BtcFourHourVolumeConfirmedDonchianBreakoutAntiChopLong

BTC 4H Volume-Confirmed Donchian Breakout with Anti-Chop Directional-Efficiency Filter Long

A long-only, single-instrument BTC strategy on BTCUSDT.BINANCE 4H bars that combines the analyst-validated Volume-Confirmed Donchian Breakout template (which scored Sharpe 3.63 + max DD 2.94% + 6/7 years positive on ADA but was killed by May-2026-window holdout noise) with an explicit ANTI-CHOP REGIME FILTER — the analyst's exact prescriptive recommendation from the abandon feedback: 'rebuilt with a defensive regime-detection filter that auto-disables entries during 2026-style chop regimes.' The anti-chop mechanism uses directional efficiency ratio = abs(net price change over N bars) / sum(absolute bar-to-bar returns over N bars). A ratio close to 1.0 means price moves directionally (trending); close to 0.0 means price oscillates with little net progress (choppy). Entries are gated to require this ratio >= 0.5 over the last 30 bars (5 days on 4H) — disabling entries during the chop regimes that caused the ADA holdout failure. Critically distinct from BTCVolumeConfirmedDonchianCalmVolLong (which uses LOW realized vol — a different regime axis): a market can be CALM AND CHOPPY (low vol grinding sideways — bad for breakouts) or VOLATILE AND TRENDING (high vol with clear direction — good for breakouts). The directional-efficiency filter discriminates regardless of vol level, complementing rather than duplicating the calm-vol filter. Implements the analyst's exact remediation for the most recent abandon. Built on BTCUSDT.BINANCE 4H — confirmed working (BtcFourHourVolumeBreakoutLong + BTCFailedBreakdownReclaimLong + BTCVolumeConfirmedDonchianCalmVolLong all validate this bar type). Single-instrument, no cross-venue infrastructure risk.

Iteration-2 fix for a config-wiring defect, not a signal defect: the diagnostic header shows the backtest ran BTCUSD.HYPERLIQUID-1-MINUTE with parameters={} (the fourth strategy this session hit by the same leaked fixture), which compresses the 20-bar Donchian and 30-bar efficiency-ratio windows to ~50 minutes and removes the very chop regime the anti-chop filter targets. The strategy code already cleared Layers 1-2 and matches the validated 4H BTC breakout family, so it is left byte-identical. The only change is an explicit, fully-populated FactoryStrategyConfig: instrument_id='BTCUSDT.BINANCE', bar_type='BTCUSDT.BINANCE-4-HOUR-LAST-EXTERNAL', and every intended parameter (donchian_period, donchian_exit_period, vol_period, volume_mult, er_period, er_threshold, entry_threshold, hard_stop_pct, max_hold_bars, risk_pct, per_trade_pct, min_notional) wired into config.parameters so nothing falls back to a leaked default. On BTC's full 4H history the anti-chop ER gate can now be evaluated against real 2026-style chop windows. Venue stays BINANCE USD-M futures at leverage 1.0 (matches the family; leverage is used in the notional cap of position_size, so it is not a leverage_set_but_unused case even though it is set to 1.0). The persistent BTCUSD.HYPERLIQUID-1-MINUTE/parameters={} pattern across four strategies should be escalated as an infrastructure-level harness bug.

