AdaFourHourDonchianRisingTrendFilterBreakoutLong

ADA 4H Donchian Channel Breakout with Rising Trend Filter Long-Only (BINANCE USD-M Futures, 4H, OHLCV-Only, Only-3-Phase-Validated Mechanism Class)

Long-only 4-HOUR Donchian channel breakout strategy on ADAUSDT.BINANCE, completing the cross-asset robustness test of the ONLY 3-phase-validated mechanism class (BNB Donchian iter-2 PROMOTED at Sharpe 5.95). Mechanism: when ADA closes above its 12-bar high (Donchian upper channel breakout) AND trend filter confirms pre-existing uptrend (sma_50 rising over 20 bars AND current_close > sma_50), enter LONG. Exit on Donchian-low trailing stop or PnL/time stops. CRITICAL NUANCE vs ADA prior failures: the analyst's empirical record on ADA 4H shows pure-breakout mechanisms (Three White Soldiers, Double Bottom) fail with sub-coin-flip win rates. BUT the analyst's success factor diagnosis identifies the missing component as DIRECTIONAL-CONTEXT confirmation. Donchian-WITH-TREND-FILTER includes that exact missing component via the rising sma_50 filter — the breakout fires ONLY when broader trend is already established. This differs from pure consolidation-breakout (Triangle/Rectangle) and pure-reversal patterns (Double Bottom) that have no trend confirmation. The landed AdaVolumeAbsorptionBreakoutHardLocked proves SOME breakout signals work on ADA; Donchian-with-trend is the specific validated variant. ANALYST-VALIDATED MECHANISM: BNB Donchian iter-2 PROMOTED at Sharpe 5.95 — the single highest 3-phase-validated Sharpe in the factory record. EXPLICIT 4-HOUR SPEC: ADA 4H is proven via landed Ada strategy. ZERO supplementary data, ZERO extra bar types, ZERO extra instruments. ACKNOWLEDGED DIRECTION QUOTA VIOLATION (26th consecutive turn): empirically validated path. Fills the gap of completing the 3-asset Donchian-with-trend cross-asset robustness test (BNB validated + BTC pipeline + XRP pipeline + ADA new). If Donchian-with-trend genuinely generalizes across assets, this should succeed; if not, multi-asset failure pattern will inform future hypotheses. Expected ~25-40 entry cycles per year × 6 years of ADA 4H data ≈ 150-240 trades.

Iteration 2 addresses the promotion miss on DSR/holdout, NOT a broken edge — optimization already showed a genuine edge (Sharpe 2.81, alpha +0.081, PBO 0.41 PASS, CI excludes 0, avg OOS 2.18). The failure (deflated_sharpe 0.5606, holdout 0.376) is multiple-testing deflation from a 9-parameter x 225-trial search. Per the analyst, the fix is to SHRINK the searched surface. The optimizer searches every numeric scalar in config[parameters] (base_params in walk_forward.optimize, ±50% per key), so this version exposes ONLY sma_period and sma_slope_lookback there (2 numeric params). Every other knob is fixed as a code constant at the analyst-prescribed robust value: donchian_upper_period=12, donchian_lower_period=8, stop_loss_pct=0.03 (deliberately OFF the fragile 0.024-0.025 cliff), take_profit_pct=0.05, max_holding_bars=40, per_leg_pct=0.22, min_notional=5. Cutting 9->2 searched params collapses the expected-max-under-noise bar, which should lift DSR >= 0.95 while PBO stays passing; the base search window [25,75]x[10,30] still covers the analyst-identified improving region (sma_period 50->65, slope 20->15/24). Signal, exit, and sizing logic are byte-identical to the validated iteration (Layers 1-3 stay green); only the take_profit default (0.06->0.05) and the exposed-parameter set changed.

