ADA Daily Multi-Week Trend Continuation Long-Only (BINANCE USD-M Futures, Daily, OHLCV-Only)

Iteration-3 fix targeting the analyst's two blockers. (1) Trade count: the prior entry required regime AND a fresh 4-day high on the same bar — a tight conjunction that fired only ~15 times in ~6 years and put ~100% of P&L on a single 2021 day. Per the concrete ask, entry is now the regime+stack STATE with a flat-bar re-entry debounce instead of the simultaneous breakout, so it fires across every uptrend regime and re-enters after each pullback stop-out — substantially raising trade count and spreading trades across the full history rather than one pump. (2) The debounce (default 3 flat bars) prevents same-bar churn/re-entry while still allowing continuation re-entries into a persisting uptrend, which chops what was one outlier ride into multiple independent legs, distributing the per-trade edge. Signal stays continuous (EMA-stack spread) to avoid the gated-to-0.0 anti-pattern; only the cached entry boolean changed. Imports, class structure, exit logic, and capital-relative ATR-risk sizing (all of which already passed Layers 1-3) are untouched. This is the last reasonable loosening: if regime-state entry still cannot reach >=30 trades with a positive ex-outlier edge, the ADA multi-week-continuation arm is structurally too sparse and should be abandoned.

Not worth optimizing: the entire result is a single 2021-03-17 bar (+96.97% daily return vs 128% total), with every other year flat-to-negative (2020 -3.2, 2023 -1.7, 2024 +11.5, 2025 +1.6, no trades in 2022). The distribution is a textbook outlier-carried artifact (return_kurtosis 673.7, skew 25.6, tail_ratio 0.0069, largest_win $93.8k vs avg_win $23.5k), so despite a +128% headline the Sharpe is only 0.577 with sharpe_ci_low -3.10 — the canonical deflated-Sharpe failure profile that best-of-225 optimization would only worsen by chasing the 2021 spike. The sample is just 18 trades. Decisively, this is iteration 3 and the developer already refactored the entry SPECIFICALLY to distribute trades across history and stop concentrating P&L in the single 2021 bar (prior analyst feedback); that fix failed — still 18 trades, still one-bar-dominated — proving the concentration is not a tunable code artifact but that ADA's daily returns outside the 2021 parabola do not offer this template's continuation edge. The multi-week continuation mechanism is validated on promoted BNB/BTC/Link siblings, so this is ADA asset-selectivity (a hypothesis-level retarget question), not a fixable implementation. FAILURE PATTERN: porting the promoted BNB/BTC/Link daily multi-week trend-continuation template to ADA yields an 18-trade, one-day-carried result (kurtosis 673, 97% of return from 2021-03-17) that will fail deflation; a param sweep cannot manufacture edge outside the single blow-off bar, and the entry-decoupling fix for exactly this concentration was already exhausted in iteration 3.

Do NOT optimize yet — two blockers remain. (1) Trade count is still 15 vs your stated 30+ target; the strategy fires too rarely on ADA daily for meaningful walk-forward fitting. (2) DECISIVE: 100% of the PnL is a single day (2021-03-17, +93.0%) — return_kurtosis 594, skew 24, tail_ratio 0.0207, largest_win $91.4k vs avg_loss $2.4k. Before resubmitting, compute total return and Sharpe with that one trade EXCLUDED; if the strategy is flat/negative ex-top-trade (it appears to be), the edge is a 2021 pump artifact, not a continuation edge, and no optimization can fix it. Concrete asks: (a) loosen the entry further (e.g. enter on the regime+stack state with a re-entry debounce rather than requiring a simultaneous fresh 4-day high) to reach >=30 trades over the full history; (b) demonstrate the per-trade edge is distributed (positive median trade, Sharpe still >0 with the single best day removed); (c) if a substantially looser entry still cannot produce >=30 trades with a positive ex-outlier edge, this ADA multi-week-continuation arm is structurally too sparse/outlier-dependent and should be abandoned rather than iterated again — this is the last reasonable loosening attempt.