Option A is correct. Screening relies on simple rankings of alphas and places assets into categories such as buy, hold, and sell. While it ignores the exact magnitude of the alphas (relying merely on their ordinal rank), it helps limit transaction costs because maintaining a "hold" list minimizes excessive turnover and trading.

Option B is incorrect because stratification is designed to match portfolio weights to benchmark weights across specific risk dimensions, not necessarily to overweight low-risk or underweight high-risk categories. Option C is incorrect because it describes stratification, which requires dimensions to distinctly and exclusively partition stocks. Linear programming can handle overlapping dimensions or constraints. Option D is incorrect because quadratic programming, which requires a full covariance matrix and expected returns (alphas) for all assets, is actually highly sensitive to noise and estimation errors, increasing the potential for poor calibration (error maximization) rather than minimizing it.