The anti-chop directional-efficiency filter — the explicit prescribed remediation for the ADA version's chop-regime holdout failure — failed at exactly its purpose. On optimization attempt 1 every forward gate fails: the untouched HOLDOUT is NEGATIVE (holdout_sharpe -2.055, ratio -2.242), i.e. the recent chop regime the filter was designed to survive still loses money, worse than the baseline it was meant to rescue. It fails deflated Sharpe decisively (DSR=0.0 vs 0.95, optimized Sharpe 1.33 far below the 225-trial expected-max luck bar of 4.74, PBO 0.627, sharpe_ci_low -0.042 straddling 0; PSR 0.987 vs DSR 0.0 is the classic multiple-testing trap). The walk-forward is is_overfitted=TRUE with avg IS 4.66 collapsing to avg OOS 0.92 and two of three OOS windows exactly 0.0 — the thin 31-trade base cannot fill the windows. Sensitivity FAILED with 2 cliffs (entry_threshold >1.0 → Sharpe 0.0). The attractive base Sharpe 2.56 rested on only 31 trades; the optimized full-sample refit dropped to 1.33 with negative 2025 (-2.9) and 2026 (-1.5). Not iterate: the anti-chop filter WAS the remediation and it did not work — the recent holdout is negative, and no parameter tuning fixes a negative forward holdout in the exact regime the filter targets on a sample too thin to walk-forward. Not revise_hypothesis: BTC 4H breakout is not a dead target and has promoted siblings; this is overfit-on-a-thin-sample plus recent-regime decay (overfit → abandon), not a proven mechanism stranded on a dead instrument. FAILURE PATTERN: bolting a directional-efficiency anti-chop gate onto the volume-Donchian breakout does NOT rescue it from chop-regime holdout failure — it thins the sample to 31 base trades (walk-forward windows go to zero), the best-of-225 optimized Sharpe (1.33) sits far below the 4.74 luck bar (DSR 0.0), and the untouched recent holdout is deeply negative (-2.055). A strong 31-trade base Sharpe and a novel regime filter do not survive deflation, an overfit walk-forward, and a negative forward holdout; the promoted 4H breakout siblings remain where this class clears the bar.

Long-only BTCUSDT.BINANCE 4H OHLCV breakout: enters when close breaks the prior 20-bar Donchian high, volume exceeds 1.5x the 20-bar average, AND the 30-bar Kaufman directional-efficiency ratio is >= 0.5 (an anti-chop regime gate that disables entries during oscillating sideways markets). Exits via a 10-bar Donchian trailing channel, a 5% hard stop, or a 60-bar (10-day) time stop. Risk-based sizing on BINANCE USD-M futures, leverage 1.0.

Faithful, direction-correct implementation of the analyst-prescribed anti-chop remediation: 31 longs / 0 shorts, clean plumbing, ER filter demonstrably throttling 2026 chop-regime entries.

Strong risk-adjusted metrics: Sharpe 2.56 with sharpe_ci_low 0.946 (>0), PF 3.80, max_drawdown 4.06%, Sortino 12.4, PSR 0.9999, positive alpha; positive in 4 of 7 years with only mild chop/bear-year negatives.

Trades spread across all 7 years (2020-2026), so walk-forward windows will each hold trades — a testable sample, unlike thin single-regime cases.

Thin-ish sample (31 trades) — the 15-day holdout may land on a sparse window, the exact failure mode that killed the ADA sibling; this is the primary risk to survive.

Outlier-dependent edge: return_kurtosis 25.6, skew 4.54, tail_ratio 4.24, avg_win (4227) >> avg_loss (1185) — a few big trend winners carry the result, which the deflated-Sharpe gate over 225 trials will stress.

Recent softening (2025 -0.2, 2026 -2.0) — mild, but the recent-window holdout remains the likely pressure point.

Faithful, direction-correct implementation of the prescribed anti-chop remediation; clean plumbing; the base backtest looked strong (Sharpe 2.56, DD 4.06%, positive most years).

Sensitivity is mostly flat away from two threshold cliffs, and the ER filter is a genuinely distinct regime axis.

Holdout FAILED and is NEGATIVE: holdout_sharpe -2.055 (ratio -2.242) — the anti-chop filter did NOT fix the recent chop regime it was explicitly built for; the untouched window loses money.

Fails deflated Sharpe: DSR=0.0 (<0.95), optimized Sharpe 1.33 far below the 225-trial expected-max luck bar of 4.74, PBO 0.627 (>0.5), sharpe_ci_low -0.042 (straddles 0); PSR 0.987 vs DSR 0.0 is the trap.

Walk-forward is_overfitted=TRUE: avg IS 4.66 collapses to avg OOS 0.92, with two of three OOS windows exactly 0.0 — thin 31-trade base produces empty/zero windows.

Sensitivity FAILED (2 cliffs: entry_threshold >1.0 → Sharpe 0.0, er_threshold); optimized full-sample refit fell to Sharpe 1.33 with negative 2025 (-2.9) and 2026 (-1.5).

