Two populations can have the same median and average score on fit for a criterion. However, if they have a different distribution, let's say population A has a 'normal' distribution while population B has a U-Shaped distribution, then after the application of a high pass filter with a cutout above the shared median, you will be left with proportionally higher residue of population B than of population A and unbiased sampling of the result will reflect this. Applying a high-pass filter below the median will show an inverse result. A high-pass filter with cut-off at the median will result in a 50/50 result between A and B remainders. As always, statistics do not say anything about any individual in any of the populations. Statistics does also not address the potential underlying causes of the distribution differences. Using statistical arguments to judge individuals is usually referred to as prejudice or discrimination. Many countries have adopted laws and regulations to curb some select forms of discrimination, while permitting others. In the scenario where a high-pass filter far above the median is applied, while at the same time a 50/50 distribution of the outcome is targeted, the observation that, regardless of the mechanisms employed, members of B are actively discriminated against would be correct. Whether that discrimination is desirable or not is debatable. How distributions are realy shaped to a criterion is measurable in theory, but in reality prone to biases especially in highly 'politicized' environments.