LtcDualTimeframeConvictionScaledTrendLS

LTC Dual-Timeframe Trend with Continuous Conviction Sizing, Long-Short (Single-Instrument LTCUSDT.BINANCE USD-M, 12H Primary + 3D Trend Confirm, Position Size Scales with Trend Agreement Strength, ATR-Trailing Exit, 3-Parameter)

A LONG-SHORT, single-instrument, pure-OHLCV directional trend strategy on LTCUSDT.BINANCE (USD-M perpetual). It is the factory's ONE surviving mechanism family (dual-timeframe momentum confluence, long-short, vol-scaled, few params) deliberately moved off the saturated BTC/ETH/ADA names onto a liquid mid-cap major (Litecoin) that has NOT been run through this mechanism (LTC has only appeared as a variance-ratio directional bet and an LTC/BCH cointegration pair, both distinct and both failed for reasons orthogonal to trend). Two DELIBERATE differences from the promoted clones, chosen so this is not a cosmetic symbol-swap: (1) a SLOWER clock — 12-hour primary bars confirmed by a 3-day trend proxy, versus the 4H+1D of the survivors — which cuts trade frequency and therefore fee drag while capturing LTC's characteristic multi-day directional swings; (2) CONTINUOUS conviction sizing instead of a binary flat/on gate: position notional scales with the MAGNITUDE of trend agreement (the product of normalized 12H and 3D trend slopes), so the book takes small size in weak/ambiguous trends and full size only in strongly aligned regimes. This is additive sizing, not a suppressive regime filter (L92): every bar with sign agreement trades, size just varies, so trade count is preserved while capital concentrates where the edge is strongest. Single instrument, no supplementary data, 3 parameters (fast lookback, slow lookback, ATR-trail multiple) to resist overfitting.

Implements the hypothesis exactly: (1) SLOWER CLOCK - 12H primary + 3-DAY confirmation instead of the survivors' 4H+1D, which on the full 6.5y LTC history yields ~32 round trips/year (fee drag ~0.10% RT is small next to the multi-day swings captured); (2) CONTINUOUS CONVICTION SIZING - not a binary flat/on gate: the agreement weight w is a smooth function of the PRODUCT of the two normalised slopes, and position notional scales with |signal| between a 20% floor and full size, so trade count is preserved (additive sizing, L92) while capital concentrates in strongly aligned regimes. Venue: BINANCE USD-M futures is required because the book goes SHORT; leverage is left at 1.0 and is genuinely unused (max notional 0.6x equity), so no inert-leverage flag. Single instrument, no supplementary data, 3 tunables (fast_lookback, slow_lookback, trail_atr_mult) with clamps mirrored in _param_bounds; slow_lookback defaults to 11 3-day bars (~1 month) and fast to 10 12H bars (~5 days), the classic month/week dual-timeframe pair rather than a sweep-maximising corner. All per-bar work is O(1) (deques with running sum/sum-of-squares, incremental ATR) so the 300s smoke budget is not at risk. A dry-run of the exact logic on the catalog's LTC 12H+3D bars produced 209 trades over 6.5 years with the entry gate firing on ~62% of bars, so zero-trade and warmup-starvation failures are ruled out. Honest caveat for the analyst: in that dry-run the mechanism's per-trade edge on LTC is positive only on the month-scale slow leg (slow_lookback>=11) and is negative for shorter slow lookbacks - the ridge may not be robust out of sample.

Near-breakeven, cost-fragile signal on a decisive 200-trade / 6.5-year sample: profit_factor 1.0316 sits in the fee/impact drag band [0.85,1.10] (impact 22.9% of gross), Sharpe 0.189 with CI [-0.599, 0.935] straddling zero, Calmar 0.027, recovery_factor 0.174. The nominal 0.275% per-trade return clears the 0.15% USD-M floor but PF 1.03 shows the gross edge barely exceeds costs. Edge is regime-confined: all profit is in the calm vol tercile (+24.1%) while the normal (-0.79%) and stressed (-9.9%, Sharpe -0.42) terciles LOSE, and the book loses in the three most recent full years (2023/2024/2025) — QA's calm-only concern is borne out on full history. This is the single-name pure-OHLCV dual-timeframe momentum-confluence clone family (L82/L93); it sits far below the sole surviving instance (ADA Binance USD-M, Sharpe 1.60 / PF 2.19 / DD 10.7%), and for this family the base backtest is the ceiling — the 3-phase deflation only lowers it. Tuning 3 parameters cannot lift a PF-1.03 signal whose edge exists in only one vol regime above costs; a 2-hour walk-forward would only curve-fit the calm-regime / 2021-2022 bull. abandon_class=negative_expectancy.

