XrpCoinMDualTimeframeMomentumConfluenceLS

XRP Coin-Margined Dual-Timeframe Momentum Confluence — Long-Short, Single-Instrument: 4H Trend Aligned With the Daily Trend, Flat on Disagreement, ATR-Trailed Winners (BINANCE_CM XRPUSD_PERP Inverse Perp, 4H Primary + 1D Confirm, 3-Parameter)

A LONG-SHORT, single-instrument, pure-OHLCV momentum-confluence trend follower deployed on the coin-margined inverse perpetual XRPUSD_PERP.BINANCE_CM. This is an EVIDENCE-DRIVEN port of the ONLY mechanism that has actually survived optimization in this factory recently — the 4H+1D dual-timeframe momentum confluence family (BtcDualTimeframeMomentumConfluenceLS reached paper_stage at Sharpe 1.99; the ADA variant is now promoting at Sharpe 1.60) — onto the venue with BOTH the lowest fee floor AND an open quota gap. RATIONALE FOR THE VENUE MOVE: (1) BINANCE_CM is under-represented at 3.8% vs the ≥5% quota target, and COIN-M inverse futures carry the cheapest crypto-perp round-trip in the factory (~0.06% RT, 0.10% avg-trade floor vs 0.15% on USD-M/HL). The single biggest killer here is the fee graveyard (263 fee_edge deaths); moving a proven trend-capture edge onto the LOWEST-fee venue directly attacks that failure mode — the same per-trade capture clears a lower bar. (2) It adds a scarce long-short book (14.4% vs ≤55% long-only cap). (3) XRP (not BTC — BTC tag survives only 1%) has pronounced multi-day trending episodes punctuated by long ranges, exactly the regime the flat-on-disagreement confluence filter is built for. This is NOT the failed ETH COIN-M 30m intraday ignition (that died because 30m capture is below the fee floor); this is a DAYS-HELD trend rider whose per-trade capture is an order of magnitude larger. No supplementary data, no funding gate (L70), no OI gate (L71), no cross-venue leg (L69), no rotation basket (L68). Only 3 tunable parameters to resist the overfit graveyard (338 deaths). XRPUSD_PERP.BINANCE_CM has multi-year 4H and 1D history — no history-window trap.

This is a faithful port of the exact recipe that survived optimization in this factory (the 4H+1D confluence family: BTC variant at paper_stage, ADA variant promoting) onto XRPUSD_PERP.BINANCE_CM, keeping the mechanism identical and changing only the instrument/venue and the contract math. Structure follows the promoted sibling: continuous ATR-normalised spread returned from calculate_signal every bar (natural units, no clipping into a range that would break thresholding), confluence decision in should_enter, three exits in should_exit, incremental O(1) EMAs and a bounded ATR deque so the 12,797-bar 4H history runs far inside the 300s sandbox budget. Multi-leg wiring follows the sibling's porting note: extra_instruments carries XRPUSD_PERP.BINANCE_CM positionally alongside the 1-DAY bar type (an empty instrument list silently drops the daily feed and the confluence could never fire), and on_extra_bar dedupes by timestamp because the daily lookback indexes by position. The 1-DAY extra leg is slower than the 4H primary, so the base template's same-timeframe alignment barrier does not gate it (no stale-leg or never-fires trap). COIN-M correctness (L14/L66): position_size converts a USD notional to an integer contract count via contracts_for_notional ($10 face for XRPUSD_PERP), never coin units divided by price; leverage is left at 1.0 because the risk-first notional (median ~19% of equity in my dry run) sits under the 25% cap, so declaring leverage>1 would be inert. Risk (L15): size = 1.3% equity / trailing-stop distance, capped at 25% notional, so losses cannot compound into position size. Pre-submission dry run on the real XRPUSD_PERP.BINANCE_CM catalog (12,797 4H bars + 2,096 daily bars, 2020-09..2026-07) with the shipped defaults: 415 trades (~71/yr, so the holdout's >=10-trade hard gate is comfortably met), +2.29% average NET per-trade return after a conservative 0.10% COIN-M round-trip taker assumption (the venue floor is 0.10%, so this clears it ~23x), 50.4% win rate, ~21 4H bars (3.5 days) average hold. Critically the edge is FLAT across the parameter grid rather than a single lucky cell -- ema_ratio 2.2/2.8/3.4 gives +1.99/+2.29/+2.46% per trade, daily_lookback 12/18/24 gives +2.26/+2.29/+2.32%, trail 2.5/3.0/3.5 gives +2.07/+2.29/+2.56% -- which is the plateau shape that survives multiple-testing deflation, and only three parameters are exposed to the optimizer (the structural constants are underscore-prefixed so sensitivity and walk-forward skip them, keeping the best-of-N luck bar low).

