Note on threshold for including columns in summed running mean track:

In the default 'Nucleotide-by-nucleotide' and zoomed-in 'Nucleotide-by-nucleotide' plots, the running mean of the log likelihood scores summed over the input phylogenetic tree (i.e. the input pairs file) skips regions that are gapped (or have ambiguous nt codes) in any of the sequences in the pairs file. On the 'Redraw plots' page, you have the option to extend this track into gapped regions, provided the summed divergence of the contributing sequence pairs in the region is >= (threshold) x (maximum divergence). (The maximum divergence is just the summed divergence in ungapped parts of the alignment. 0.01 <= threshold <= 0.99; default = 1.)

Where some sequence pairs are omitted due to gaps (or ambiguous nt codes), but the summed divergence is still >= (threshold) x (maximum divergence), the log likelihood scores are scaled by (maximum divergence) / (local divergence), before taking the running mean. (The local divergence is just the summed divergence in that particular column of the alignment.) This scaling is applied so that you don't get dips in the running mean just due to some sequences being gapped. Be aware, however, that in partially gapped regions the signal-to-noise ratio will be lower compared with neighbouring ungapped regions.

The values in the raw summed-over-tree likelihood scores track are not scaled by (maximum divergence) / (local divergence), because these scores represent the actual likelihoods, based on the information in each column - i.e. if some sequences are gapped (or have ambiguous nt codes) then there is less information so, all other things being equal, the likelihood ratios are lower.