L2: Scaling Ethereum

:::section{mins=2} Why L2 :::

:::lead If you opened MetaMask in November 2021 and the network asked $52 for a Uniswap swap, you already understood, viscerally, why [[gloss:layer 2|L2]] had to exist. By May 2026 that same swap on Base costs about $0.03, and the trade-off behind that 1,700× drop is what this card is about — what [[gloss:layer 2|L2s]] actually are, what they inherit from [[gloss:ethereum|Ethereum]], and what they give up to be that cheap. :::

:::p{html} [[gloss:ethereum|Ethereum]] [[gloss:l1|L1]] has a hard architectural ceiling: a new [[gloss:block|block]] roughly every 12 seconds, and inside it room for 15–30 [[gloss:transaction|transactions]] per second. We covered the reason in → 2.2 §3 "[[gloss:gas|Gas]] Economics": every full [[gloss:node|node]] has to re-execute every [[gloss:transaction|transaction]], which is what makes the network trustworthy, and also what caps its throughput. You can't speed it up without either weakening security or pushing [[gloss:node|nodes]] off most home machines, and the [[gloss:ethereum|Ethereum]] community has consistently refused both. That's the constraint everything else has had to work around. :::

:::p{html} An [[gloss:layer 2|L2]] is a separate chain that does its own execution and posts a compressed record back to [[gloss:ethereum|Ethereum]]. The [[gloss:layer 2|L2]] runs the [[gloss:transaction|transactions]], packs the resulting state into a small commitment, and writes that commitment to [[gloss:l1|L1]] along with the data needed to reconstruct it. [[gloss:ethereum|Ethereum]] doesn't re-execute any of it — it only stores the data and verifies one short proof or one short challenge window. That's why [[gloss:layer 2|L2s]] can be 100× cheaper than [[gloss:l1|L1]] and still inherit most of its security: an attacker who wants to rewrite [[gloss:layer 2|L2]] history has to attack [[gloss:ethereum|Ethereum]] itself. :::

:::p{html} The strategy was named in October 2020, when Vitalik published "A [[gloss:rollup|rollup]]-centric [[gloss:ethereum|Ethereum]] roadmap" and proposed that [[gloss:l1|L1]] stop trying to scale execution at all. Instead, [[gloss:l1|L1]] would specialize in two things: settlement (where final balances live) and [[gloss:data availability|data availability]] (where [[gloss:layer 2|L2s]] post their data so anyone can rebuild state). Execution moves to a competitive layer of [[gloss:layer 2|L2s]] sitting on top. That was the official story for five years. :::

:::p{html} By April 2026 the [[gloss:layer 2|L2]] ecosystem holds roughly $30–40 billion in [[gloss:tvl|TVL]] across about 40 active networks, and three or four of those — Base, Arbitrum, OP [[gloss:mainnet|Mainnet]] — process around 90% of all [[gloss:layer 2|L2]] [[gloss:transaction|transaction]] volume. Combined [[gloss:layer 2|L2]] throughput sits around 5,600 [[gloss:transaction|transactions]] per second after the [[gloss:fusaka|Fusaka]] upgrade in December 2025, with another step planned through 2026. But the story has also gotten more complicated than the original 2020 plan: monthly active [[gloss:layer 2|L2]] addresses peaked at 58 million in mid-2025 and fell to about 30 million by February 2026, and on February 3, 2026, Vitalik himself said the [[gloss:rollup|rollup]]-centric roadmap "no longer makes sense" in its original form (we'll come back to this in §8). :::

