Recently Arbitrum’s DAO passed a vote to determine a feature of the new and upcoming sequencing replacement, named: Timeboost™️. It’s my job by the end of this article to plead for the Arbitrum DAO and team to reconsider their implementation of Timeboost™️, but I likely know this is likely a fool’s errand. There are a few major issues I see, both with the greedy and shortsighted rationale for pushing it through quietly and quickly, as well as the likely catastrophic effects it will have on users in the Arbitrum ecosystem.

When talking to CeFi people, I often cited Arbitrum as the lone beacon of hope in an ever degenerating MEV environment. Sadly, that is no longer possible. First come first serve was truly a special ordering structure and led to Arbitrum becoming a haven for users not wishing to have their transactions exploited in every which way, but that is fundamentally broken with this “upgrade” (if you can call it that).

Why did they build this monstrosity?

First and foremost, let’s check out some data on where other L2’s profit comes from:

And let’s also look at the last year of revenue and profit data from Arbitrum, and Base:

Let’s look at two datapoints specifically, since this is a very interesting set of charts from our lovely friend 0xRenaud on his dune dashboard located here.

These two datapoints stuck out to me:

Talk about a come from behind! This in my opinion is the scariest data point for the top dog Arbitrum, as Base has already passed them in last 9 month, and 6 month profit benchmarks. As you can clearly see from the SMG data posted above as well, the majority of the Base profit is coming from priority fees, aka what Searcher’s and MM’s pay for to get priority ordering of their arbitrages and MEV over regular users. To date, Arbitrum has had no priority gas fee mechanism and is still making ETH hand over fist.

This is fundamentally the reason the Foundation and DAO want to build a priority gas fee mechanism, and let me make one thing explicity clear, I DO NOT HAVE A PROBLEM WITH THAT! The thing I despise is this over engineered shitboost mechanism that will likely crush retail on Arbitrum and cause more MEV extracted per dollar of volume than any other chain in the Ethereum ecosystem as a whole. I’ll detail some alternatives I would highly suggest the foundation and DAO explore at the end of the article, but I really do understand the important of ripping apart the design of this “mechanism.” So let’s do that.

The Mechanism Itself

Here’s a snippet of the FAQ on Arbitrum’s own notion document here:

So Timeboost™️ sells off the right to place your transactions at the beginning of a block for a full minute. Given that this will extend the Arbitrum blocktime by 200ms to 450ms (also noted in that document), that means on average the auction winner will receive exclusive first transaction control of ~133 blocks. I made this point on twitter a few days ago, and now I really want to dive into a few ways just off the top of my head that a serious searcher could absolutely fucking destroy this system:

First, let’s take a look at an example the Chaos Labs team excellently detailed in their risk assessment of Timeboost™️, but I believe they significantly underrated the risk of:

Here’s my attempt at drawing this for you smooth brained people:

Make sense? Great. So you are beginning to get a sense of the serious fucking risks of giving someone I consider so evil that I had to google “Demon Wojak” to get an accurate depiction of.

A member of the Arbitrum team even tweeted this after Chaos Labs had posted that writeup:

That isn’t a complicated searcher attack. It’s so simple a low skill ETH Mainnet searcher could writeup a strategy to do this, and absolutely will. On my twitter I linked several examples of strategies orders of magnitude more complicated that are happening today.

But something that is incredibly important to note, that I think even Chaos seemed to miss is that minus the very first and last blocks there is ZERO difference between Timeboost™️ including your transaction at the beginning or END of the block. Ok, so why point that out? Think about what a sandwich attack is. It’s including a transaction at the beginning and END of your transaction that extracts value in some way from it… So if I own the right to the first transaction of this block… and the first transaction at the end of next block… I own the right to sandwich you. I have no fucking clue how this simple level of analysis was ignored at every level of the supply chain for this proposal, but you can imagine the pain my jaw was in when it hit the floor after reading the Timeboost™️ specs.

