DonchianBreakoutNetDirectionalTrendBasketLS

Donchian Channel-Breakout Net-Directional Trend Basket Across Liquid Majors (Turtle-Style, Long-Short, Binance USD-M, Equal-Weight, Weekly Rebalance, Full-Flatten-and-Reopen, Pure OHLCV)

A LONG-SHORT, MULTI-INSTRUMENT, NET-DIRECTIONAL trend basket across ~10 confirmed-full-history Binance USD-M majors (BTC, ETH, SOL, BNB, XRP, ADA, DOGE, AVAX, LINK, LTC) whose signal is the classic DONCHIAN / TURTLE CHANNEL BREAKOUT -- a PARALLEL classical trend system to the promoted equal-weight Absolute-Momentum basket (holdout Sharpe 2.17), NOT a refinement of it. KEY DISTINCTION: the promoted basket trades trend STATE (price vs SMA -- profitable in trend MIDDLES); this trades trend INITIATION (breaking an N-day price range -- profitable at trend STARTS). This is the single most important diversification axis in managed futures: real CTA funds run BOTH moving-average and breakout systems precisely because their entry timing DECORRELATES, so this is portfolio-ADDITIVE rather than a weighting/signal tweak of the same SMA signal. It avoids every closed/prohibited path: NOT single-asset (alt-TSM and BTC-trend are dead -- single legs, including the single-asset ETH Donchian that collapsed, do not generalize; the DIVERSIFIED basket is the validated structure); NOT a WEIGHTING refinement (inverse-vol falsified -- this stays EQUAL-WEIGHT, the promoted weighting); NOT reversion/microstructure/options/funding/seasonality (all dead/walled). MECHANISM (per leg, classic dual-channel Turtle state): a leg is LONG when its most-recent entry_channel-day HIGH breakout is more recent than its most-recent exit_channel-day LOW breakout, and SHORT in the mirror case -- i.e. each leg holds the direction of the last range breakout until the opposite (shorter) channel breaks. This is dense (every leg is always positioned), net-directional (signed sum of legs), and the canonical breakout trend-follower. EVERY design choice respects the hard engine walls: (1) FULL-FLATTEN-AND-REOPEN sizing every weekly rebalance (close ALL legs, recompute states, reopen equal-weight summing to <=100% gross) -- the PROVEN pattern that bounded the promoted basket at ~24%, NOT the long-only held-winners pattern that proved uncappable. (2) EQUAL-WEIGHT (the validated, promoted weighting), NOT inverse-vol (falsified). (3) NET-DIRECTIONAL. (4) PURE OHLCV (no ticks/2-day-wall, options/unwired, funding/data-ends-2024). (5) WEEKLY rebalance -> trade count and fee drag near the promoted basket's, not the inverse-vol basket's 2860-trade bleed. Per the 'favor simpler' guidance, 3 tunables (entry_channel, exit_channel, rebalance_period).

Implements a Donchian/Turtle channel-breakout basket as a PARALLEL classical trend system to the promoted equal-weight Absolute-Momentum basket (holdout Sharpe 2.17), not a refinement of it: the promoted basket trades trend STATE (price vs SMA, profitable mid-trend) while this trades trend INITIATION (breaking an N-day range, profitable at trend starts), the canonical decorrelated entry-timing axis that real CTA funds run alongside MA systems -- so it is portfolio-additive. The per-leg signal is the classic dual-channel state computed by scanning each leg's history for the four breakout event indices (entry-channel high/low and exit-channel high/low) and taking long when the last entry-high break is more recent than the last exit-low break, short in the mirror, defaulting to the most-recent entry breakout direction so each leg is always positioned (dense, net-directional). It honors every validated constraint: EQUAL-WEIGHT (the promoted weighting; inverse-vol was falsified) with each active leg sized sign * equity*gross_exposure/N, so gross is bounded at <=gross_exposure (90%) and the book's net direction is the signed sum; exposure is the bounded full-flatten-and-reopen end-state via single netting delta orders per leg plus an aggregate gross-cap backstop (no held-winners balloon). All order management is centralized in calculate_signal, fired by the base only after the cross-leg alignment barrier confirms every same-daily-timeframe leg is contemporaneous; should_enter/should_exit/position_size are inert. The Turtle state is recomputed only on weekly rebalance bars (intra-week channel chatter never churns the book), keeping trade count and fee drag near the promoted basket's. min_bars_required = entry_channel+5 and a bar-count warmup gate the first rebalance until channels are computable. The tanh log-return proxy is a continuous per-bar signal so Layer-2 frozen-signal detection passes while trades stay schedule-gated. Three tunables (entry_channel, exit_channel, rebalance_period) keep the overfit surface small. All 10 constituents are 2020-listed liquid majors with full daily history (insufficient-history/price<=0 legs skipped that rebalance), so no leg risks data_unavailable; pure OHLCV avoids the tick/options/funding walls. leverage stays 1.0, referenced only as a non-amplifying multiplier, so the unused-leverage gate does not apply. If the multi-instrument daily-return aggregation artifact flagged elsewhere recurs, this strategy's own gross is correctly capped <=90%, so that is the harness accounting bug to escalate, not a strategy sizing error.

