The way I see it, the blog post focuses on the “false impression” one might get after reading the Avalanche whitepaper. Indeed, one shouldn’t believe that it is a plug & play algo. The rebuttal imho is the Avalanche mainnet going live & the roadmap (with pruning) in the works.
Well, almost two years since that post have passed, and the main net has launched, and it is obvious that a working model has been produced from the paper.
IANACT (I Am Not A Consensus Theoretician), but as far as questions raised in the paper go the answers are: No, transactions don’t all reference the genesis, because of the UTXO model. And metastability of the consensus is not a problem in practice, as even if it were practically gameable by malicious actors, all they would be able to do is stall their own attempts at double-spend, so who cares?