Valid inference for regression with best subset selection

Best subset selection is widely implemented in statistical software and is routinely used by practitioners in scientific fields for variable selection. However, classical confidence intervals and $p$-values constructed after model selection generally fail to achieve their nominal frequentist guarant