Short (Binance USD-M, Fee-Favorable Slower Cadence of the Promoted ETH-1H Expression, Clamped Impulse Threshold, Tight Initial Stop, Trailing-Stop Winners, Discrete Capital-Capped)

A LONG-SHORT, single-instrument, pure-OHLCV CONTINUATION strategy on ETHUSDT.BINANCE USD-M perpetual using 4-HOUR bars — a CADENCE extension of the factory's single strongest validated expression, the PROMOTED ETH-1H impulse-bar continuation (Sharpe ~1.40), onto the FEE-FAVORABLE slower timeframe. The family is now mapped to two failure axes: it is FEE-SENSITIVE (sub-1H died — BTC-15m and ETH-15m both fee-dominated, more fee-incidence per smaller move) and ASSET-SENSITIVE (works on high-vol clean-flow majors ETH/SOL; BTC is fee-bound-zero across every trigger/venue; LINK/XRP whipsaw). This hypothesis attacks the FEE axis from the unexplored favorable side: a 4H cadence makes each impulse move LARGER, clearing the ~0.10% round-trip fee with wide margin (the inverse of why sub-1H failed), while keeping the validated asset (ETH, the strongest member). It is genuinely additive — 4H impulse events are a distinct, larger-move trade population with low correlation to the promoted 1H stream — and it directly honors the analyst's directive to 'pursue the validated ETH expression.' It is NOT a coin-flip on an unproven alt (ETH is THE proven asset; the in-flight AVAX-1H already tests alt-generalization), NOT a BTC variant (dead all triggers/venues), NOT a funding/basis/carry/cross-venue strategy (the entire family is fee-dominated/event-sparse — confirmed across BTC/ETH/SOL), NOT a daily-momentum basket refinement (all correlated with promoted AbsoluteMomentum), NOT relative-momentum (decayed), NOT options (engine), NOT reversion (dead), NOT HL intraday (data wall). Single-venue (no cross-venue fee floor), pure OHLCV, deep ETH 4H history (validatable), engine-safe (discrete enter-once/exit-once, single position, hard-capped). It carries the SOL-sibling parameter hardening to pre-empt the LINK overfit trap: impulse_mult CLAMPED [1.0, 1.8], strong_frac FROZEN, <=4 tunables, flat-between-signals.

Iteration-2 fix for the Layer-2 failure 'TypeError: _true_range() takes 2 positional arguments but 3 were given' across all six synthetic scenarios. Root cause: _true_range was declared @staticmethod def _true_range(cur, prev) but called as self._true_range(cur, prev); in the Layer-2 synthetic proxy environment the staticmethod descriptor was not honored as a plain function, so self got prepended and the call passed 3 args (self, cur, prev) to a 2-parameter function. Smallest fix: drop the @staticmethod decorator and make it a normal instance method `def _true_range(self, cur, prev)`, so self._true_range(cur, prev) correctly binds (self, cur, prev) in both normal execution and the proxy. No other code, imports, signal logic, sizing, or parameters were changed — every earlier-passing check (Layer 1 static, the rest of Layer 2 construction) remains intact.