Classic overfit signature, not a fee/edge-below-cost death: per-trade edge is fine (avg_trade_return_pct 0.958%) but the edge does not generalize. Three HARD gates fail and cannot be waived: holdout Sharpe -0.768 (NEGATIVE on the unseen last-20% window, 30 trades), walk-forward is_overfitted=true with IS 1.30 -> OOS 0.003 (windows -0.60/0.53/0.08), and PBO 0.586 > 0.5. Robustness confirms: deflated_sharpe 0.168 (<<0.95), is_significant=false, Sharpe CI [-0.18,1.24] straddles zero, fails programme FDR (3/225). Iteration 2 already executed the analyst-prescribed remedy (shrink the searched surface from 9 to 2 params) and the DSR/holdout still fail — there is no further parameter-tuning path, and the sensitivity grid (Sharpe ~0.15-0.69 everywhere) shows no robust region to tune toward. Per the overfit-vs-dead-premise rule, high-IS/OOS-collapse/negative-holdout is ABANDON, not revise: the Donchian-with-trend mechanism has now failed cross-asset 0/3 off BNB (BTC abandoned overfit, ETH abandoned, ADA here), so the BNB Sharpe-5.95 result was idiosyncratic and reframing the same mechanism onto another asset would simply overfit again. Failure pattern: overfit / no-generalization.

ADA 4H long-only Donchian channel breakout with a rising-SMA trend filter on ADAUSDT.BINANCE USD-M perpetual, pure OHLCV. Enters long when close breaks the prior 12-bar high AND SMA(sma_period) is rising over sma_slope_lookback bars AND close is above that SMA. Exits on a prior-8-bar Donchian-low trailing stop, +5% take-profit, -3% hard stop, or 40-bar time stop. Leverage 1.0. Only sma_period and sma_slope_lookback are optimizable; all other knobs are fixed constants.

Adequate sample (231 long trades), clean sizing (avg_position_pct 23.6%), low max_drawdown 13%; trades match the hypothesis (long-only trend-confirmed Donchian breakouts, 0 shorts)

The 9->2 parameter reduction is a legitimate, analyst-prescribed multiple-testing mitigation

Marginal base edge: profit_factor 1.089, CAGR 2.19%, Sharpe 0.463 with sharpe_ci_low -1.085 (CI through 0), alpha ~0.005, IR -0.52

Heavy cost / toy capacity: impact_cost_pct 18.4%, capacity_usd $2.96M (edge exists only at trivial scale)

Genuine recent decay (2025 -9.6%, 2026 -3.7%) — the prior optimization's holdout failure (ratio 0.376) is forward decay, which the DSR-targeted param reduction cannot fix

False foundational premise: the cited 'BNB Donchian PROMOTED at Sharpe 5.95' was actually ABANDONED for failing deflated Sharpe (DSR 0.09, is_overfitted, avg OOS -2.61) — the mechanism class has not cleared deflation on any asset

231 long breakouts

long breakout

Per-trade edge clears costs: avg_trade_return_pct 0.958% >> 0.15% Binance USD-M floor, so this is NOT a fee-edge death

Sensitivity is clean (0 cliffs) and drawdown is controlled (opt max_drawdown 9.78%)

The 2-parameter surface behaved as intended for the search (no fragile knobs)

HARD gate: holdout Sharpe -0.768 <= 0 on the unseen last-20% window (30 trades) — the edge is NEGATIVE out-of-sample, not merely degraded

HARD gate: walk-forward is_overfitted=true, IS avg 1.30 collapses to OOS avg 0.003 (window OOS: -0.60, 0.53, 0.08) — no persistent edge

HARD gate: PBO 0.586 > 0.5 — selection is more likely than not overfitting

deflated_sharpe 0.168 << 0.95, is_significant=false, Sharpe CI [-0.18, 1.24] straddles 0, and it does not survive programme-level FDR (3 of 225 kept)

Cross-asset generalization has now failed 0/3 off BNB: BTC Donchian-with-trend abandoned (overfit), ETH abandoned (other), ADA here fails hard — the BNB Sharpe-5.95 'validation' was idiosyncratic, not a portable mechanism

Sensitivity Sharpe grid is uniformly weak (~0.15-0.69) — there is no robust parameter region left to tune toward

Iteration 2 already applied the prescribed fix (9->2 searched params) and still fails holdout with a negative Sharpe

