I often see that when putative oversplitting occurs there are "duplicate spikes" that get assigned to both clusters, giving a very large peak near 0 in the cross correlogram (see example below). This makes it difficult for SLAy to find such merges given the refractory period violation penalty (indeed, it doesn't find the one below unless you set k2 to 0). We should probably filter out duplicate spikes before calculating the refractory period penalty. Also, these duplicate spikes should be removed after automerging (regardless of the algorithm used).
I often see that when putative oversplitting occurs there are "duplicate spikes" that get assigned to both clusters, giving a very large peak near 0 in the cross correlogram (see example below). This makes it difficult for SLAy to find such merges given the refractory period violation penalty (indeed, it doesn't find the one below unless you set k2 to 0). We should probably filter out duplicate spikes before calculating the refractory period penalty. Also, these duplicate spikes should be removed after automerging (regardless of the algorithm used).