LTCUSDT Binance USD-M perpetual, long-short dual-timeframe trend on a deliberately slow clock: 12-HOUR primary bars for the fast trend read and 3-DAY bars (extra leg) for the slow confirmation. Both legs are expressed as volatility-normalised trend slopes (log return over the lookback divided by the realised stdev of that timeframe's returns scaled by sqrt(lookback)), so the two clocks are directly comparable across LTC regimes. The traded signal is the mean of the two slopes multiplied by a smooth agreement weight built from their PRODUCT (w = 0.5*(1+tanh(k*z_fast*z_slow))): sign = agreed direction, magnitude = strength of agreement, recomputed every 12H bar and never gated to a constant. Entry is a small threshold on that signal (long above, short below), so essentially every sign-agreement regime is traded; what varies is SIZE. Position size is risk-first (equity*risk_pct*conviction / (trail_atr_mult*ATR), notional capped at max_notional_frac*equity*conviction) with conviction = clip(|signal|/conviction_full, 0.2, 1.0), so weak/ambiguous trends take ~20% of full size and strongly aligned 12H+3D trends take full size. Exits: ATR trailing stop from the best close since entry, or the agreement signal crossing zero against the position; after a trailing-stop exit the same direction is blocked until the signal crosses back through zero (no immediate re-buy of a stopped-out trend). Three tunables only.

No change required; noted for awareness. If the base template ever relies on its own min_bars_required for dispatch gating, confirm the override value is >= what the base expects.

min_bars_required() and _submit_exit() override framework methods (base_shadow static warnings). Both overrides are deliberate and correct here: min_bars_required returns a valid warmup count used only as an internal entry gate, and _submit_exit calls super()._submit_exit(position) before resetting local trailing state, so base behavior is preserved.

No fix needed; documented for the optimizer's awareness that conv compounds through both terms.

Conviction appears in both the risk-based qty (equity*risk_pct*conv/stop) and the notional cap (equity*max_notional_frac*conv). This double-application of conv is intentional per the hypothesis (weak agreement -> small size) and is not a bug, but it means the effective notional floor at min_conviction_frac=0.2 is ~12% of equity — verify sizing still clears the $5 min-notional at low LTC prices (it does at any realistic equity).

Sandbox regime attribution is fragile: all PnL comes from the calm vol tercile (+29.2%, Sharpe 1.16) while the normal (-6.1%) and stressed (-7.9%) terciles lose, and profit-factor is only 1.12 with Sharpe CI [-0.63, 0.94] straddling zero. The developer's own honest caveat states the per-trade edge on LTC is positive only for slow_lookback>=11 and turns negative for shorter slow lookbacks, so the parameter ridge may not survive out-of-sample. This is a single-name pure-OHLCV dual-timeframe momentum family whose surviving instance (ADA Binance USD-M) sits well above this; the analyst should weigh whether the LTC clone clears the bar on full history. Not a correctness issue — flagged for the backtest review where performance is judged.

Faithful implementation of the stated mechanism: 200 signaled = 200 submitted, balanced 109 long / 91 short, continuous conviction sizing and ATR trailing exit all present; this is a fair test, not a broken/mismatched mechanism.

avg_trade_return_pct 0.275% is nominally above the 0.15% Binance USD-M fee floor, and the 12H clock kept trade count low (200 over 6.5y, commission only 3.07% of gross).

profit_factor 1.0316 sits inside the fee/impact drag band [0.85,1.10] and below the review-guard PF>=1.05 viability floor — gross edge barely clears costs (impact 22.9% of gross).

Sharpe 0.189 with bootstrap CI [-0.599, 0.935] straddling zero — indistinguishable from no-skill; Calmar 0.027, recovery_factor 0.174.

Regime-confined: ALL profit is in the calm vol tercile (+24.1%, Sharpe 0.97) while normal (-0.79%) and stressed (-9.9%, Sharpe -0.42) LOSE — QA's concern borne out on full history.

Losing in the three most recent full years (2023 -0.05, 2024 -2.76, 2025 -12.33); the opposite of a robust trend edge.

Single-name pure-OHLCV dual-timeframe momentum-confluence clone (L82/L93). Sits far below the only surviving instance (ADA Binance USD-M: Sharpe 1.60 / PF 2.19 / DD 10.7%). The base backtest is the ceiling for this family — deflation only lowers it.

LtcDualTimeframeConvictionScaledTren-53d450061b

Clearing the per-trade fee floor isn't enough when PF is ~1.03 and the edge lives in a single volatility regime.

Abandoned at the pre-optimization backtest-review gate: PF 1.03 below the 1.05 floor, Sharpe CI straddling zero, and profit confined to the calm vol tercile while normal/stressed terciles and 2023–2025 all lose.

A long-short, single-instrument LTCUSDT.BINANCE USD-M trend strategy pairing a 12H primary clock with a 3-day trend confirmation, sizing position notional continuously by the strength of dual-timeframe trend agreement and exiting on an ATR trailing stop.

Over a 200-trade, 6.5-year backtest it returned 11.91x total (CAGR 52.5%), but Sharpe was only 0.189 (CI [-0.599, 0.935]), profit factor 1.032, Calmar 0.027, max drawdown 19.3%; avg per-trade return 0.275% with impact 22.9% of gross.