So let’s look at how I, if I was a demon wojak, would fuck you in the ass if I so chose to for 133 blocks in a row!

Now if you read the Arbitrum docs/FAQ of Timeboost™️, this is actually something they advocate. They believe that the simple front and backrunning of transactions in each block is somehow a good thing because in theory the auction will capture a large portion of this. But that is ONLY the case if the auction was for ONE block. The auction is for 133 BLOCKS. This means, that effectively they’re selling the right to sandwich you for 132 (-1 since the first block is a mulligan) blocks IN A ROW. This fact will lead to a massive amount of Multi-Block MEV likely of an order of magnitude never before seen. I’d also like to point people to one more potentially devious game theory situation that could occur, and I likely expect to happen at least once.

If you’ve ever heard of the prisoner’s dilemma, you know the solution is always be the last one to drop the soap… no wait that can’t be right. Wait a second…

This is a snippet from the wikipedia article on the prisoner’s dilemma. Pretty basic stuff. In any infinite game, cooperate! In any finite game, fuck your counterparty in the ass. This is how things like Celsius or FTX occur. But also how Ebay has incredibly low fraud rates. What does this mean for Arbitrum Timeboost™️? Well it means simply that if I’m an anonymous bidding searcher with the ability to spin up 2^256 different public keys to bid on this auction, I can be as evil as I want since I’m not playing an infinite game, I’m playing an infinite number of finite games. A potential vector of attack here in my opinion is major price manipulation similar to the Chaos Labs example, but over multiple blocks.

Let’s assume the searcher is sophisticated enough to have some model that based on market flows can at a better rate than 50% predict the movements of a token’s price on a centralized exchange. This is effectively the job of a market maker, take in some data, make some prediction based on said data and adjust spreads/orders appropriately. Let’s also assume that the Chaos Labs team is correct in saying that because of Timeboost, no one will attempt CEX-DEX arbitrage during the 200ms delayed transaction period of every block:

Let’s also note that shorter blocktimes reduce extractable MEV from “LVR” (stale price arbitrage) onchain (paper here):

The searcher in that case knows definitively that all the orderflow in a block is retail or low sophistication. So that means that they can effectively treat all blocks in their 1 minute epoch as one gigantic block that runs for a full minute. That means, that they can strategically hold off on arbitraging blocks and let stale prices run wild, then execute a massive arbitrage after T+n blocks versus every single block. This is impossible to model without Timeboost™️ being live, but watch out for searchers looking at this. This will likely be a value extraction method that is tested on select blocks sparsely before getting put fully into production. We’ll likely see some kind of an arms race between the searchers that win the auction and those that don’t.

For example, the auction winning searcher “A”, in their first minute chooses to not arbitrage the last block n-2, but chooses to combine n-2 and n-1 into 1 block through the method I just described. They then throw their backrunning tx into the last block, thus treating the last two block before it as one more profitable bigger block.

Let’s say a searcher “B” notices this. They then realize that they can include a tx in block n-1 in the delayed period that “arbitrages” the block before “A” can. So then on the next minute, “A” includes a backrun tx in n-2 that steals the backrun from “B” before B can execute the arbitrage, therefore turning “B” into a victim of the next block’s first transaction arbitrage!

Apologies if that is a little confusing, but it all turns into a prisoner’s dilemma. If A knows B is NOT going to fuck them on the next block, they betray A before A has an opportunity to. If A finds this out, next time he’ll be smarter.

In conclusion

Please God do not implement this mechanism. If you do, give me free popcorn to watch the fireworks. But more seriously, I believe there are some interesting rabbit holes in the Timeboost mechanism, but the multi block MEV opportunities are just too serious to ignore. Arbitrum is genuinely my favorite chain, and as I mentioned earlier the piece I tell every TardFi alumnus to look at it for a great MEV environment where users are given great pricing, liquidity, and are protected from frontrunning.

That all gets washed down the drain with this proposal.