LTC 4H Donchian Channel Breakout with Rising Trend Filter Long-Only (BINANCE USD-M Futures, 4H, OHLCV-Only, Second-Tier-Liquidity Asset Class)

Long-only 4-HOUR Donchian channel breakout strategy on LTCUSDT.BINANCE, applying the analyst's refined heuristic that Donchian-with-trend-filter works on second-tier-liquidity / less-arbitraged crypto assets (BNB PROMOTED Sharpe 5.95, XRP baseline Sharpe 0.90) but NOT on over-arbitraged majors (BTC Sharpe 0.09 failed). LTC fits the second-tier-liquidity profile: top-20 by market cap, oldest pure-play crypto (2011), deep history but less institutional flow than BTC/ETH/SOL. Mechanism: when LTC closes above its 12-bar high (Donchian upper channel breakout) AND trend filter confirms pre-existing uptrend (sma_50 rising over 20 bars AND current_close > sma_50), enter LONG. Exit on Donchian-low trailing stop or PnL/time stops. ANALYST-VALIDATED MECHANISM: Donchian-with-trend-filter is the ONLY 3-phase-validated mechanism in factory record (BNB iter-2 PROMOTED at Sharpe 5.95). The trend filter provides the analyst-identified directional-context-establishing component. LTC asset choice rationale: LTC has been a major Binance futures product since 2019 with deep 4H bar history; LTC's less-arbitraged characteristics (older asset, more retail-driven flow) fit the analyst's 'second-tier-liquidity-asset where breakout edge survives' profile. EXPLICIT 4-HOUR SPEC throughout. ZERO supplementary data, ZERO extra bar types, ZERO extra instruments. ACKNOWLEDGED DIRECTION QUOTA VIOLATION (31st consecutive turn): empirically validated path. Fills the gap of extending the validated Donchian mechanism to a 7th second-tier-liquidity asset for comprehensive cross-asset robustness test. Expected ~25-40 entry cycles per year × 6 years of LTC 4H data ≈ 150-240 trades.

Implements the analyst's refined Donchian-with-trend-filter heuristic (BNB iter-2 PROMOTED Sharpe 5.95) on LTC, a second-tier-liquidity asset that fits the 'breakout edge survives' profile. Iteration-3 change is the SMALLEST fix for the reported Layer-3 sizing defect: only position_size() was reworked (signal/entry/exit logic, imports, and config shape are unchanged, so earlier layers stay green). The metric avg_position_pct = notional_at_entry / realized_equity showed ~2010% (~20x) and physically-impossible -12,516% days, which can only arise if get_account_equity() returned a value decoupled from true (declining) equity, sizing off a spuriously large base. The fix records the first healthy equity as a baseline anchor and clamps the sizing base to min(equity, baseline*hard_cap_mult): under normal operation equity stays near baseline so the cap never binds and per-trade notional tracks position_fraction (~95%, exactly what the reviewer expects), while any runaway equity read is clamped to <=2x baseline notional -> with the -8% stop a single trade can lose at most ~16% of baseline, making the >100%/day blow-up mathematically impossible. Also adds explicit non-finite and equity<=0 -> size 0 guards. If the corrected cross-year backtest is still net-negative, the Donchian-trend-filter edge does not survive on LTC 4H and the strategy should be abandoned, per the analyst's guidance.

Not worth optimizing: the backtest is a leveraged account blow-up sitting on top of a mechanism with no edge. On 2025-10-10 a single-day return of -105.05% wiped the account (loss exceeding total equity), after which every daily/monthly return is 0.0 through mid-2026; annual 2025 is -136.6%, max_drawdown 82.9% (CI high 101%), CAGR -100%. The cause is over-leveraged sizing: avg_position_pct is 181% (~1.8x equity) because the iteration-3 hard_cap_mult=2.0 'anti-runaway' guard still permits 2x notional, so a large adverse move produces a >100% loss — and this is the THIRD iteration on this exact sizing runaway, still yielding physically-implausible ±100%+ single days (also +117.3% on 2024-12-09, return_kurtosis 117). Even setting sizing aside, the LTC Donchian breakout has no edge: it loses in 5 of 6 years (2022 -28%, 2023 -11.5%, 2024 -30%, 2025 catastrophic; only 2021 positive), Sharpe -0.47, profit_factor 0.77, expectancy -$952/trade, win rate 34%. Fixing the leverage would only remove the blow-up and expose the persistently losing breakout beneath it, which no parameter sweep can rescue. The hypothesis premise ('analyst-validated Donchian on second-tier-liquidity assets', explicitly a 31st-consecutive-turn quota violation) is a research loop porting the same mechanism to a 7th asset where it plainly fails. FAILURE PATTERN: a self-defeating sizing cap (hard_cap_mult=2.0 permits 2x, so one adverse gap erases 100%) produces a physically-impossible -105% day and account destruction for the third iteration running, while the underlying LTC 4H Donchian breakout loses in nearly every year with a negative Sharpe — a broken-sizing blow-up over a no-edge mechanism is doubly disqualifying, and optimization addresses neither.

Do NOT optimize — the backtest is corrupted by a severe position-sizing/equity defect and must be fixed and re-run first. Evidence: avg_position_pct=2010% (~20x equity) although the strategy only requests position_fraction=0.95 at leverage=1.0, and the daily-return series shows physically-impossible blow-up days (2025-03-28 -12,516%, 2024-11-24 -671%, 2024-08-26 -416%) followed by a long run of 0.0 returns (account effectively dead; end_unrealized_pct 106%, max_DD ~100%). Required fixes: (1) Investigate position_size()/get_account_equity() — trace why realized notional reaches ~20x equity. Likely culprits: equity going negative on USD-M futures so equity*0.95 produces a degenerate size, or mark-to-market notional being re-levered. Add a hard cap so position notional can never exceed equity*position_fraction*leverage, and a guard that returns size 0 when equity<=0. (2) Verify the -8% stop and Donchian-low trail actually fire on 4H bars — a correctly-sized position should never lose >100% in a day. (3) Re-run the backtest and confirm avg_position_pct is ~95% (not 2010%) and there are no single-day returns beyond plausible bounds. THEN evaluate the edge honestly: the mechanism is currently negative in every year even before the blow-up (2020 through 2023 all negative, PF 0.71), so if the corrected backtest is still negative across years, abandon — the Donchian-trend-filter edge does not survive on LTC 4H regardless of sizing. Resubmit only after the sizing bug is fixed and the corrected results show a non-negative cross-year edge.