LtcDualTimeframeMomentumConfluenceLS

LTC Dual-Timeframe Momentum Confluence, Long-Short (Single-Instrument LTCUSDT.BINANCE USD-M, 4H Primary Aligned With the 1D Trend, Flat on Disagreement, ATR-Trailing Exit, 3-Parameter)

A LONG-SHORT, single-instrument, pure-OHLCV momentum-confluence trend follower on LTCUSDT.BINANCE (USD-M perpetual), 4H primary bar + 1D trend confirm. This is a DELIBERATE stay-inside-the-survivor-class move: the ONLY mechanism that has ever reached paper in this factory is exactly this recipe on a DEEP major (BTC Sharpe 1.99, ETH 1.05, ADA 1.60 all promoted). The graveyard shows the recipe dies overfit when ported to NOISY alts (XRP, LINK, SOL, BCH, ATOM, DOT all overfit/fee_edge) and dies fee_edge when sped up (ETH-1H). The real discriminator between survivors and failures is INSTRUMENT NOISE, not the mechanism. LTCUSDT is the obvious un-mined candidate that shares the survivors' character: 2020-listed on Binance USD-M with multi-year 4H/1D history, top-15 perpetual liquidity, and distinct multi-week halving-cycle trend regimes — a low-noise, mean-persistent price series far closer to BTC/ETH than to the SOL/LINK failures. LTC is completely absent from the failure list and the corpus (which is 62% BTC recently), so this simultaneously exploits the survivor family AND diversifies away from BTC concentration. Kept to 3 parameters (fast_ema, ema_ratio, atr_mult) to match the survivors' complexity — NO regime/efficiency/volume gate is added, per L92 which shows gates reduce trade count without adding edge.

Implements the hypothesis exactly: the surviving BTC/ETH/ADA recipe, unchanged in mechanism, ported to LTC as a low-noise 2020-listed major with deep 4H and 1D history (both confirmed present in the catalog). Kept to the 3 tunables the hypothesis names (fast_ema, ema_ratio, atr_mult), each with the identical clamp declared in _param_bounds so the optimizer never tests a value the code silently rewrites; daily_lookback, atr_period and the two sizing parameters are structural and bounded narrowly around the survivors' values so search complexity matches the survivor class. NO regime, efficiency or volume gate was added, per the hypothesis and L92/L25 — the entry is a two-condition confluence that fires often enough on 4H bars to clear the sandbox's >=1-trade requirement (warmup ~73 4H bars ~ 12 days, a small fraction of the sandbox window). calculate_signal returns the raw continuous ATR-normalised spread every bar (never a gated constant), and the entry decision lives in should_enter. Futures venue is correct here because the strategy shorts and uses leverage>1; leverage is genuinely consumed in position_size via the max_notional cap, not merely declared. The extra_instruments list repeats LTCUSDT.BINANCE so the daily bar type is paired positionally and the slow leg actually receives bars.

Negative expectancy on a decisive 475-trade / 2432-day sample: avg_trade_return_pct -0.247% (below zero, under the 0.15% Binance USD-M floor), profit_factor 0.843 (<1), total_return -22.6%, Sharpe -0.206 with CI straddling zero, expectancy -$71.6/trade. Commission is only 5.25% of gross, so the signal is gross-negative, not fee-fragile — this cannot be optimized above costs. The strategy LOSES in all three vol terciles (calm/normal/stressed) and in 4 of the last 5 years, falsifying the 'LTC is a low-noise major like BTC/ETH' premise; it is the exact port-to-alt failure the graveyard shows (XRP/LINK/SOL/BCH/DOT), and QA's edge concern is borne out on full history. Not worth 2 hours of optimization. abandon_class=negative_expectancy (premise_falsified).

Long-short dual-timeframe momentum-confluence trend follower on LTCUSDT.BINANCE (USD-M perpetual). The fast leg reads 4H bars as an ATR-normalised EMA spread ((EMA_fast - EMA_slow)/ATR, computed incrementally every bar); the slow leg reads 1-DAY LTC bars and takes the sign of the 18-day return. It goes LONG only when both are positive, SHORT only when both are negative, and stays FLAT on any disagreement. Exits on an ATR trailing stop (give-back of atr_mult x entry ATR from the best close since entry), on the daily trend flipping against the position, or on the 4H spread flipping against it. Sizing is risk-first: equity * risk_pct / (atr_mult * ATR), capped at max_notional_frac * equity * leverage (2x) so the requested leverage is actually consumed and per-trade risk stays ~1.3% of equity regardless of the equity path.