Failed deflated Sharpe: DSR=0.0652 (vs 0.95 bar), with optimized Sharpe 0.5845 BELOW the 225-trial expected-max luck bar 1.1141 (is_significant=false, sharpe_ci_low -0.1651 straddling 0, PBO 0.5106 >0.5) — the selected config is statistically indistinguishable from best-of-N noise. Two independent forward gates concur and are decisive: the time-ordered HOLDOUT FAILED and is NEGATIVE (holdout_sharpe -0.865, ratio -2.282 — the untouched recent window loses money), and the walk-forward is is_overfitted=TRUE with avg IS 1.154 collapsing ~3x to avg OOS 0.379 (windows [-0.70, 0.60, 1.24], one negative, the average carried by a single window). This is confirmed by genuine recent decay: optimized annual returns fall to 2025 -5.97% and 2026 -17.76%, with rolling Sharpe running to -10/-11 through mid-2026 — the impulse-continuation edge has stopped working forward on the 4H cadence. The clean sensitivity grid (0 cliffs) and PSR 0.9525 measure consistency, not significance or forward-viability, and cannot override a negative holdout + DSR 0.065 + is_overfitted + CI through 0. The 4H cadence also did not escape the fee axis it targeted (commission_pct_of_gross 5.35% + impact_cost_pct 11.68%, capacity ~$7.3M). Not iterate (attempt 1 of 2): sensitivity is already flat with 0 cliffs (optimizer on its plateau) and the best-of-225 Sharpe sits below the luck bar, and the binding failure is a NEGATIVE forward holdout driven by 2025-2026 regime decay — not an under-explored parameter region — so a second best-of-225 sweep re-selects another sub-significant, decaying config. Not revise_hypothesis: ETH intraday impulse continuation is the validated/PROMOTED sibling family (ETH-1H, Sharpe ~1.40), so this is the 4H-cadence variant on the same proven asset failing the multiple-testing gate (cadence-variant selectivity), not a sound mechanism stranded on a dead target. FAILURE PATTERN: extending a promoted single-asset intraday continuation trigger to a slower 'fee-favorable' cadence on the SAME asset does NOT inherit promotability — the larger-move 4H trade population is a distinct but modest edge that fails best-of-225 deflation (Sharpe 0.58 below the 1.11 luck bar, DSR 0.065, is_overfitted IS 1.15 -> OOS 0.38) and, critically, its untouched recent holdout is NEGATIVE (-0.865) as the edge decays through 2025-2026. This is the same signature as the abandoned ETH sustained-multi-bar-drive variant: a clean sensitivity grid and PSR~0.95 do not rescue a cadence/trigger variant whose forward holdout loses money. The promoted ETH-1H instance remains the configuration where this mechanism clears the bar.

Long-short single-bar impulse continuation on ETHUSDT.BINANCE USD-M perp, 4-HOUR bars, pure OHLCV. When a single 4H bar's true range is a magnitude outlier (TR/ATR >= clamped impulse_atr_mult in [1.0,1.8]), it enters WITH the impulse direction (signed body-in-ATRs sign), applies a tight initial ATR stop, then trails winners with a wider ATR trailing stop so avg_win > avg_loss. Flat between signals, discrete enter-once/exit-once, single position, hard capital cap. The 4H cadence makes each impulse move large enough that the ~0.10% round-trip fee is a small fraction of the move (the fee-favorable inverse of the failed sub-1H members), while keeping ETH, the proven clean-flow major.

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['risk_frac', 'atr_period', 'min_notional', 'stop_atr_mult', 'trail_atr_mult', 'impulse_atr_mult', 'max_notional_frac'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: TypeError: EthFourHourImpulseBarContinuationTrendLS4H._true_range() takes 2 positional arguments but 3 were given (bar timestamp: 1735690560000) - steady_downtrend: TypeError: EthFourHourImpulseBarContinuationTrendLS4H._true_range() takes 2 positional arguments but 3 were given (bar timestamp: 1735690560000) - flat_ranging: TypeError: EthFourHourImpulseBarContinuationTrendLS4H._true_range() takes 2 positional arguments but 3 were given (bar timestamp: 1735690560000) - volatility_spike: TypeError: EthFourHourImpulseBarContinuationTrendLS4H._true_range() takes 2 positional arguments but 3 were given (bar timestamp: 1735690560000) - zero_volume: TypeError: EthFourHourImpulseBarContinuationTrendLS4H._true_range() takes 2 positional arguments but 3 were given (bar timestamp: 1735690560000) - price_gap: TypeError: EthFourHourImpulseBarContinuationTrendLS4H._true_range() takes 2 positional arguments but 3 were given (bar timestamp: 1735690560000)

Trades match the hypothesis: 598 impulse-outlier continuation trades (272 long / 326 short), trailing winners, with the pre-registered geometry holding (avg_win $5,763 > avg_loss $2,627, ~2.2:1) — the 4H cadence clears the ~0.10% fee with margin

Regime-consistent, NOT single-outlier-carried: positive in 5 of 7 years (2020 +42%, 2021 +21%, 2022 +20%, 2024 +51%, 2026 +8%), only mild negatives in 2023/2025

Healthy risk profile and clean sizing: max_drawdown 21.8% (calmar 9.6), Sortino 1.50, avg_position_pct 39.9% (no exposure inflation), ample sample (598 trades / ~5.6 yr)

Member of the factory's strongest validated family (promoted ETH-1H impulse, Sharpe ~1.40) on the proven asset, single-venue pure OHLCV, low-parameter with the impulse threshold clamped vs the LINK overfit trap

