XrpDonchianBreakoutTrendCaptureLS

XRP USD-M 4h Donchian Breakout Trend-Capture, Long-Short (Single Instrument XRPUSDT.BINANCE Perp, 4h Bars, Enter 20-Bar Channel Breakouts Aligned with the 100-Bar Trend, Chandelier ATR Trailing Exit, Fires Frequently — 3-Parameter)

A LONG-SHORT, single-instrument BREAKOUT trend-capture on XRPUSDT.BINANCE (USD-M perp, 4h bars). Momentum/trend is the ONLY family with a demonstrated live survivor in this factory (the deployed cross-sectional momentum rotation, Sharpe 2.6) — every non-momentum family I have tried (intraday reversion fades, OI/positioning fades, order-flow) has died fee_edge/no_edge, so this leans into the one edge that works, but as a SINGLE-NAME absolute breakout rather than a basket rotation (L32 exhausted) or an always-in TSMOM (my pending ETH sibling). Two explicit fixes for the deaths that killed my prior breakout attempts: (1) TRADE COUNT — my SOL DAILY breakout died 'no trades produced' because a 200-day filter + 55-day channel almost never fired inside the ~363-bar (~1yr) sandbox; this uses 4h bars (thousands of sandbox bars) with a SHORT 20-bar channel (~3.3 days) and a 100-bar (~16d) trend filter, guaranteeing dozens-to-hundreds of trades in both sandbox and full history. (2) IMPLEMENTATION — my CM inverse breakout had a PnL-sign code defect and my BNB flow strategy timed out; this stays on a standard linear USD-M perp with pure OHLCV and O(1) incremental rolling max/min (deque) + running ATR, no full-series rescans. It is NOT reversion (opposite polarity), NOT a basket, NOT carry/options/COIN-M/liquidation. XRP is chosen for its strong episodic multi-percent trends (ideal for breakout capture) and low usage in the current book. 3 parameters.

Leans into momentum/trend -- the only family with a live survivor here -- as a single-name absolute breakout rather than a basket rotation (L32 exhausted) or an always-in TSMOM (the pending ETH sibling), and is explicitly NOT reversion/carry/options/COIN-M/flow. It fixes the two failure modes that killed prior breakout attempts: (1) TRADE COUNT -- on 4h bars the sandbox holds thousands of bars, and with a short 20-bar channel (~3.3d) plus a 100-bar (~16d) trend filter the warmup (min_bars = max(100,20)+14+5 = 119) is a tiny fraction of the window, so breakouts fire dozens-to-hundreds of times (the opposite of my SOL DAILY 55/200 breakout that died 'no trades' because warmup approximated the whole 1yr sandbox); (2) IMPLEMENTATION -- a standard LINEAR USD-M perp avoids the inverse-contract PnL-sign trap that zeroed my CM breakout, and the exit uses the proven entry-anchored chandelier pattern (favourable-extreme ratchet + ATR frozen at entry + a minimum-hold guard) that eliminates the immediate-close-at-entry defect I hit on the ETH squeeze sibling; execution is O(1)-ish bounded-window rolling max/min + mean-TR ATR with no full-series rescans, so no smoke-test timeout. calculate_signal returns a CONTINUOUS breakout score (close vs channel midpoint in ATR units) that varies every bar, with the entry decision thresholded in should_enter (breakout AND trend alignment), per the continuous-signal guidance. The chandelier width clears the ~0.10% USD-M round-trip by capturing multi-percent XRP trends. Leverage (3x) is genuinely wired into position_size as the notional cap (binds in low-vol regimes), so it is not an inert leverage setting. XRP chosen for strong episodic trends and low usage in the current book.

Failed deflated Sharpe and walk-forward overfit: DSR=0.0, PBO=0.54 over 225 trials, optimized Sharpe 0.233 vs expected-max 1.98 — the selected config is indistinguishable from best-of-N noise, and the Sharpe CI [-0.37, 0.87] straddles zero. The walk-forward is a textbook overfit signature (IS avg 1.266 → OOS avg -0.004, is_overfitted=TRUE, per-window OOS two-of-three negative), and the holdout failed catastrophically (Sharpe -0.305, ratio 0.0 vs 0.70 floor). Four unwaivable hard-gate failures (OOS≤0, overfit, PBO>0.5, holdout fail). Optimization made it worse, not better (Sharpe 0.35 → 0.233; 2025/2026 annual returns negative), confirming the pre-opt thinness I flagged. The 225-trial sweep covered the whole 3-parameter space and found no robust region, so this is overfit → abandon, not revise_hypothesis: reframing the same single-name absolute-breakout mechanism onto another instrument would just overfit again (the ETH turtle and SOL spot breakout siblings died the same way), and the only trend survivor in this book is the structurally different basket rotation. No generalizable edge exists to tune toward.

Long-short single-instrument Donchian breakout trend-capture on XRPUSDT.BINANCE (USD-M perp), 4-hour bars, pure OHLCV. Enters a 20-bar channel breakout only when aligned with the 100-bar trend SMA (close above prior 20-bar high AND above the SMA -> long; below prior 20-bar low AND below the SMA -> short), then rides it with a chandelier ATR trailing stop from the favourable extreme since entry (using the ATR frozen at entry) plus a trend-filter-break exit. Sizing risks ~2% of equity over the chandelier stop distance, capped at leverage(3x) x equity. 3 tuned params (channel_len, trend_len, atr_mult).

Analyst/optimizer: check whether tuning channel_len/trend_len/atr_mult lifts avg_trade_return_pct clearly above 0.15% and PF above ~1.2 OOS; if per-trade return stays pinned near the ~0.10% fee line, abandon rather than promote a marginal edge.

