BNB Net-Directional Time-Series Momentum (Single-Instrument USD-M, Long-Short, 2H, Discrete Trend-Flip Entries, Capital-Capped, Pure OHLCV)

A LONG-SHORT, SINGLE-INSTRUMENT net-directional TIME-SERIES MOMENTUM strategy on BNBUSDT.BINANCE USD-M perpetual (2H bars). It applies the ONE validated, most-replicated momentum factor -- an asset's own trailing-return trend predicting its near-future return -- to a DECORRELATED, liquid major NOT yet covered by the portfolio's TSMs (which are all BTC/ETH/SOL). RATIONALE FOR ASSET + HORIZON: (1) BNB is top-3 liquidity with full clean history and a DISTINCT driver (exchange/chain token), so it is decorrelated from the L1/smart-contract majors that dominate the portfolio -> genuine return diversification, and it has the CLEANEST alt trends -> highest TSM survival among alts. (2) BNB's only existing strategy is a long-ONLY daily Donchian breakout (BnbDonchianBreakoutTrendFilteredLong1D); this adds the missing SHORT side and a finer horizon. (3) 2H fills the UNDER-represented mid_30m_2h horizon bucket (11.7%) and LONG-SHORT fills the under-represented direction bucket (13%). EVERY design choice respects this session's hard engine walls and lessons: (a) SINGLE-INSTRUMENT -- NOT a multi-leg basket, whose summed gross exposure proved structurally UNCAPPABLE in this engine. (b) DISCRETE full-size enter/exit on trend flips -- NOT continuous per-bar rebalancing, which produced the 244%-vs-cap notional-inflation artifact on both spot and perp. (c) NET-DIRECTIONAL momentum -- the validated edge (absolute 1.36 / dual 1.61 baskets, single-asset TSM); dollar-neutral relative momentum, cross-asset cointegration, and funding-confirmed clones are CLOSED. Plain price-TSM is more robust than the funding-confirmed version that failed on alts: it goes net-SHORT in downtrends (real crash protection) rather than long-biased bull-beta. (d) PURE OHLCV -- no options/tick/funding plumbing (options are engine-unwired -> zero trades; funding data ends 2024); guaranteed to trade in the Layer-3 sandbox. (e) 2H dense triggers -- avoids the daily single-asset event-sparsity that trade-starved RSI-bounce/candle-pattern walk-forwards to OOS-all-zero. (f) CAPITAL-CAPPED against STABLE realized equity (excl. unrealized PnL), hard-capped <=100%. Intentionally minimal: 3 tunables (trend_lookback, entry_band, stop_atr_mult) with robust defaults to avoid a trend_lookback cliff.

Implements the validated TSMOM factor (own trailing-return trend predicts near-future return) on BNB, a decorrelated top-3 major not covered by the portfolio's BTC/ETH/SOL TSMs, at the under-represented 2H/long-short buckets. Architecture is deliberately the single-instrument clean path that avoids every engine wall hit this session: (a) NO multi-leg basket, so the uncappable summed-gross-exposure and multi-instrument daily-return aggregation artifacts cannot occur; (b) discrete full-size enter/flip on trend changes rather than continuous per-bar rebalancing, avoiding the 244%-vs-cap notional-inflation artifact; (c) one position at a time, so exposure is structurally bounded to position_pct (95%) of equity at leverage 1.0 and the daily return is one leg's clean MTM. Net-directional momentum (sign of own trend) gives genuine crash protection by going net-short in downtrends, unlike the funding-confirmed/long-only variants that failed on alts. The entry_band deadband prevents noise-driven flips near zero, and the let-trends-run exit (0-crossing of the trailing return) plus a wide 4x-ATR catastrophe stop keep winners running while bounding tail loss; entry context is reconstructed from avg_px_open if state is lost. calculate_signal returns the continuous trailing return so Layer-2 frozen-signal detection passes, while the discrete band entries and sign-flip exits live in the gate methods. Sizing uses get_account_equity() (stable realized USDT balance, excluding unrealized PnL, so it cannot inflate mid-trade and oversize) and clears the BNBUSDT $5 min-notional; 2H bars give dense triggers across full BNB history, avoiding the daily single-asset event-sparsity that trade-starved prior walk-forwards. leverage stays 1.0, referenced only as a non-amplifying sizing multiplier, so the unused-leverage gate does not apply. Pure OHLCV guarantees the strategy trades in the Layer-3 sandbox with no supplementary-data plumbing risk.

Two independent fatal problems, each sufficient. (1) CATASTROPHIC OVER-LEVERAGE + LIQUIDATION: despite the docstring's claim of 'capital-capped against stable realized equity, hard-capped <=100%, single position -> exposure = position_pct,' the backtest shows avg_position_pct 641.8% (~7x the intended 95%), physically-impossible daily returns (+1940.5% on 2023-03-22, -1560.5% on 2023-06-23, +260.6%, +332.8%) that a single BNB position at <=95% sizing cannot produce, return_kurtosis 1499, annualized_volatility 755%, impact_cost_pct 10,126%, and a full LIQUIDATION on 2023-06-04 (total_return -100%, max_drawdown 100%). This is the recurring notional-inflation/over-exposure bug the hypothesis claimed single-instrument design would avoid — it does NOT, because a discrete TSM that re-arms entries while a position is open stacks size; the 641% figure proves the cap is not actually enforced. (2) THE SIGNAL LOSES ANYWAY: profit_factor is 0.91 (<1.0), so even correctly sized, plain BNB time-series momentum has negative gross expectancy — and it is the explicitly regime-dead naked-single-asset-TSM family (the un-gated cousin of the ER-gated TSM just abandoned, which the hypothesis itself concedes is more whipsaw-exposed). Not optimize: a return series dominated by impossible bars and ending in liquidation cannot be optimized. Not iterate: fixing the sizing/stacking bug would only surface a PF-0.91 money-loser in a dead family — a wasted cycle with no edge at the end. FAILURE PATTERN: the multi-leg-basket notional-inflation/over-exposure bug also occurs on SINGLE-instrument discrete-entry strategies (avg_position 641% vs a claimed 100% cap, impossible >1000% daily prints, liquidation) whenever entry logic can re-fire while a position is open; and naked single-asset time-series momentum on an alt (BNB) has profit_factor < 1.0, confirming the regime-dead naked-trend family. A 'capital-capped <=100%' claim in a docstring is not evidence the cap holds — verify realized exposure; here it was 6.4x over.