Base Sharpe modest at 0.578 with CI [-0.11, 1.11] grazing 0 — the directional edge is real but not strong pre-optimization

return_kurtosis 103 / skew 8.0 — elevated fat-tail dependence; must be watched against deflated Sharpe / PBO at the post-opt stage

Cadence sibling of an already-promoted ETH-1H impulse strategy — correlation/redundancy must be verified at the portfolio stage even if it optimizes cleanly (max_consecutive_losses 12 is a real live-drawdown caveat)

Genuinely additive, engine-clean design: single-instrument ETH 4H, discrete enter-once/exit-once, capital-capped, clean entry diagnostics (595/598 entries, 0 dropped), moderate kurtosis (55) and avg_position_pct ~50% — no leverage/outlier blow-up

0 sensitivity cliffs across all 7 parameters (clamped impulse threshold held) — the SOL-sibling hardening worked as intended

Real long-short participation (270 long / 325 short) on the proven ETH asset

Failed deflated Sharpe: DSR=0.0652 with optimized Sharpe 0.5845 BELOW the 225-trial expected-max luck bar 1.1141 — indistinguishable from best-of-N noise (is_significant=false)

Holdout FAILED and is NEGATIVE: holdout_sharpe -0.865, ratio -2.282 — the untouched recent window loses money

Walk-forward is_overfitted=TRUE: avg IS 1.154 -> avg OOS 0.379 (~3x collapse), OOS windows [-0.70, 0.60, 1.24] with one negative

PBO 0.5106 (>0.5) and sharpe_ci_low -0.1651 (CI straddles 0); PSR 0.9525 / DSR 0.065 is the trial-count trap

Recent regime decay: optimized annual returns 2025 -5.97%, 2026 -17.76%, rolling Sharpe to -10/-11 through mid-2026 — the edge has stopped working forward

Fee-fragile at 4H after all: commission_pct_of_gross 5.35% + impact_cost_pct 11.68% and small capacity (~$7.3M optimized)

Built as a deliberate 4-hour cadence extension of the factory's single strongest promoted expression — the ETH-1H impulse-bar continuation — this strategy bet that larger 4H impulse moves would clear the ~0.10% round-trip fee that had killed the sub-1H members. Its backtest was encouraging: 598 trades, +234% return, a modest 21.8% drawdown, positive in 5 of 7 years and clean enough sizing to earn passage to optimization. But optimization exposed a modest, decaying edge: deflated Sharpe collapsed to 0.065 with the Sharpe below the luck bar, the walk-forward was overfit (IS 1.15 → OOS 0.38), and — most decisively — the untouched recent holdout was outright negative (-0.865) as the impulse edge stopped working through 2025-2026, while fees still ate 5.35% of gross. After two iterations the analyst abandoned it, concluding the promoted ETH-1H instance remains the only configuration where this mechanism clears the bar.

Extending a promoted single-asset intraday continuation trigger to a slower 'fee-favorable' cadence on the same asset does not inherit promotability — a clean sensitivity grid and high probabilistic Sharpe cannot rescue a cadence variant whose untouched recent holdout loses money and whose deflated Sharpe fails after best-of-N selection.

The analyst abandoned it at the post-optimization ANALYZING stage for failing deflated Sharpe (0.0652 vs the 0.95 bar) with the optimized Sharpe of 0.5845 below the best-of-225 luck bar of 1.1141, compounded by two decisive forward-gate failures — a negative time-ordered holdout (Sharpe -0.865, ratio -2.282) and an overfitted walk-forward (avg IS 1.154 collapsing to OOS 0.379) — reflecting genuine 2025-2026 regime decay, plus fee-fragility it was meant to escape (commission 5.35% + impact 11.68% of gross).

A long-short, single-instrument impulse-bar continuation strategy on ETHUSDT Binance USD-M perpetual using 4-hour bars — a fee-favorable slower-cadence extension of the factory's promoted ETH-1H impulse-continuation strategy (Sharpe ~1.40) — entering with magnitude-outlier ATR-impulse bars, using a tight initial stop and trailing winners, with a clamped impulse threshold ([1.0, 1.8]) to resist overfitting.

The base backtest returned +234.2% over 598 trades (272 long / 326 short) at Sharpe 0.578 with a 21.8% max drawdown, positive in 5 of 7 years and ~2.2:1 payoff (avg_win $5,763 vs avg_loss $2,627); optimization held return at +235.0% and Sharpe 0.5845 but max drawdown rose to 33.8% with deflated Sharpe 0.0652.