Thin per-trade edge, near the fee floor. Sandbox avg_trade_return_pct is 0.1026% NET — positive (gross ~0.20% does clear the ~0.10% USD-M round-trip), but BELOW the 0.15% futures viability floor the analyst enforces, and profit_factor is a razor-thin 1.043. Sharpe 0.064 with CI [-1.69, +1.65] straddles zero, so the in-sample edge is statistically inconclusive. Not a code defect and not the gross<cost fee-death pattern (net expectancy positive), but the margin over costs is slim enough it may not survive the analyst's floor or the walk-forward.

Optimizer: confirm profitability isn't a single-trade artifact via per-window stability; watch fee/slippage sensitivity given the thin PF and high loss count.

Return is concentrated in the right tail, typical of trend-following: win_rate 0.33, avg_win 2401 vs avg_loss 1151, skew 2.43, kurtosis 20.5, largest_win 14381 vs total_return only +1.99%. The positive result leans heavily on a few large XRP trends, so it is fragile to those episodes and to slippage on the many small losing breakouts. Edge-robustness matter for the analyst/walk-forward, not a correctness defect.

avg_trade_return_pct 0.90% comfortably clears the 0.15% futures fee floor — genuine per-trade edge net of costs.

profit_factor 1.20 with positive expectancy (+$219/trade), max_drawdown 23.4% (well under 50%), calmar 3.10.

354 well-sampled trades, balanced 180 long / 174 short, all signals submitted with zero drops — statistically meaningful for optimization.

Positive annual returns in 6 of 7 years across bull/bear/chop — not a single-regime artifact; headline is realized (end_unrealized only 8.1%), capacity $7.9M.

Momentum/trend family — the one family with a demonstrated live survivor in this factory.

Headline Sharpe only 0.35 with CI [-0.26, 1.00] straddling zero — risk-adjusted quality is thin and not yet significant.

information_ratio -0.52 — underperforms buy-and-hold XRP (expected for a long-short trend system, but must be judged on absolute Sharpe after tuning).

Win rate 34.7% — right-tail-dependent (largest_win $27k, kurtosis 11.8); reliant on episodic big trends, so vulnerable to trend-poor stretches.

Well-sampled (415 trades, balanced 208 long / 207 short) and per-trade edge still clears fees (avg_trade_return_pct 0.58%).

Sensitivity was mostly clean (only atr_mult flagged as a cliff) and drawdown stayed modest (21.4%) — the failure is generalization, not risk control.

Walk-forward overfit: IS avg Sharpe 1.266 collapses to OOS avg -0.004 (is_overfitted=TRUE); per-window OOS [-0.60, -0.71, +1.30], two of three negative.

Holdout failed catastrophically: Sharpe -0.305, ratio 0.0 vs 0.70 floor.

PBO 0.5436 > 0.5 and deflated_sharpe 0.0 — selected Sharpe indistinguishable from best-of-225-trials noise (expected-max 1.98); Sharpe CI [-0.37, 0.87] straddles zero.

Optimization degraded the result: Sharpe 0.35 → 0.233; recent annual returns negative (2025 -0.8%, 2026 -5.4%).

Four unwaivable hard gate failures (OOS≤0, overfit, PBO>0.5, holdout fail).

XrpDonchianBreakoutTrendCaptureLS leaned into the one family with a live survivor in this factory, trading 4h XRP Donchian breakouts aligned with a trend filter and riding them with a chandelier stop. Its first backtest passed the pre-optimization gate with 354 trades, a 0.90% per-trade edge clearing fees, profit factor 1.20 and a modest 23.4% drawdown, though headline Sharpe was only 0.35 with a confidence interval straddling zero. The three-parameter optimization then exposed the thinness: a 225-trial walk-forward showed textbook overfit (in-sample Sharpe 1.27 → out-of-sample -0.004), the holdout failed catastrophically (Sharpe -0.305, ratio 0.0), PBO hit 0.54 and deflated Sharpe was 0.0. The analyst abandoned it on four unwaivable hard-gate failures, concluding no generalizable edge existed to tune toward and that reframing the same single-name breakout onto another instrument would only overfit again.

A thin headline Sharpe near zero on a right-tail-dependent single-name breakout is a warning sign that survives to optimization: a wide parameter sweep can expose that the apparent edge is best-of-N noise with no generalizable region, so single-name absolute-breakout reframes tend to overfit where only the structurally different basket rotation survives.

The strategy cleared the backtest-review gate to optimization on a genuine per-trade edge, but the post-optimization analyst issued an abandon verdict on four unwaivable hard-gate failures — OOS≤0, walk-forward overfit, PBO>0.5, and a catastrophic holdout fail — showing the selected config was indistinguishable from best-of-225-trials noise and no robust parameter region existed.

A long-short, single-instrument Donchian breakout trend-capture on XRPUSDT.BINANCE 4h perp bars that entered 20-bar channel breakouts aligned with a 100-bar trend filter and rode them with a chandelier ATR trailing stop, leaning into the momentum/trend family as a single-name absolute breakout.

The initial backtest looked promising — 354 well-sampled trades (180 long / 174 short), avg_trade_return_pct 0.90% clearing the fee floor, profit factor 1.20, positive expectancy (+$219/trade), 23.4% max drawdown and profitable in 6 of 7 years — but headline Sharpe was only 0.35 with CI [-0.26, 1.00] straddling zero. Optimization over 225 trials then degraded it: Sharpe fell to 0.233, walk-forward in-sample Sharpe 1.266 collapsed to OOS -0.004 (overfit), the holdout failed (Sharpe -0.305, ratio 0.0), PBO was 0.54 and deflated Sharpe 0.0.