Timeboost on trial: Arbitrum’s auctions and the next fix

The news: a first big look at Timeboost in the wild

Last week delivered the clearest picture yet of how Arbitrum’s Timeboost behaves under load. On September 26, 2025, researchers published the first large scale empirical study of Timeboost auctions, covering 11.5 million express lane transactions and more than 150,000 auctions from April through July. Their headline findings are blunt: two entities won the vast majority of auctions, revert rates for time boosted transactions were high, and auction competition fell over time. The authors also observed that profitable opportunities clustered at the ends of blocks, which dampened the value of paying for early inclusion. Read the full analysis in the paper The Express Lane to Spam and Centralization, which anchors this piece and its conclusions about auction dynamics and incentives on Layer 2 networks. Read the Timeboost study.

If you build, trade, or govern on Arbitrum, this matters. Sequencing is not just a back office function. It is the control plane for user experience, fairness, and revenue. The study brings fresh, data backed insight to a question that has simmered since Timeboost launched in April: does selling short bursts of exclusive priority tame latency races or simply concentrate control and shift costs elsewhere?

Timeboost in plain English

Imagine a busy toll road with a special gate that opens instantly. Every minute, the right to operate that gate changes hands through a sealed bid, second price auction. The winner controls the express gate for the next minute. They can let their cars through without delay. All the other cars face a small, fixed turnstile pause at the normal booths before they are waved through.

That is Timeboost. On Arbitrum chains that enable it, auctions run in short rounds. By default an auction closes 15 seconds before the next round begins. The winner becomes the express lane controller for that next round and pays the second highest bid into a beneficiary account controlled by the chain’s owner. The sequencer treats the express lane as immediate. Non express transactions are timestamped with an artificial delay of about 200 milliseconds, then follow the normal block cadence. Round length, the closing buffer, the minimum bid, the artificial delay, and the beneficiary are all parameters that governance can tune.

The design attempts to redirect two kinds of behavior. First, it asks sophisticated actors to spend on open auctions rather than burning money on private latency arms races that can flood the network. Second, it gives the chain’s owner a way to internalize some of that value as revenue for a public treasury. In theory, this trades chaos for a visible market and a new funding source.

What the study says and why it happened

Key takeaways that should shape Arbitrum’s next decisions:

Control clustered fast. Two entities won well over 90 percent of rounds in the sample window. That is not surprising once you map the mechanics. Success depends on knowing which opportunities will materialize, having the capital and tooling to bid continuously, and operating infrastructure with near zero variance. Bigger players with better inventory and more stable pipelines dominate this equilibrium. The sealed bid, second price format does not change that. It can even reinforce it by rewarding the best forecasters with cheaper effective wins over time.
Spam did not vanish. Roughly a fifth of time boosted transactions reverted in the sample, which is consistent with a world where the express lane controller experiments aggressively during its minute of control. The cost of trying and failing is often lower than the cost of missing a rare but high value backrun or liquidation. When the right to go first is exclusive, the right holder pushes more attempts through the gate. That shows up as revert heavy throughput rather than clean wins and lower network load.
Auction heat cooled. The authors observed declines in competition and lower auction proceeds as the months wore on. If a small set of actors win most rounds and secondary markets to resell the right underperform, outsiders learn to stop bidding. The game becomes predictable. Treasury revenue falls alongside the number of active participants.

None of this means Timeboost is broken beyond repair. It means the defaults guide the system toward a concentrated, professionalized market. If that is the intention, governance can lean in and capture more value with the right fee floors and controls. If the intention is a wide, competitive market that also curbs spam, the parameters and the surrounding incentives need work.

The levers Arbitrum governance can pull now

Arbitrum’s architecture exposes several clear dials. Each has a cause and effect.

