BNB Price–Open-Interest Confirmed Trend Continuation, Long-Short (Single-Instrument BNBUSDT.BINANCE USD-M, 4H Bars — Take Directional Trend ONLY When Open Interest is EXPANDING With the Move = Fresh Committed Leverage, Stand Flat When OI Contracts = Short-Covering/Washout Without Conviction, ATR-Trailing Exit, 3-Parameter)

A LONG-SHORT, single-instrument, single-venue trend-CONTINUATION strategy on BNBUSDT.BINANCE USD-M perpetual (4H bars) whose signal is the JOINT state of price direction and OPEN-INTEREST change — a positioning tell, not a pure-OHLCV filter. The core idea (which is the mechanism that PASSED as SolPriceOpenInterestQuadrantTrendLS and is live as DogeOpenInterestConfirmedTrendContinuationLS): a directional move is only worth riding when it is accompanied by EXPANDING open interest, because rising OI means new capital is committing on that side and the move is being funded by fresh leverage rather than by position-closing. When price trends up AND OI is rising, go LONG (genuine accumulation). When price trends down AND OI is rising, go SHORT (fresh shorts pressing, genuine distribution). When OI is FALLING, the move is being driven by de-risking (a rally on falling OI is short-covering; a selloff on falling OI is a long-liquidation washout) — there is no committed conviction behind it, so STAND FLAT. This is explicitly the OPPOSITE polarity of the abandoned 'BTC Price–OI Divergence Reversal' (which FADED moves on falling OI and died in coding); here we only take CONFIRMED continuation and never fade. Routed to BNB — a highly liquid major whose price series is essentially absent from the factory's momentum book (dominated by BTC/ETH/ADA/SOL) — and to Binance USD-M specifically because the mechanism genuinely requires the reliable, densely-collected OI feed that only Binance provides in the catalog (HL/COIN-M OI coverage is thinner and HL multi-history is timeframe-capped). 3 parameters only, to fight the overfitting that is the #2 killer here.

Direct implementation of the hypothesis's price/OI quadrant logic with the exact polarity requested: continuation ONLY on expanding OI (both directions), flat on contracting OI, never a fade. It reuses the mechanism and code shape that passed as SolPriceOpenInterestQuadrantTrendLS / DogeOpenInterestConfirmedTrendContinuationLS, minus the daily confirmation leg, because this hypothesis is explicitly single-instrument with 3 parameters (mom_lookback, oi_lookback, trail_atr_mult, bounds declared in _param_bounds). Routed to Binance USD-M as required: shorts need a MARGIN venue and the OI feed the mechanism depends on is densely collected only for Binance (data/supplementary/metrics/BNBUSDT and open_interest_hist/BNBUSDT both exist). leverage stays 1.0 - the edge needs the futures venue for shorts and the OI feed, not amplified notional, so no unused leverage is declared. OI is read strictly at-or-before each bar close via a forward-only pointer (no lookahead, O(bars + readings) so the 300s smoke budget is safe), and 30-minute thinning keeps the merged series small. Entry is a 2-condition conjunction on a 4H series with an ATR-normalised threshold, which fires on the order of tens of times a year rather than the near-mutually-exclusive 3+ gates that cause zero-trade sandbox failures.

Negative expectancy on a decisive 487-trade / 2400-day (6.5-year) sample: profit_factor 0.806 (<1, gross-NEGATIVE before costs — commission is only 9.48% of gross, so this is a losing signal not fee-fragility), avg_trade_return_pct -0.126% (below ZERO, not merely below the 0.15% Binance USD-M floor), expectancy -$86.66/trade, total_return -42.1%, Sharpe -0.145 with CI [-0.898, 0.562] straddling zero. Optimization cannot lift an edge that is negative before costs. The code faithfully implements the OI-confirmed continuation hypothesis (487 signaled = 487 submitted, balanced 255 long / 232 short, ATR trailing + OI-contraction + trend-flip exits) so this is a fair test, and QA's concern is confirmed on full history: the OI-expansion gate adds no edge on BNB. The failure is regime-WIDE — the book loses in all three vol terciles (calm, normal, stressed) and in every calendar year 2021 through 2026 — the opposite of a robust continuation edge. Tuning 3 parameters cannot rescue a PF-0.806 signal that is gross-negative in every regime; not worth 2 hours of walk-forward. abandon_class = negative_expectancy (premise falsified on BNB).

Long-short 4H trend-continuation on BNBUSDT.BINANCE USD-M perpetual whose entry is gated on the JOINT state of price direction and OPEN-INTEREST change. The signal is a continuous ATR-normalised momentum read, trend = (close/close[-mom_lookback] - 1) / (ATR/close), recomputed every bar. Open interest is merged point-in-time from the `metrics` (deep 5-minute Binance Vision series) and `open_interest` (recent fapi mirror) supplementary feeds and reduced to oi_change = OI(t)/OI(t - oi_lookback bars) - 1. Entry: trend > entry_thresh AND oi_change > oi_expand_thresh -> LONG (fresh leveraged longs funding the rally); trend < -entry_thresh AND oi_change > oi_expand_thresh -> SHORT (fresh shorts pressing). Any falling-OI state is a hard no-trade: a rally on contracting OI is short-covering and a selloff on contracting OI is a long-liquidation washout, neither carries committed capital, so the strategy stands flat and never fades. Exits: ATR trailing stop (give back trail_atr_mult x entry-ATR from the best close since entry), OI contraction beyond oi_contract_thresh, or the trend flipping against the position. Sizing is risk-first (risk_pct of equity across the trailing-stop distance) with a hard notional cap, so losses shrink size instead of compounding it. If no fresh OI reading exists for a bar the strategy opens nothing rather than degrading into a price-only trend follower.