As MrBayes searches for trees it saves increasingly likely trees to the tree file (here named "mrtree.t"). Other parameter estimates are saved to the file mrtree.p. At first the likelihoods of the trees increase rapidly. Eventually the increase in likelihood declines, and the likelihoods converge on a steady value (a "plateau"). After that, the saved trees are roughly equally likely, and it is these roughly equally likely trees that we will use to construct a consensus tree. The "burnin" value is the number of trees (not the number of generations) that will be ignored when the consensus is created.