"Fat" clients need to talk to many "RPC"s at once and get final vote on "two phase" commit to their transaction.
Client is like Builder in Proposer builder separation.
Each representative stores all data for all accounts. They do DA.
Here I present a formal Quint specification for its consensus protocol, model-checked under a Byzantine fault model. Safety was found to be preserved within the fault-bounds and under a constant weight model, and the possibility of FastPay style lockouts was reproduced as expected.
The direction for future research depends on the direction Keeta takes; checkpoints and epochs (similar to Sui) are the features to watch there.
Prerequisite paper: https://keeta.com/whitepaper.pdf
Google marketing slop: https://cloud.google.com/blog/topics/financial-services/how-...