The backtest ran on the WRONG data, so the anti-chop remediation was never evaluated. Fix the FactoryStrategyConfig wiring before re-testing: (1) Set instrument_id='BTCUSDT.BINANCE' and bar_type='BTCUSDT.BINANCE-4-HOUR-LAST-EXTERNAL' — the strategy is designed for BTC 4H (donchian_period=20 bars=80h, er_period=30 bars=5 days, max_hold_bars=60=10 days), but the backtest ran BTCUSD.HYPERLIQUID-1-MINUTE, degrading every window to minutes and producing impact_cost 111.8% of gross with only 7 trades in 4 days. (2) The config passed parameters={} — ensure the intended parameter block (donchian_period, donchian_exit_period, vol_period, volume_mult, er_period, er_threshold, hard_stop_pct, max_hold_bars, risk_pct, per_trade_pct) is wired into config.parameters, not left empty. (3) Ensure the run spans BTC's full 4H history so the anti-chop ER filter and the 2026-style chop regime it targets are actually present in the data. CRITICAL — SYSTEMIC INFRASTRUCTURE DEFECT: this is the FOURTH strategy this session (after BTC price-volume divergence, ADA 50-day breakout, and BNB 4H volume breakout) to run with the IDENTICAL wrong config (BTCUSD.HYPERLIQUID-1-MINUTE, parameters={}). This is 4-for-4 and is conclusively NOT a per-strategy coding error — the backtest harness/config fixture is leaking a hardcoded default instead of reading each strategy's own instrument_id/bar_type/parameters. Escalate this to fix the backtest wiring at the infrastructure level immediately; until it is fixed, EVERY strategy review will be meaningless. Once BTC 4H data runs correctly with the intended parameters, re-submit for review.

BtcFourHourVolumeConfirmedDonchianBreakoutAntiChopLong faithfully implemented the analyst's prescribed remediation, adding a directional-efficiency anti-chop gate to the volume-Donchian breakout that had scored well on ADA but died on a chop-regime holdout. Its base backtest was attractive — Sharpe 2.56, PF 3.80, 4.06% max drawdown, PSR 0.9999 — so it advanced to optimize, but the sample was only 31 trades and outlier-carried. Optimization exposed the failure at the filter's own purpose: the untouched recent holdout was deeply negative (-2.055), deflated Sharpe was 0.0 with the optimized 1.33 far below the 4.74 luck bar, the walk-forward was overfit with two of three OOS windows at zero, and sensitivity showed two cliffs. On iteration 2 the analyst abandoned it, concluding that the anti-chop filter did not fix the chop regime it targeted and that no tuning rescues a negative forward holdout on a sample too thin to walk-forward — the promoted 4H breakout siblings remain where this class clears the bar.

Bolting a novel anti-chop regime filter onto the volume-Donchian breakout did not rescue it — the filter failed at its exact purpose (the recent chop-regime holdout was still deeply negative) and thinned the sample to 31 base trades so the walk-forward windows collapsed to zero; a strong 31-trade base Sharpe cannot survive deflation, an overfit walk-forward, and a negative forward holdout.

It cleared the backtest-review gate to optimize but the analyst abandoned it on optimization attempt 1 for failing every forward gate: a negative untouched holdout (holdout_sharpe -2.055, ratio -2.242), deflated Sharpe 0.0 (optimized Sharpe 1.33 far below the 4.74 luck bar, PBO 0.627, sharpe_ci_low -0.042), walk-forward is_overfitted=TRUE (avg IS 4.66 → avg OOS 0.92 with two of three OOS windows exactly 0.0), and 2 sensitivity cliffs.

A long-only BTCUSDT.BINANCE 4H volume-confirmed Donchian breakout that added an anti-chop directional-efficiency (Kaufman ER ≥ 0.5 over 30 bars) gate — the analyst's exact prescribed remediation to auto-disable entries during the chop regimes that had killed the ADA sibling on holdout.

The base backtest looked strong on 31 long trades (0 shorts): Sharpe 2.56 (sharpe_ci_low 0.946), profit factor 3.80, max drawdown 4.06%, Sortino 12.4, PSR 0.9999, 51.6% win rate and positive alpha — but the edge was outlier-dependent (kurtosis 25.6, skew 4.54, avg win 4,227 vs avg loss 1,185) with recent softening (2025 -0.2, 2026 -2.0); the optimized refit rose to 69 trades but fell to Sharpe 1.33.