Either restrict _param_bounds to the 3 named tunables (fixing the structural params at their defaults) to honor the survivor-complexity constraint, or update the hypothesis to acknowledge the 7-param surface. Watch PBO / deflated Sharpe at the holdout gate.

The hypothesis states the strategy is 'Kept to 3 parameters (fast_ema, ema_ratio, atr_mult)' to match the survivor class's complexity and avoid the overfitting that killed alt ports. However _param_bounds declares 7 tunable ranges (adds daily_lookback, atr_period, risk_pct, max_notional_frac), so the optimizer will search a 7-dimensional space. The structure (which exits/stops exist) matches the hypothesis exactly, so this is not a mechanism mismatch — but the wider search surface is at odds with the stated overfit-avoidance rationale.

Two advisory concerns for the analyst, neither blocking. (1) The sandbox smoke test is negative (total_return -22.8%, Sharpe -0.22, PF 0.89, avg_trade_return_pct -0.257% over 471 trades) and, unlike the survivors this is cloned from, LTC loses across ALL three vol terciles (calm -14.9%, normal -6.6%, stressed -10.5%) rather than earning in a specific regime — so the 'LTC is a low-noise major like BTC/ETH' premise is not yet supported by even the unoptimized draw. This is a single unoptimized parameter set over a truncated window, so it does not block, but it is the exact port-to-alt failure signature (XRP/LINK/SOL/BCH/DOT all died overfit/fee_edge). (2) The hypothesis pre-registers a 3-parameter surface to match survivor complexity and avoid overfitting, but _param_bounds exposes 7 optimizable parameters (fast_ema, ema_ratio, atr_mult PLUS daily_lookback, atr_period, risk_pct, max_notional_frac). The extra degrees of freedom widen the search space the thesis explicitly wanted narrow; worth watching for PBO/deflated-Sharpe overfit at the optimization/holdout gate.

Reliable, decisive sample: 475 trades over 2432 days (metrics_reliable=true); trades match the hypothesized dual-TF confluence mechanism (263 long / 212 short, entry_signaled 475 = entries_submitted 475, no drops).

Code is clean and faithfully implements the survivor recipe (3 core tunables, ATR-trailing exit, risk-first sizing).

Negative expectancy: profit_factor 0.843 (<1), total_return -22.6%, avg_trade_return_pct -0.247% (below ZERO and far under the 0.15% Binance USD-M fee floor), expectancy -$71.6/trade.

Sharpe -0.206 with CI [-0.96, 0.53] straddling zero; Sortino -0.25; win_rate 0.295.

Loses across ALL vol regimes (calm -4.7%, normal -16.8%, stressed -7.4%) and in 4 of last 5 years (2022 -13.7%, 2023 -7.3%, 2024 -15.4%, 2025 -17.7%) — no regime where the edge exists.

Commission only 5.25% of gross, so this is a gross-negative signal, not fee-fragility — the LTC 'low-noise major' premise is falsified; the port-to-alt failure signature (XRP/LINK/SOL/BCH/DOT) repeats.

max_drawdown 44.8%, recovery_factor -0.68.

LtcDualTimeframeMomentumConfluenceLS-f44b46b287

LTC did not behave as a low-noise major for this mechanism; the port-to-alt failure signature repeated, so noise assumptions need full-history validation, not inference from listing age/liquidity.

Abandoned at the pre-optimization backtest-review gate for negative expectancy (-$71.6/trade) across all vol regimes and 4 of 5 recent years — no edge to optimize above costs.

Port the factory's only surviving mechanism—a dual-timeframe momentum-confluence trend follower (4H primary aligned with the 1D trend, flat on disagreement, ATR-trailing exit, 3 tunables)—onto LTCUSDT.BINANCE, betting LTC is a low-noise major like the promoted BTC/ETH/ADA.

Over 475 trades / 2432 days it was gross-negative: -22.6% return, Sharpe -0.206, PF 0.843, win rate 29.5%, avg_trade_return_pct -0.247%, 44.8% max drawdown; commission was only 5.25% of gross.