[tim333]

Here's kind of how they deal with it from their FAQ

>To allow high scalability a Tempo ledger is split into a very large shard space, allowing a huge degree of concurrency. To avoid a double spend across any of the shards, the shard a wallet lives on is determined by its public key. This makes sure that any spend from a wallet will always start on the same shard. When combined with the logical clocks and gossip, this Tempo to always find the total ordering of related events, allowing double spends to be quickly detected and ignored.

[SI_Rob]

I don't think this actually addresses the problem at all unfortunately, as a gossip protocol that has to flood a global network is going to require, sooner or later the affirmation or rejection of a truth value from all nodes in that network, and it is the accumulation of latency between all of those nodes that becomes the upper bound on transaction frequency.

To have full decentralization you must not bypass, obscure or exclude any cohort of nodes from having a proportionally fair influence over the settlement of a common shared truth value, and this means you run smack into hard limit of geography-scoped cumulative latency as a scalability dead end in one direction, and decentralization dead end in the other.

There's a reason why high frequency trading shops have all settled on the same solution to their own (far simpler) latency-scaling network dilemma, which is to house their nodes as close as geometrically possible to the settlement forum. This is the terminal stability state toward which any purportedly decentralized truth settlement network will progress given enough time, as the people, er, "actors" with the best geography are fated to win the latency leverage race and with this advantage, come to control the definition of in-network truth values.