In Fig. T1 we compare the top and diboson backgrounds after detector simulation. The top backgrounds peak below the excess region (indicated by the vertical line).
We compare the MJJ distribution for single top + t-tbar samples generated with PYTHIA and MC@NLO. MC@NLO only generates the t-channel but as can be seen from the rightmost plot of Fig. T2 which compares the single-top MJJ distributions this makes little difference.
The number of top events predicted using PYTHIA (scaled to the NLO cross sections) is 1233 which compares very well with the estimate from MC@NLO of 1261.
We have also fit the data when we artificially increase the expected t-tbar contribution by 50%. This is an unreasonably large increase in the background, but nevertheless the excess remains with a statistical significance of 3.9 standard deviations.