Failed deflated Sharpe and overfit selection. DSR=0.258 (vs 0.95 bar), is_significant=false, and the optimized Sharpe 0.654 is BELOW even the low best-of-225 expected-max luck bar of 0.8335 — the edge cannot clear the multiple-testing penalty. PBO=0.6415 (>0.5) says the parameter selection is more likely than not overfit, and the walk-forward confirms it: is_overfitted=TRUE, avg IS 0.932 collapsing to avg OOS 0.381 with one OOS window negative (-1.03), so the OOS average rests on a single strong window (+1.93). sharpe_ci_low=0.0132 pins the CI at 0. The strong holdout (ratio 1.979) does NOT override these — a passing holdout with PBO>0.5 and an overfit walk-forward is the family's recurring 'holdout-luck' signature, and the holdout window predates the regime that produced the negative OOS window. The optimizer again breached the design's own cap (best_params max_gross 1.42 > 1.0) to chase 2021 return (annual +229%), and the book is fat-tailed (kurtosis 198) with 77.5% open-MTM at backtest end. Not iterate (attempt 1 of 2): sensitivity is flat with 0 cliffs (optimizer on its plateau) yet the Sharpe sits below the luck bar and the walk-forward is overfit with a negative OOS window — there is no robust region to tune toward, and a second best-of-225 sweep re-fits 2021 and re-breaches the gross cap. Not revise_hypothesis: the net-directional momentum-basket STRUCTURE is already validated on the promoted Absolute Momentum sibling; the Donchian/Turtle breakout SIGNAL family has repeatedly failed forward (single-asset ETH Donchian collapsed, the BTC breakout-with-trend sibling abandoned on a negative holdout), so this is signal-variant selectivity, not a sound mechanism stranded on a dead target. FAILURE PATTERN: swapping the validated SMA-state signal for a Donchian breakout signal on the same 10-major net-directional basket does NOT survive deflation — the breakout entry-timing edge is genuine but modest and regime-concentrated (2021), so best-of-225 selection lands a Sharpe (0.654) below even a low luck bar (0.83) with PBO 0.64 and an is_overfitted walk-forward, while the optimizer levers gross above the design cap to recover in-sample return. A clean sensitivity grid and a high-ratio holdout do not rescue an edge that fails DSR with overfit selection; the promoted Absolute Momentum basket remains the configuration where this structure clears the bar.

Long-short net-directional Donchian/Turtle channel-breakout trend basket across 10 full-history Binance USD-M majors (BTC, ETH, SOL, BNB, XRP, ADA, DOGE, AVAX, LINK, LTC), daily, weekly rebalance. Each leg's direction is the classic dual-channel Turtle breakout state: long when its most-recent 20-day high breakout is more recent than its most-recent 10-day low breakout, short in the mirror case, defaulting to the direction of the most recent entry breakout (always positioned). Equal-weight (each active leg = gross_exposure/N notional, signed), so the book is net-directional with gross <=90%. Every rebalance moves each leg to its fresh signed target via a single netting delta order with an aggregate gross-cap backstop. Pure OHLCV, leverage 1.0. Three tunables: entry_channel, exit_channel, rebalance_period.

