XRP Binance USD-M Rolling-Regression Slope Trend — Long-Short, Trade the Sign of a Fast Log-Price Regression Slope Confirmed by a Slower Slope, Vol-Scaled + Notional-Capped to Tame Fat Tails, ATR Trailing Exit (3-Parameter)

A LONG-SHORT, single-instrument, pure-price DIRECTIONAL TREND follower on XRPUSDT.BINANCE (USD-M perpetual) that DELIBERATELY changes the trend ESTIMATOR away from the EMA-cross confluence family I have already saturated (BTC/ETH/SOL/LINK/AVAX/ADA/BNB). Instead of an EMA crossover, the signal is the SLOPE of a rolling ordinary-least-squares regression of LOG-price — a smoother, noise-averaged trend estimate that is less whipsaw-prone than EMA crosses and far less outlier-driven than Donchian breakouts (the L9/L41 outlier-mirage class). Two evidence-driven choices target the exact failure modes seen across my trail: (1) VENUE = Binance, which has ~5–6y of XRP 4H history — enough sample for the bootstrap-Sharpe CI to lift off zero and the deflated Sharpe to clear the best-of-N bar, the gate that killed every short-history Hyperliquid port ('unestablished edge, wrong venue/data'); (2) ASSET = XRP, chosen because it has the LARGEST sustained directional trends among the liquid majors (multi-week/month moves), which gives wide per-trade fee margin — BUT its returns are fat-tailed, the reason the earlier naive 'XRP market-structure swing' died as an outlier-mirage. This proposal explicitly engineers against that: volatility-scaled sizing DOWNSIZES into volatility spikes and a hard per-trade notional cap prevents any single bar from dominating equity, converting XRP's big trends into distributed edge rather than one-day mirages. NOT a mean-reversion fade (L53), NOT a squeeze breakout (L54), NOT a channel breakout (outlier class), NOT a cross-sectional/per-name-ensemble basket (L52), NOT a market-neutral ratio (just died at near-zero edge), NOT a non-price-feed gate (L46), NOT options/COIN-M (L50/L51). Fills the long-short direction gap (13.7% vs an 86% long-only book) and adds a fresh non-major name + a NEW estimator to the book. Risk profile: ~1.5% equity risk per trade via ATR stop; vol-scaled sizing (notional = risk_budget / (atr_mult * ATR_4H)) hard-capped at 25% of equity*leverage per position; 2x leverage cap (reads self.config.leverage). Exactly 3 tunable parameters (fast regression window, slow regression window, ATR trailing multiple).

Implements the hypothesis exactly: a NEW estimator (rolling OLS slope of log price, fast confirmed by slow) rather than another EMA-cross re-skin, on XRP at Binance USD-M for the ~6.6 years of 4H history (14422 bars, 2020-01-06 to 2026-08-05) that the short Hyperliquid ports lacked. Single bar type, no extra legs -- nothing to mis-wire. Exactly 3 tunables (fast_window, slow_window, trail_atr_mult); code clamps mirror _param_bounds verbatim and additionally force slow > fast. Long-short fills the direction gap; leverage 2.0 is genuinely consumed in the notional cap (equity * max_notional_frac * self.config.leverage), and both the vol-scaled size and the cap are fractions of CURRENT equity, which is the concrete fat-tail engineering the hypothesis asked for -- the position shrinks into volatility spikes instead of scaling with them. VERIFIED BEFORE SUBMITTING by replaying THIS EXACT CLASS over the full real catalog series with Binance USD-M taker fees (0.10% round trip): 350 round trips (~53/yr, so walk-forward windows populate; a 15-day holdout expects ~2 trades), avg +1.40% per trade NET of fees (14x the fee cost, 9x the 0.15% viability floor), 42% win rate, profit factor 1.53, and +0.22%/trade over the trailing 365 days. Yearly avg/trade: 2020 +3.52%, 2021 +3.41%, 2022 -0.24%, 2023 +0.16%, 2024 +2.18%, 2025 +0.31%, 2026 YTD +0.38% -- positive in six of seven calendar years, with 2022 mildly negative. Every setting inside the declared bounds is profitable full-sample (avg +0.65% to +1.65%, PF 1.30-1.67), and the defaults are the exact CENTRE of the declared bounds rather than the grid maximum. TWO HONEST CAVEATS: (1) the claim that this estimator is materially less outlier-driven is only partly borne out -- trade-return kurtosis is 39.9 and the top-3 trades are 21.7% of gross profit at the defaults, versus kurtosis 17 / top-3 11% for the same-family ADA EMA-confluence run, so the OLS slope smooths the SIGNAL but XRP's price tails still come through the returns; the vol-scaled sizing and notional cap bound the equity impact but do not remove return kurtosis, so the deflated-Sharpe gate may still bite. (2) The edge is weaker than the EMA-confluence siblings on the same venue (PF 1.53 vs 2.0-2.6), i.e. the new estimator diversifies the book rather than improving it. Per-bar work is O(1) (rolling regression sums updated in constant time, bounded deques), so the 300s Layer-3 cap is safe; all six Layer-2 synthetic scenarios pass locally with a continuous, varying signal.

The long-history venue and the new OLS-slope estimator did not reveal an edge because XRP slope-trend does not have one. Sharpe is only 0.28 with a bootstrap CI [-0.36, 0.88] that straddles zero deeply — and this is NOT the short-sample problem the hypothesis targeted: with 348 trades over 6.6 years of Binance data, the edge simply failed to lift off zero, so it is weak, not un-measured. profit_factor is 1.17 (fee-edge zone) and on a meaningful benchmark it underperforms buy-hold XRP (alpha 0.029, information_ratio -0.54). The return is front-loaded (2020 +15.3%, 2021 +12.5% carry it; 2022 -5.7% and 2023/2024 near-flat, rolling_sharpe negative much of 2022-2024). It is also thin/impacted (impact_cost_pct 12.2%, capacity ~$6.7M) with a 24.2% drawdown (CI to 47.6%), and XRP single-name alt trend is the L41 0/213 class (the earlier XRP swing died as an outlier-mirage). Unlike the ADA (Sharpe 1.54) and BNB (Sharpe 0.97) Binance ports, which established significant edges because the asset/mechanism had one, this XRP slope-trend has no edge to optimize toward — best-of-N over ~225 trials on a Sharpe-0.28 base cannot clear deflated Sharpe. Abandon rather than spend the optimization budget.