Factor returns obtained from characteristic-based sorting are not only a function of their signal, but also on the procedure and choices of the factor construction method. Since there is no consensus on which choices to make when constructing factors, researchers face a ad-hoc number of degrees of freedom, potentially allowing for p-hacking. We focus on a wide-range of construction choices that are widely used in financial research on cross-sectional equity factors, and on purpose data mine over 500 different versions of each factor in our sample. We find that Sharpe ratios exhibit large and significant variation within a factor due to construction variation. Consequently, this variation impacts model selection exercises. We attribute the variation in factor performance, across construction choices, to variation in liquidity, factor breadth, and diversification.