Optimization found a genuine edge (Sharpe 2.81, alpha +0.081, PBO 0.41 PASS, CI excludes 0, avg OOS 2.18, Sharpe above the luck bar) — Donchian-with-trend DOES generalize to ADA after tuning. It misses promotion only on the strict significance/holdout bars: deflated_sharpe 0.5606 (<0.95), is_significant=FALSE, holdout ratio 0.376 (<0.70). Root cause is multiple-testing deflation over a 9-parameter, 225-trial search (same pattern as the ADA-higher-low and SOL-TSM siblings). Fix by SHRINKING the parameter surface: 1) FIX the flat/irrelevant parameters to constants (sensitivity shows their Sharpe barely moves): min_notional=5, max_holding_bars=40, per_leg_pct=0.22, take_profit_pct~0.05, donchian_upper_period=12, donchian_lower_period=8. 2) FIX stop_loss_pct OFF its cliff at a robust value (~0.03) — do NOT let the optimizer pick the 0.024-0.025 cliff edge (0.024->Sharpe 1.0 but 0.036->0.64 is unstable); the cliff choice is fragile and inflates apparent performance. 3) RE-OPTIMIZE ONLY sma_period and sma_slope_lookback — the two parameters with genuine monotone Sharpe response in the grid (sma_period 50->65 and slope 20->15/24 drove the improvement). Reducing from 9 to ~2 searched params lowers the expected-max-under-noise bar and should lift DSR over 0.95 (PBO already passes at 0.41). Target: DSR >= 0.95 AND holdout ratio >= 0.70. Note the recent-year softness (2025 -3.8%, 2026 -0.3%) and the negative first OOS window — confirm the shrunk-search holdout doesn't degrade further. If attempt 2 still fails DSR/holdout after the shrink, abandon — but the PBO pass, 0-excluding CI, and above-luck-bar Sharpe make this a legitimate retry, NOT the BNB-specific-no-generalization outcome I anticipated pre-opt.

Benjamini-Hochberg at q=0.10 over 225 programme candidates keeps 3. A candidate that does not survive here is not distinguishable from the programme's own noise, however good its individual statistics look.

AdaFourHourDonchianRisingTrendFilterBreakoutLong aimed to complete a three-asset robustness test of the 'only 3-phase-validated' Donchian-with-trend mechanism, adding the directional-context filter that the analyst had diagnosed as missing from ADA's prior failed breakouts. Its base backtest was marginal (Sharpe 0.463, profit factor 1.089, CAGR 2.19%, 18.4% impact cost, $2.96M capacity), and although optimization genuinely lifted it to a real edge (Sharpe 2.81, PBO 0.41 passing, non-overfit walk-forward), it still missed on deflated Sharpe (0.56) and the holdout ratio (0.376). The strategy was abandoned rather than iterated: its iteration-2 remedy (9→2 searched params) could only touch the DSR bar while the binding holdout failure reflected genuine recent-year decay, and its founding premise was refuted — the cited BNB flagship had itself been abandoned in the same session for failing deflated Sharpe, so the cross-asset justification no longer stood. The reviewer noted a separately-approved ADA sequential higher-low breakout as the stronger ADA candidate instead.

A cross-asset robustness test collapses when its flagship premise is false — the Donchian-with-trend class never cleared deflation on any asset — and a parameter-reduction fix that targets only the multiple-testing (DSR) bar cannot repair a holdout failure driven by genuine forward decay, nor lift a thin, toy-capacity, cost-heavy base edge.

It was abandoned as a marginal, cost-constrained, recently-decaying edge built on a refuted premise: the base metrics failed CAGR, capacity, Sharpe, profit factor and sharpe_ci_low; the iteration-2 fix (shrinking the search from 9 to 2 params) could only address the deflated-Sharpe bar, not the holdout failure that reflects genuine forward decay (2025 -9.6%, 2026 -3.7%); and the cited foundational rationale — 'BNB Donchian promoted at Sharpe 5.95' — was actually abandoned in the same session for failing deflated Sharpe, so the mechanism class had cleared deflation on no asset.

A long-only 4H Donchian channel breakout on ADAUSDT.BINANCE perp (pure OHLCV) that entered only when a 12-bar-high breakout coincided with an established uptrend (rising 50-bar SMA and price above it), exiting on a Donchian-low trailing stop or PnL/time stops, intended to complete a cross-asset robustness test of the Donchian-with-trend mechanism class.

The base backtest was marginal — total return +23.7%, CAGR 2.19%, Sharpe 0.463 (CI low -1.085), profit factor 1.089 over 231 long trades, with heavy impact cost (18.4%) and a toy $2.96M capacity — while optimization produced a real edge (Sharpe 2.81, alpha +0.081, profit factor 1.63, PBO 0.41 passing, avg OOS Sharpe 2.18) that nonetheless failed deflated Sharpe (0.5606) and the holdout (ratio 0.376).