share of samples nearest each class centre, and median distance from the origin (data sits at 3.0)
settings                                 cat    dog    fox    distance
cfg 1, positive any, no negative         0.335  0.328  0.336  3.02
cfg 1, positive any, negative fox        0.335  0.328  0.336  3.02
cfg 3, positive any, negative any        0.335  0.328  0.336  3.02
cfg 1.5, positive any, negative fox      0.427  0.418  0.156  5.00
cfg 3, positive any, negative fox        0.494  0.490  0.016  12.39
cfg 7, positive any, negative fox        0.502  0.498  0.000  32.54
cfg 3, positive dog+fox, negative fox    0.000  0.923  0.077  12.31
cfg 3, positive dog, negative any        0.000  1.000  0.000  3.63
cfg 3, positive dog, negative fox        0.000  1.000  0.000  13.00

cat and dog keep sharing the mass that fox lost, but nothing says the split stays even: it depends on where each path starts.