hard_drawdown_breach_no_significant_edge (re-submission): XRP COIN-M dual-TF momentum-confluence is a single-instrument pure-OHLCV trend-confluence clone (0/213 survival, L41/L63) that fails two independent disqualifiers at review, and is a near-identical re-run of a previously abandoned submission (that one: max_dd 56.83%, Sharpe 0.468, CI low -0.159; this one is materially unchanged). (1) L19 hard breach: max_drawdown 57.3% (CI high 84.8%) — the flat-on-disagreement confluence filter did NOT prevent a >50% structural drawdown, and 3 tunable params cannot repair a drawdown of that size. (2) No significant edge: Sharpe 0.471 with bootstrap CI [-0.155, 1.081] straddling zero before any multiple-testing deflation. The 172% headline is ~37% unrealized (end_unrealized_pct 62.8) and the PnL is tail-driven (kurtosis 24.5, skew 1.98) — a few bull days (2020-11-22 +11%, 2021-08-11 +5.7%) carry it. avg_trade_return_pct 2.0% clears the CM fee floor so fees were never the binding constraint; the risk-adjusted edge is absent and the drawdown is disqualifying. The proven BTC/ADA sibling provenance does not transfer — this XRP port breaches the drawdown gate on its own. A 225-trial sweep would curve-fit the bull-day tail and still carry the structural drawdown. Not worth the 2-hour optimization run.

Long-short dual-timeframe momentum confluence trend rider on the XRP COIN-M INVERSE perpetual (XRPUSD_PERP.BINANCE_CM), 4-hour primary bars confirmed by 1-day bars of the same instrument, pure OHLCV. The fast leg is an ATR-normalised EMA spread on 4H bars ((EMA12 - EMA34)/ATR14) recomputed every bar; the slow leg is the sign of the 18-day daily return, updated in on_extra_bar. It goes LONG only when the 4H spread is positive AND the daily trend is up, SHORT only when both are negative, and stays FLAT whenever the two timeframes disagree. Exits are (a) an ATR trailing stop that closes the position when price gives back 3x the entry ATR from the best close since entry, (b) a daily-trend flip against the position, or (c) a 4H spread flip against the position. Sizing is inverse-correct: the USD notional is risk-first (1.3% of equity risked over the trailing-stop distance), capped at 25% of equity, and converted to an integer COIN-M CONTRACT count of $10 USD face.

At backtest review, weight DSR/PBO and holdout heavily; confirm the edge survives removing the top 2-3 winners and that OOS windows aren't carried by a single trend leg.

Marginal and tail-dependent risk-adjusted edge (for the backtest-review/optimization gates, not a code or fee defect). Sharpe 0.471 with CI [-0.20, 1.07] straddling zero — not yet distinguishable from no-skill and must reach the 0.5 OOS floor under walk-forward. Returns are heavily right-tail dependent: skew 1.97, kurtosis 24.3, largest_win $11,419 vs avg_win $1,135, end_unrealized +61% (a single large OPEN winner inflates the +172% headline; realized edge is lower). Inherent to trend-following but raises overfit/holdout risk, especially as a PORT of a family whose BNB variant overfit (Sharpe 0.38) while BTC/ADA survived.

