Pretty interesting reply thread btw @sam_battenally and @aeyakovenko about latency vs validator distro/MCP.

Here's the thinking:

If you want to minimize your round trip time (RTT) you need to geographically colocate your node with a proposer.

In a single sequencer or geo-concentrated validator setup, this is trivial and extremely fast since you just colocate with the sequencer/validators and you're done.

In a geo-distributed single leader setup your average RTT would be the average stake-weighted RTT between all validators, which would end up being pretty slow no matter where you are.

But with MCP you could get consistently low RTTs by colocating in an area where there is a concentration of validators & stakeweight such that you have a high likelihood of there being at least one proposer there per slot.

This seems to get you the best of both worlds: geographically decentralized validators, good CR, low latency.

However I'm still confused how conflicting txs confirmed by two different proposers in MCP get resolved. It seems that the guarantee that you get from a proposer in MCP is fundamentally different from what a single sequencer or single leader gives you.