When is proof-of-transactions better than proof-of-work?

A transient blockchain can process more data than a non-transient blockchain. But what if a proof-of-work blockchain was also operating atop a transient ledger? Would it be just as scaleable as Saito? Asking this question forces us to address the fundamental difference between proof-of-work and proof-of-transactions.

We have explained elsewhere that proof-of-transactions splits network fees according to market forces between miners (who provide security), routing-nodes (that route transactions) and full-nodes (that produce blocks). This differs from proof-of-work systems where the network gives all of its fees to miners and expects them to pay for everything. Looked at from this perspective, proof-of-transactions would seem to be a superior version of proof-of-work, but is it necessary to pay for bandwidth separately? Advocates of proof-of-work tell us that market forces take care of this problem in their system as well. A miner that wishes its blocks to propagate quickly, for instance, will pay for the bandwidth to send blocks to its peers. Similar incentives push miners to collect transactions from users. Surely we can assume that proof-of-work is basically the same thing as proof-of-transactions, they insist?

Unfortunately, this is not true.

There are many reasons for this, but the biggest is that as hashing and bandwidth costs grow more extreme miners are subject to perverse incentives that push them to maximize their own income instead of supporting the network. To list a few examples of this, miners have incentives hoard high-fee transactions at scale if doing so gives them a competitive edge. When deploying hashpower, miners have incentives to pool their machines in densely-connected regions since doing so increases their profitability at the expense of their less-connected peers. And when bandwidth becomes scarce miners have incentives to send blocks to their most powerful counterparts first. All of these behaviours systemically impoverishes and drives smaller miners out of the market, leading to greater centralization over time.

Another problem with proof-of-work is that any miner that can free-ride on the network paid for by its counterparts will be able to spend a much higher percentage of its revenue on mining and earn superior returns, money that can be poured back into buying more hash power in the Bitcoin equivalent of a fee-recycling attack. This is a classic collective action problem that can only be solved by miners destroying the open-access properties of the network. In practice, it means miners will be laggard in paying for non-mining costs, and will be OK with a certain degree of under-provision in the public network.

By splitting-up the payments for transaction sourcing, block creation and mining, Saito avoids all of these problems. Saito nodes have no incentive to hoard transactions because they are paid for routing them regardless of where the transaction ends up being bundled into a block. There are likewise no incentives for nodes to form dense mining clusters, since parts of the network that have *just* produced a block are less not more likely to produce the next one. And when bandwidth or mining or transaction sourcing becomes scarce the network can allocate more funds to the behaviour in a dynamic ways that does not require discrimination or collusion by any particular actor.

In short, the proof-of-work mechanism distributes network fees in ways which encourage bad behaviour while the proof-of-transactions mechanism incentivizes optimal economic behaviour. Proof-of-work allows centralizing pressures in fee-sourcing / mining / routing to reinforce each other and make it harder for independent nodes to survive. Proof-of-transactions pits those centralizing pressures against each other, so that routing-nodes act as a bulwark against centralizing pressures in mining.

From this we can conclude that proof-of-work systems are hard to destroy but suffer centralizing pressures over time. Proof-of-transactions systems are easier to attack in situations where hashpower is hard to accumulate, but are more resilient to centralization pressures. The problems with proof-of-work become acute as the costs of non-mining operations grow, and in situations where bandwidth costs or transaction sourcing become first-order costs in running the network, proof-of-transactions is far superior.