Classification: (a) concrete trading idea. Source: QuantConnect Strategy Library 'Momentum-Short Term Reversal Strategy' (after Quantpedia). The source is a dollar-neutral cross-sectional book over all NYSE/NASDAQ stocks: rank by 12-month return, and within the winner tercile go long the 15 names with the LOWEST ratio of last-month geometric return rate to 12-month geometric return rate (winners that are not in the final, accelerating stage of overreaction). Within the loser tercile, short the 15 names with the highest ratio. Rebalance monthly, 50/50 long/short. It CANNOT be run as written here. Lesson L137 bans multi-leg neutral books (0 survivors in 150 multi_instrument runs). L151 bans hand-typed universes. L128 bans USEQ daily-bar books (35-500 lifetime rebalances, and deflation strips them). The source's 'all stocks' universe also depends on survivorship. The mechanism that carries over is the ACCELERATION SCREEN on momentum: follow the trend, but not while the recent return rate is running ahead of the trend's own average rate (a blow-off), because that is where short-term reversal eats the momentum payoff. It is restated as a SINGLE-INSTRUMENT time-series rule on the asset it is measured on (no signal-on-A/trade-B, L143). Crypto momentum decays faster than equity momentum, so the horizons are compressed: the '12 months' becomes 60 days (360 4H bars) and the 'last month' becomes 5 days (30 4H bars), keeping roughly the source's 12:1 ratio. Instrument selection (L122: stated explicitly): BNBUSDT.BINANCE USD-M. It has multi-year Binance history from early 2020 at 4H (about 6.5 years, ~14k bars). It is a top-5 liquid perp with lower BTC beta than alts. It is not BTC (62% of recent corpus share) and not ETH (the most crowded non-BTC name in the current pipeline). The mechanism is price-only, so it stays on Binance rather than HL (L168). The 4H cadence was chosen to reach >=150 lifetime trades; daily bars would give about 70. Declared flat state (L150): flat whenever the screen says the move is overextended or the trend is too weak. Expected time in market is about 45-55%, flat about 45-55%. Pre-registered comparison, required by L130 because the screen is effectively a gate: the ungated variant (plain sign of the 60-day return, same exits) must be run as a SEPARATELY scored control. The hypothesis is FALSIFIED if the screened version does not beat the ungated control on OOS Sharpe AND on avg trade return. The screen is the whole claim, so it must earn its place. Point-of-caution note on the source code: its loser leg takes the highest GARR ratio with a NEGATIVE denominator, which actually selects accelerating losers and contradicts the stated 'less likely to be in final overreaction' logic. This restatement applies the text's logic symmetrically (short only DEcelerating declines, skip capitulation legs) and does not copy the code's sign quirk. Risk profile: 1x-1.5x leverage, vol-scaled size (target 2% equity daily-vol contribution), max 30% of equity at risk-weighted notional, and an ATR stop capping loss at about 2.5% of equity per trade.