Eric observed that most of our triplet multiplets had signals with widely varying periods (i.e. the spacing of the triplets), which were unlikely to represent the same source.

So I added a processing step to limit triplet multiplets to periods within 1% (for barycentric) or 10% (for non-barycentric). This works as follows:

- given a triplet multiplet, find the two signals that are closest in period.
If these are outside the limit (1% or 10%) discard the multiplet.
- go through the other signals in the multiplet and discard those that
are more than the limit away from the mean of the two closest ones.

This is obviously a kludge. For example, if the multiplet contains several close pairs, it should be split into several multiplets. We discussed this, and concluded that (at least for triplets, maybe other types) we need a multiplet-finding algorithm based on cluster analysis. We have started a conversation with experts in this area.

-- D