Real (if thin) edge: profit_factor 1.110 (>1.0), CAGR +24%, alpha +0.22 — above the sub-1.0 abandoned siblings (consensus 0.968, inverse-vol 1.0005, long-horizon 0.96)

Diversified net-directional basket = the VALIDATED structure (promoted absolute-momentum basket Sharpe 2.17); a legitimate parallel breakout-entry system, not the dead single-asset Donchian

Recent years positive (2025 +27%, 2026 +64%), so the chronological holdout has a genuine chance; low fee drag (1.47% of gross), sane per-leg sizing (avg_position 18.5%)

2021-concentrated: +366% (2021 mania) carries the entire result; only 3 of 7 years positive

Three-year losing streak 2022 -30% / 2023 -12% / 2024 -34% — the net-directional crash-protection thesis FAILED in the 2022 bear (breakout whipsaws, not protection)

Outlier-carried + MTM-inflated: return_kurtosis 375, +135.6% single day (2024-12-24), end_unrealized_pct 224.6% (headline mostly open mark-to-market)

Thin PF (1.110), modest capacity ($36M), Sharpe 0.398 with CI straddling 0 (sharpe_ci_low -0.31) — matches the abandoned single-asset breakout-family signature in basket form

3 of 7

majority

no single-leg domination

Clean sensitivity surface: 0 cliffs across all 7 parameters

Holdout passed with a high ratio (1.979, holdout_sharpe 0.753 vs WF-OOS 0.381)

Optimization improved depth/efficiency: max_drawdown 57.2% -> 30.9%, profit_factor 1.11 -> 1.44, Sortino 0.87 -> 1.48, recovery_factor 10.1, and cut turnover/fees

Genuinely differentiated signal (Donchian/Turtle breakout entry-timing) with a sound CTA-diversification rationale vs the SMA-state promoted basket

FAILS deflated Sharpe: DSR=0.258 (vs 0.95), is_significant=false; optimized Sharpe 0.654 is BELOW even the low expected-max luck bar 0.8335 over 225 trials

PBO=0.6415 (>0.5) — parameter selection is more likely than not overfit

Walk-forward is_overfitted=TRUE: avg IS 0.932 -> avg OOS 0.381 with one window negative (-1.03); the OOS average rests on a single strong window (+1.93)

sharpe_ci_low=0.0132 — 95% CI is pinned at 0, edge not distinguishable from no-skill

Optimizer breached the design's own gross cap (best_params max_gross 1.42 > 1.0); end_unrealized_pct 77.5%, return_kurtosis 197.8

Edge concentrated in 2021 (annual +229%) with a brutal 2024 base-config year (-34%) — regime-dependent

DonchianBreakoutNetDirectionalTrendBasketLS applied the classic Turtle dual-channel breakout to a 10-major equal-weight net-directional basket, pitched as a portfolio-additive CTA diversifier to the promoted SMA-momentum sibling. Its initial backtest showed a real but thin edge (+24% CAGR, profit factor 1.11, Sharpe 0.398), concentrated in 2021 with a three-year losing streak through 2022-2024, and the analyst gated it to optimization. Optimization sharpened the surface — Sharpe 0.654, profit factor 1.44, drawdown cut to 30.9%, a passing holdout — but the robustness suite told the real story: deflated Sharpe 0.258, PBO 0.64, an overfit walk-forward with a -1.03 OOS window, and an optimizer that levered gross past the design cap to re-fit 2021. After one iteration, the strategy was abandoned: the breakout signal family repeatedly fails forward, and the basket structure already clears the bar only on the promoted Absolute Momentum configuration.

Swapping a validated structure's signal (SMA-state) for a genuinely decorrelated but modest, regime-concentrated breakout signal does not survive the multiple-testing deflation penalty — a clean sensitivity grid and a high-ratio holdout cannot rescue an edge whose Sharpe falls below the luck bar with PBO>0.5 and a negative OOS window.

After the backtest-review 'optimize' verdict, the analyst abandoned it at the post-optimization ANALYZING stage: it failed deflated Sharpe (0.258 vs 0.95), is_significant=false, optimized Sharpe 0.654 sat below the best-of-225 luck bar of 0.8335, PBO 0.64 and an is_overfitted walk-forward confirmed the parameter selection was overfit, and the optimizer breached the design's own gross cap (1.42 > 1.0) to chase 2021 returns.

An equal-weight, net-directional long-short Donchian/Turtle channel-breakout trend basket across ~10 liquid Binance USD-M majors, intended as a CTA-style breakout (trend-initiation) diversifier to the already-promoted SMA-state Absolute Momentum basket.

The base backtest produced +24.1% CAGR and +464% total return over 816 trades but only a Sharpe of 0.398 (CI low -0.31), profit factor 1.11, 57.2% max drawdown, and just 3 of 7 positive years. Optimization improved the headline figures (Sharpe 0.654, profit factor 1.44, drawdown 30.9%, holdout ratio 1.979) but failed the robustness gates with deflated Sharpe 0.258, PBO 0.6415, and walk-forward avg OOS Sharpe 0.381 (one window -1.03).