I built a backtesting lab with preregistration, out-of-sample splits, and a cost model measured from real order-book data. Over 3 years of Binance USDⓈ-M futures (12 symbols), I tested ~25 ideas: cross-sectional momentum, funding carry, taker-flow imbalance, liquidation reversal, spot-futures basis, cross-predictability, clock-phase effects.

Consistent finding: several signals have a small positive gross edge (~8 bps per trade), but round-trip taker cost (~15 bps) kills them. Own-asset momentum (14–28d) is the most robust, surviving out-of-sample with modest Sharpe, but net profit is thin after costs.

My measured order-book data suggests maker execution on USDC pairs could cut round-trip cost to ~4 bps, which would flip some signals positive — but maker fills carry adverse-selection risk (you get filled right when price turns against you).

Question for those who've actually traded live on a retail account: is the edge ever in the signal, or is it entirely in execution (maker vs taker, venue, timing)?…