Move the artificial delay. The 200 millisecond head start is the core prize. Increasing it makes the express lane more valuable and will likely raise short term auction revenue. It also makes the baseline user experience slower and could amplify the payoff to concentrated control. Decreasing it narrows the gap and may reduce spam incentives, but it risks lowering revenue. A data driven approach is to test a few week trial at 150 milliseconds and 250 milliseconds on different days and publish the impact on revert rates, auction take, and time to inclusion for retail transactions.
Raise or program the reserve price. A minimum bid sets the floor for treasury revenue and shields against collusive lowball equilibria. Rather than a single static number, governance can direct the auctioneer to update the reserve once per day based on a trailing percentile of winning bids. For example, set the next day’s reserve at the 30th percentile of the past seven days of winning second prices. This is simple and transparent. It avoids intraday meddling while reacting to real demand.
Cap bids per sender and require meaningful deposits. Limiting the number of bids per round per address reduces denial of service pressure on the auctioneer. A non-trivial minimum deposit and a short withdrawal delay can filter low commitment spammers from the bidder set. If one or two entities truly dominate, these rules will not dethrone them, but they will lighten operational noise and make it easier for new entrants to participate credibly.
Shorten or randomize round length. If profitable backruns cluster at the end of blocks, a one minute exclusive window may be too blunt. Cutting round length to 30 seconds or randomly jittering it within a band can make the prize less all or nothing, which tends to soften concentration at the margin. This change has engineering implications for auction resolution and should be tested off chain first.
Publish weekly auction and quality of service reports. The fastest way to restore confidence is visibility. A simple package that shows winning concentration by entity, average and median second price, DAO revenue by day, revert rate by lane, and p95 time to inclusion for non express transactions gives builders and users a signal they can plan around. It also gives governance an objective basis for parameter changes.
Keep the kill switch credible. If the study’s patterns worsen after attempted fixes, pausing Timeboost for a period is better than letting bad incentives harden. Owners should define ahead of time what metrics would trigger a pause and how quickly the network would revert to first come first served ordering.

These are not political statements. They are controls the protocol already has. The value is in choosing a clear objective and measuring the result.

The OP Stack counterpoint: fast confirmations without exclusivity

While Arbitrum refines an exclusivity auction, the Optimism ecosystem is steering in a different direction with its OP Stack. In August, Optimism and Flashbots announced a collaboration to make configurable, verifiable sequencing standard across Superchain networks and to expose fast confirmation features to builders. Chains such as Base and Unichain already run sub second confirmations using Flashbots’ Flashblocks, and the plan is to expand those features to more OP Stack chains. The promise is simple: give users near instant feedback while letting each chain program how blocks are built and verified. Read the announcement for the specifics and examples. Flashbots on OP Stack sequencing.

This is not only about speed. It is about where competition happens. OP Stack chains can differentiate on confirmation targets, external block building policies, and how they prioritize order flow. If Arbitrum’s Timeboost is a market to buy a minute of priority, OP Stack’s approach is a menu of sequencing modes that builders pick and tune. That difference will matter as users grow to expect predictable confirmation under load. See how broader stack choices affect throughput in Polygon’s stateless speed push.

Neutral and shared sequencers are arriving on schedule

A third path is to take sequencing out of the exclusive control of a single chain and run it as a shared, neutral service. For context on decentralizing ordering inside a stack, see Starknet’s leap to decentralized sequencers.

Espresso Systems. Espresso launched Mainnet 0 in late 2024 and is on track to upgrade to Mainnet 1 in the fourth quarter of 2025. The key change is moving to permissionless proof of stake. That means any qualified operator can help provide confirmations, which increases resilience and credibility. Espresso positions itself as rollup native infrastructure with fast confirmations, cross chain composability, and compatibility across stacks. The company has also collaborated with Offchain Labs on research related to Timeboost style ordering in a shared context.
Astria. Astria went live with a decentralized shared sequencing layer in 2024 and has been onboarding production networks. In June 2025, for example, the Forma network migrated its rollup to Astria’s decentralized sequencer via the Palette hard fork. Astria’s pitch is turnkey censorship resistance and fast confirmations with a path to cross rollup atomicity, which appeals to application chains that want reliability without running their own sequencer set.

Both efforts converge on the same promise. Make ordering verifiable and independent. Offer fast, predictable confirmations without relying on a single operator’s goodwill. Let chains retain sovereignty over their rules while outsourcing timely, fair ordering to a specialized network.

The next 60 to 90 days: what to expect and where to look

Rollup teams will compete on ordering fairness and builder experience through the end of the year. Here is what will likely change by December and how to track it.

