The accusation, in a nutshell, is that conflicts that appear on the DAG on parallel blocks are only resolved „locally“ by nodes, but not in „global“ consensus, and hence Kaspa is „not really secure“.
So let me make this poignantly clear: there is nothing „local“ about Kaspa’s consensus. The GHOSTDAG protocol orders blocks in consensus, which means all nodes reach the same ordering. Once this ordering is agreed upon, all nodes resolve conflicts the same way. The confirmation time is essentially how long you have to wait before you know with very high confidence that (assuming honest majority) the block containing your transaction will not be reordered, that it is included in the part of the ordering that is „agreed upon“. This is exactly the same security as Bitcoin’s Nakamoto consensus, the only difference is that the word „orphaned“ is replaced with the word „reordered“. In the GHOSTDAG paper I provide a rigorous mathematical proof [1, Appendix A] not only that the ordering converges in consensus, but that it converges rapidly: the time this takes is only governed by network latency, regardless of block production rates. This is Kaspa’s instant confirmations: a rapid and global agreement on the DAG ordering.
A less formal but more approachable explanation is available in the GHOSTDAG 101 workshop video [2].
There are only two ways to dispute this security claim: either point out an error in the mathematical analysis, or describe an explicit attack. Making up narratives and semantics is just cowardly raising „doubts“ about our analysis without actually addressing it or pointing out any flaws. (In contrast, when I criticized the confirmation times and scalability of another project, I provided the math to back it up.)
Disappointingly, this false narrative originates from a main dev of a „competing“ project, who is often referred to as a „Bitcoin OG“, leaving little doubt that this mislead is intentional. He even leveraged this vacuous claim to publicly call Kaspa’s security a „marketing gimmick“ in front of his community. I would expect more from a person with actual pedigree who is currently leading a project, but I guess that’s why we can’t have nice things.
So whenever you see anyone trying to FUD Kaspa with this „client consensus“ vs. „global consensus“ nonsense, know that the person you are arguing with is either disingenuous or gullible.
[1] eprint.iacr.org/2018/104.pdf
[2] https://youtu.be/nhI2zo44dfc?si=-qJMIAM9hUoZ_UQx