multcompare contradicts friedman test

I used a friedman test on a (24,2) matrix, MatForAnova, whose columns represented the factor (2 levels) of interest and whose rows represented the other factor present in the experiment (3 levels, 8 observations per level). I did

[P_friedman1,TABLE,STATS_friedman1] = friedman(MatForAnova,8,'off');

which yielded:

P_friedman1 = 2.1921e-005

TABLE =

    'Source'         'SS'          'df'    'MS'          'Chi-sq'     'Prob>Chi-sq'
    'Columns'        [408.3333]    [ 1]    [408.3333]    [18.0147]    [2.1921e-005]
    'Interaction'    [ 28.6667]    [ 2]    [ 14.3333]           []               []
    'Error'          [     583]    [42]    [ 13.8810]           []               []
    'Total'          [    1020]    [47]            []           []               []

STATS_friedman1 =

       source: 'friedman'
            n: 3
    meanranks: [5.5833 11.4167]
        sigma: 4.7610

I then did

multcompare(STATS_friedman1)

which yielded

ans = 1.0000 2.0000 -13.4522 -5.8333 1.7856

The two columns overlapped in the graph that multcompare generates: only blue and gray, no red different groups. Meaning that the two columns were not seen as different by multcompare, despite the low pval=2.1921e-005 as seen in the table. What is wrong with this picture? Help. Neurobiologically yours, E