Till now, families with different perturbation parameters were processed independently. Unperturbed ELS's were treated as perturbed ELS's with zero perturbation parameters. However, the only purpose of perturbed ELS's is to be compared with "true", unperturbed ELS's. This is the differential measurement. Perturbed ELS's should be very close to true ELS's in all statistical parameters. Their only distinction is that they are not exactly equidistant! What of it? It should not matter. Strangely, it matters, which is the WRR effect.

This last step of the computation is very simple. The measure of proximity for unperturbed ELS's (perturbation parameters (0,0,0)) is compared with the measure of proximity for each other combination of perturbation parameters. That is, given the "Data 5" file,

[  -2 -2 -2  2.377977e+05 ]
[  -2 -2 -1  3.450332e+05 ]
[  -2 -2  0  2.672815e+05 ]
[  -2 -2  1  3.251481e+05 ]
  .   .   .   .   .   .
[   0  0 -1  2.727624e+05 ]
[   0  0  0  1.488219e+06 ]
[   0  0  1  3.029864e+05 ]
  .   .   .   .   .   .
[   2  2 -1  9.820894e+04 ]
[   2  2  0  8.988155e+04 ]
[   2  2  1  5.933152e+05 ]
[   2  2  2  2.263009e+05 ]

Though, some numbers may be absent (since some families may be empty, as was noted in "Step 4-5"). We count all available numbers, and how many of them exceed the "central" number. Sometimes the very "central" number is missing (no "true" ELS matches one of the two given words). In such case no result is produced.