This article analyses the design decisions taken by the TAIKO team in terms of rollup decentralisation. It is based on the recently published rollup decentralisation analysis framework.

Taiko has chosen the sequencer role to be completely permissionless and free for anyone to assume at any time. When constructing the L2 blocks, the L2 nodes exclude/ignore duplicate transactions from multiple sequencers. In a sense, the first sequencer “wins” the profit.

This design decision favours the speed of the network as multiple sequencers at any given time make sure the user transactions are sequenced.

The main tradeoff of this design decision is capital efficiency - the cost of the system (and consequence cost of transaction) and making it profitable for the sequencers to participate even if some/most of their transactions get ignored inside the L2 (due to being sequenced by someone else).

Ultimately, this design decision favours the speed of the rollup over transaction costs.

Taiko has decided to use a sequencer-led leader election with fallback to “free for all” for selecting the prover of a sequence. The sequencers negotiate off-chain with provers in order to “couple” themselves for this sequence. If the prover fails to post a proof in an allotted timeslot, the sequence becomes available for any prover to prove.

The “Leader election” part of the spectrum of design decisions, favours the cost efficiency. In the happy path, no wasted computation is done by the provers, thus ensuring that the system is very efficient in terms of the prover's resources.

The main tradeoff here is with the possibility to stall the network finality (but not the network throughput), if the leader exhibits liveness issues. That being said, the backup of “Free for all” proving ensures that the liveness issues are contained to a single timeslot.

Ultimately, this design decision favours the cost of the network, trading off possible finalization delays.

Taiko has chosen to have a single fork with a merging of competing forks. The protocol has defined the way to merge competing blocks, by allowing the L1 to order the blocks.

Single fork choice in Free for All sees the eligible sequencers competing to be the first one to propose sequence to the L1. It ultimately leads to the highest possible speed of the rollup, but it can lead to capital inefficiency as multiple sequencers might be competing for a single slot. Furthermore, the latency of L1/DA synchronization for sequencers will play a huge role as the proposed sequence by a sequencer might be based on a past fork choice quite quickly. Single fork choice, however, simplifies the design by giving a single criterion for fork choice.

Ultimately, this design decision favours the speed of the network and the simplicity of the system, but trades it off for increased transaction costs.

Not applicable as there is a single fork at a time.

Taiko is enabling asynchronous proving of the posted sequences. This means the rollup accepts proofs for sequences for any future not finalized slot.

Asynchronous proving, allows the provers to “fire and forget” any proofs they have generated thus virtually ensuring that whenever the finalization of the cannonical chain catches up with this slot, it will be able to instantly finalize it and move on. It however opens up the possibility for finalization gaps on the canonical chain of the rollup. Any gap delays the finalization of the canonical chain of the rollup, regardless of all the proofs submitted for future slots. Furthermore it complicates the L1 onchain rules for finalization of the canonical rollup chain introducing complexity for the system. Such a complexity for the system can result in increased L1 costs for proofs submission or chain finalization.

Last but not least, depending on the choices (will come to this later) in crypto economical incentives decisions, gaps can decouple the payment associated with fork/slot finalization and the submission of its proof.

Ultimately, this design decision favours speed of finalization, but trades off complexity of the system and possibly increases the L1 costs for provers.

As stated in the “Prover Choice” decision, Taiko has decided to hard couple sequencers and prover. “Hard coupling” introduces the possibility to slow the finalization of system via liveness problems, however leads to even greater system efficiency.

The increased capital efficiency for the system comes due to two factors. Firstly, due to the actors being able to synchronize offchain ahead of time. Second, because there is no implementation overhead for the L1 to implement “leader election” for provers.

Ultimately, this design decision favours cost efficiency and system simplicity, over possible delayed finalization.

Two types of rewards are given to the sequencers. Firstly, the sequencers receive all the EIP1559 tips of the L2 block in the L2. Secondly, the sequencers are minting TKO tokens as a reward for submitting sequences.

This means that the sequencers are paid early (as soon as they submit a sequence). Early payments are somewhat safer and more capital efficient for sequencers - they do not need a huge inventory to cover the L1 tx costs, except that they need a way to exchange the received reward for ETH. Early payments, however introduce systemic risks through misbehaviour, that needs to be addressed via concrete sequencer punishment mechanisms. An example of this might be a malicious actor monitoring the L1 mempool and frontrunning honest sequencers and taking their profits. In order to avoid this the system need to be extended to account for such attack vectors.

Ultimately, this design decision favours capital efficiency for the sequencers and speed of the network, while trading it off for system complexity in the sequencer punishment section.

While sequencers are paid via L2 block tips and L1 TKO mints, they are also required to pay to the provers for the service of proving. Effectively, this payment acts as a stake for the sequencer.

If the sequencer acts maliciously (f.e. proposes an invalid block), the provers will be able to prove that and the sequencer will lose on the L2 tips and would have paid the prover. No further explicit punishment is introduced.

Ultimately, this design decision favours security of the system, but might increase transaction costs as it requires the sequencer to put up capital upfront (for offchain payment of the prover).

Provers are paid by the sequencers offchain, however they need to post a bond onchain, in order to indicate that they will be proving the sequence. If they act correctly in the allotted timeframe, they get back the bond and effectively their complete reward.

This kind of payment effectively acts as a stake for the prover, ensuring the security of the system.

Ultimately this decision promotes correct behaviour without requiring putting major capital upfront or sacrificing the speed to finalization of the network.

The crypto dynamics of the protocol are balanced in a way, where the sequencers are getting paid with tips and TKO, and the provers are paid by the sequencers and require purchasing TKOs from them. The protocol is paid by the basefee of the L2 protocol making it possibly self sufficient.

The TKO token, however might become inflationary when provers and sequencers keep accumulating it over time. This might see a gradual raise in the required tips to make the system viable.

Ultimately this dynamic nicely balances the incentives of the various actors, however it might lead to increased transaction costs over time.

Overall the architectural approach of TAIKO is well balanced. The design decisions made are favouring for the transaction framework of the network, without needlessly sacrificing too much transaction costs and finalization speed. The design upholds to a high standard of security and decentralisation.

The large amount of effort put into the crypto-economical design of the rollup is quite easily seen, however it can be further improved to tie loose ends. Such loose ends are mostly around the inflation of the TKO token and the rewards for empty blocks.