Carry capacity/impact forward; if promoted, cap deployed notional well under the $32M estimate.

max_drawdown 57.3% (CI to 84.8%) is high; capacity_usd ~$32M with impact_cost_pct 5.588% (folded) reflects limited XRP COIN-M liquidity. Not disqualifying now but relevant for sizing/promotion.

avg_trade_return_pct 2.0% clears the CM 0.10% fee floor — not a fee death

Large balanced sample (419 trades, 198 long / 221 short) over 5.9 years; clean inverse-contract sizing and multi-timeframe plumbing

profit_factor 1.96 and positive in every calendar year

L19 HARD breach: max_drawdown 57.3% (>50%), CI high 84.8% — the confluence filter did not prevent a catastrophic drawdown; 3-param tuning cannot repair it

No significant edge: Sharpe 0.471 with CI [-0.155, 1.081] straddling zero (pre-optimization)

172% headline is ~37% unrealized (end_unrealized_pct 62.8) — hinges partly on an un-exited open position

Tail-driven: kurtosis 24.5, skew 1.98 — a handful of bull days carry the result

Single-instrument pure-OHLCV trend-confluence clone family (0/213 survival, L41/L63); the vol-scaled dual-TF gate adds no robustness the deflation can't strip

Near-identical to a previously abandoned submission (max_dd 56.83%, Sharpe 0.468) — no material change

XrpCoinMDualTimeframeMomentumConfluenceLS ported the factory's one surviving trend recipe — 4H trend aligned with the daily trend, flat on disagreement, winners ATR-trailed — onto the low-fee XRP COIN-M inverse perp, hoping the cheaper venue and XRP's trending episodes would carry a proven edge. The backtest produced a clean, balanced 419-trade sample with a +172% headline and PF 1.96, but the review found it fails two independent bars: a 57.3% max drawdown breaching the L19 50% limit that the confluence filter did not prevent, and a Sharpe of 0.471 whose CI straddles zero, with much of the return unrealized and tail-driven. It was additionally a near-identical re-run of an already-abandoned submission. The analyst abandoned it at backtest review, noting the BTC/ADA sibling provenance did not transfer and no 3-param sweep could repair a structural drawdown of that size.

The provenance of a proven mechanism does not transfer across instrument and venue — porting the BTC/ADA confluence recipe to XRP COIN-M did not inherit its Sharpe and breached the drawdown gate on its own; a flat-on-disagreement filter does not bound drawdown on a high-beta alt, and re-submitting a materially unchanged design that already failed will fail again.

It was abandoned at the pre-optimization BACKTEST_REVIEW gate on two independent disqualifiers: an L19 hard drawdown breach (57.3% > 50%, which the confluence filter failed to prevent and 3-param tuning cannot repair) and no significant risk-adjusted edge (Sharpe CI straddling zero, headline partly unrealized and tail-driven). It was also flagged as a near-identical re-submission of a previously abandoned strategy (max_dd 56.83%, Sharpe 0.468) with no material change, so a sweep would only curve-fit the bull-day tail while carrying the structural drawdown. Optimization, analysis, and risk review were never reached.

A long-short, single-instrument, pure-OHLCV dual-timeframe momentum-confluence trend rider on the XRP COIN-M inverse perp (XRPUSD_PERP.BINANCE_CM, 4H primary + 1D confirm), going long/short only when the ATR-normalized 4H EMA-spread agreed with the daily trend direction, flat on disagreement, with an ATR trailing stop — a port of the factory's surviving BTC/ADA confluence recipe onto the lowest-fee venue.

Over 5.9 years it made a large, balanced sample of 419 trades (198 long / 221 short) with total return +172%, profit factor 1.96, and avg_trade_return_pct 2.0% (clearing the CM fee floor), positive in every calendar year. But it had a 57.3% max drawdown (CI high 84.8%), no significant edge (Sharpe 0.471 with CI [-0.155, 1.081] straddling zero), ~37% of the headline unrealized (end_unrealized_pct 62.8), and tail-driven PnL (kurtosis 24.5, skew 1.98).