The abstract makes one testable claim for a tradeable rule. Most risk premia earn their Sharpe by carrying negative skew, which is crash risk. Trend following is the exception: it earns positive excess return WITH positive skew. We test the crypto version of that claim on ETHUSDT.OKX (OKX USDT perpetual swap, 4-hour bars, data from 2019-12-25, about 14.8k bars). OKX is 0.3% of the corpus and is a pure-OHLCV venue, as lesson L160 requires. Signal: s_t = (close_t - EMA_100(close)_t) / (close_t * sigma_100_t), where sigma_100 is the rolling standard deviation of 100 four-hour log returns. So s is the distance from trend measured in per-bar volatility units. The strategy is FLAT whenever |s| has not exceeded 2 since the last exit. That deadband is the flat state required by L150, not an extra regime gate. It is long when s crosses above +2 and short when s crosses below -2, and it exits when s crosses zero. Beyond return, the hypothesis also predicts the SHAPE of the trade distribution: a low win rate (~30%), a right tail of large winners, and positive skew. That is the paper's diagnostic. If the backtest shows a profit but negative skew, the premise is falsified even if Sharpe is positive. Point-in-time pre-study (close-to-close, 0.10% round trip deducted, done before any code): 259 trades, +2.17% mean net per trade, t = 2.8, trade-return skew +2.9, win rate 32%, 7 of 7 calendar years positive. The result is robust across a 3x3 grid: EMA span 20/50/100 x threshold 1/2/3 gave t between 1.7 and 2.9 on every 4H cell, with skew positive in all of them. We pre-register the centre-right cell (100, 2) and will NOT optimise it away. Selection choices, stated per L122: ETH because it is the most liquid OKX swap with 2019+ history (BNB OKX starts 2022, others ~2 days of data); OKX to spend the under-used venue; 4H because 1D gives only 35-135 trades, below the sample floor.