I am not 100% sure, but I think it's MMseqs behaviour. Linclust will use the parameters you gave it to create clusters, and put sequences together based on those parameters, but that won't mean that there won't be sequences in other clusters that wouldn't match your cluster with these parameters, if this makes sense. It means that you will never get a 0.79 identity sequence in the created cluster .The 'cluster' option is slower, but is doing a bit better job regarding this, but you ll still find matches across families

Thanks!! @Evangelos Karatzas I found an issue describing that during cluster workflow because of the iteratively behavior of MMseqs2 "it can happen that some members that were already close to a clustering do not fulfill the clustering criteria anymore" and they added the an option --cluster-reassign but I do not know if it can have significant power in this case. Its just that because of the way the clustering method works, I was expecting to have unique cluster fulfilling the min identity and cover criteria 🫠