Strange Splitting

I have been using Kilosort3 on Neuropixel recordings, and downloaded the last version of phy. I need to do the phy extract-waveforms command to be able to look at the FeatureView.

On two different datasets, I have the same issue (with also another issue: I don't see anything on the "AmplitudeView" plot):

When I try to clean a cluster to get a better refractory period for example, once I have selected the "core" of the cluster, I am left only with the minority of the initial number of spikes (for example 485 against 5000 initially).
I did some tests, and in fact, when I select all but one displayed points in the Feature Fiew, I always end up with a new cluster containing 499 spikes. And the last point (also a single trace on WaveformView - further away from the core of the cluster - contains all the remaining spiking events...

Yet, if I then re-iterate the phy extract-waveforms command on the new dataset (with the two new clusters), I can see more than just a single trace, and new points on the FeatureView, but the same problem re-emerge when I want to split again the cluster containing a high number of spikes.

Do you have any idea of what I should change to clean properly and easily these datasets?
Thank you!

Start by reproducing the split behavior with Kilosort3 data after running phy extract-waveforms, comparing FeatureView, WaveformView, and AmplitudeView. Check why selecting displayed points creates one small cluster and assigns the remaining spikes to another; done means cluster splitting retains the expected spikes and the related views remain consistent.