Parameter experiments on Arbitrum. Expect governance or the Arbitrum Foundation to propose a small set of Timeboost parameter trials by mid October, with results published before Thanksgiving. Watch for tweaks to the artificial delay, reserve price policy, and per sender bid limits. The signal to watch is the trio of DAO revenue, express lane concentration, and revert rate. A successful configuration will reduce revert share and concentration while keeping revenue stable or higher. If two out of three move in the right direction, the configuration is a keeper; otherwise, iterate.
OP Stack chains expand fast confirmations. Base, Unichain, and World Chain have already leaned into external block building and fast confirmation settings. Over the next two months, expect at least one more OP Stack chain to enable sub second confirmations as a headline feature. Developers should monitor user facing time to inclusion and the rate of transaction replacement conflicts to ensure the new settings do not increase failed transactions during peak hours. For recent ecosystem context, see World Chain’s cross chain push.
Espresso’s upgrade and Astria integrations. Espresso’s Q4 upgrade to permissionless proof of stake will open the door for more independent operators and for chains to trial confirmations that are not tied to a single sequencer. Astria, already in production, will likely announce one or two additional migrations or pilots. Builders should evaluate integration paths now so they are not scrambling in December.
Ordering fairness as a marketing line. Chains will publish clear, user friendly explanations of how they order transactions, how they protect against sandwiches and censorship, and how developers can customize block building. Expect dashboards that show confirmation targets, percent of blocks built externally, and the share of order flow that uses public versus private channels.

What users and developers should watch in their own data

Fees. If you are a user, watch your effective fees rather than the sticker gas price. Fees include base gas plus any premium you or your application route payer is spending to gain priority. The easiest proxy is the difference between your expected gas spend and the final spend reported by your wallet. If that spread grows during peak times, the chain’s ordering policy is pushing you into more costly lanes.
Failed transactions and reverts. For power users, track your personal revert rate by day and by application category. For developers, instrument revert reasons and lane attribution. If more than 10 percent of your submitted transactions revert during peak periods, work with your order flow provider or bot logic to reduce duplicate submissions. If you do not control order flow, ask your wallet and relayer what changes they will make in response to the study.
DAO revenues. Treasury inflows from Timeboost give you the best read on whether the auction is attracting real competition. A steady decline over weeks alongside high concentration suggests that the market is thin and needs governance action. A steady or rising line with flat or falling concentration suggests a healthier equilibrium.
Time to inclusion. For non express users, measure p95 time from submission to inclusion during the busiest hour of your app’s day. If it consistently sits above one second on networks that advertise quarter second blocks, the artificial delay and queueing are biting. Applications that depend on quick feedback, such as games or fast trading, should consider routing strategies that can use express capacity when needed but avoid flooding it.
Builder experience. If you are building on OP Stack, test programmable block building and fast confirmation settings in staging and run a gradual rollout. Specifically, monitor user reports of stuck transactions, compare simulation versus actual inclusion ordering for sensitive flows, and measure your help desk ticket volume before and after enabling fast confirmations. If you are building on Arbitrum, measure the impact of Timeboost parameter changes on your own flows and be ready to publish feedback during governance discussions.

The real competition: fairness you can feel

This is not a purity contest between auctions, Flashbots, and shared sequencers. It is a competition to deliver fairness you can feel in the wallet and dashboard. Users want fast, predictable outcomes and protection from predatory order flow. Builders want programmable access to order flow and confirmation guarantees they can design around. Treasuries want durable revenue that does not depend on a handful of bidders.

The Timeboost study forces a decision. If the goal is revenue from a professionalized market, say so and tune for it. If the goal is a broad, competitive market that also tames spam, adjust the parameters and publish the results. If the goal is neutral, verifiable ordering, invest in shared sequencers and standardized fast confirmations. Over the next 90 days, the chains that move first and measure honestly will win developer trust. The rest will inherit whatever equilibrium the fastest few actors choose.

The good news is there are credible paths forward. Arbitrum has levers to pull and the data to evaluate them. OP Stack chains are proving that fast confirmations and flexible sequencing can be product features, not just infrastructure details. Espresso and Astria are maturing into real options for neutral ordering. The race is no longer to be the cheapest. It is to be the fairest at speed, with the least surprise.