:::callout{icon=💡} Imagine a stock exchange and the brokers who trade on it. The exchange itself — the central clearinghouse — is slow and expensive. Every trade that lands on its book has to be reconciled, settled, and recorded in the master registry of who owns what. That's why brokerage commissions exist: a broker doesn't send each of your trades straight to the exchange, they match orders against other clients inside their own internal book all day, and at the end of the session they ship a single netted statement to the clearinghouse — buy 12,000 shares of one ticker, sell 8,400 of another, net cash flows like this. The exchange records that statement into the master ledger, and now those positions are legally yours. If the broker goes bankrupt overnight, you don't lose your shares: the clearinghouse statement says exactly who is entitled to what, and you can claim your positions directly. The broker is fast and cheap because they don't have to clear every trade individually. They are trusted only up to that final netted statement. :::

:::keypoints{label="Key Points"}

An [[gloss:layer 2|L2]] is not a "faster version of [[gloss:ethereum|Ethereum]]" — it's a separate execution environment that uses [[gloss:ethereum|Ethereum]] as its court of final appeal, with all the asymmetry that implies
"Inherits [[gloss:ethereum|Ethereum]]'s security" is half-true: only the final commitment is anchored on [[gloss:l1|L1]]. Day-to-day, you trust the [[gloss:layer 2|L2]]'s [[gloss:sequencer|sequencer]] to order your [[gloss:transaction|transactions]] honestly and not censor you
The "[[gloss:rollup|rollup]]-centric roadmap" was a deliberate strategic bet by the [[gloss:ethereum|Ethereum]] core team, not an accident of evolution — and as of February 2026 even Vitalik considers it overdue for revision
Cheap fees buy you exit friction: withdrawing from an Optimistic [[gloss:layer 2|L2]] back to [[gloss:l1|L1]] takes about seven days, and the cost of a [[gloss:transaction|transaction]] does not capture the cost of getting your assets out :::

:::details{badge=Advanced title="EIP-4844 blobs, and why "L2 fees are nearly free" stopped being true in December 2025"} :::p{html} Before March 2024, [[gloss:layer 2|L2s]] posted their data to [[gloss:ethereum|Ethereum]] as ordinary calldata — the same expensive [[gloss:transaction|transaction]] payload that any regular [[gloss:l1|L1]] contract uses. In December 2023 alone, [[gloss:layer 2|L2s]] collectively spent about 15,000 [[gloss:ethereum|ETH]] (around $34 million) just to write data into [[gloss:l1|L1]]. That was the single largest line item in [[gloss:layer 2|L2]] economics, and it set a hard floor under the cost of any [[gloss:layer 2|L2]] [[gloss:transaction|transaction]]. :::

:::p{html} [[gloss:eip-4844|EIP-4844]] (the [[gloss:dencun|Dencun]] upgrade, March 13, 2024) introduced a new data type — the [[gloss:blob|blob]]. A [[gloss:blob|blob]] is 128 KiB, the [[gloss:evm|EVM]] cannot access it, [[gloss:node|nodes]] store it for about 18 days and then delete it, and its fee lives on a separate fee market that does not compete with regular [[gloss:gas|gas]]. That was exactly what [[gloss:layer 2|L2s]] needed: their data doesn't have to live on [[gloss:l1|L1]] forever, only long enough for anyone to reconstruct the rolled-up state. [[gloss:layer 2|L2]] fees dropped 90–99% within weeks. A typical Arbitrum swap went from around $0.50 to about $0.01, and on Base from over $0.10 to roughly $0.02 — exactly the cost drop the [[gloss:rollup|rollup]]-centric roadmap had promised. :::

:::p{html} What happened next is less talked about. For most of 2024 and 2025 the [[gloss:blob|blob]] base fee sat at its minimum (1 wei), because not enough [[gloss:blob|blobs]] were being posted to push the price up. Then on December 3, 2025, the [[gloss:fusaka|Fusaka]] upgrade activated PeerDAS (Peer [[gloss:data availability|Data Availability]] Sampling) and tripled [[gloss:blob|blob]] capacity to roughly 12 target / 16 max [[gloss:blob|blobs]] per [[gloss:block|block]], with further increases scheduled. Demand caught up to supply almost immediately, and the [[gloss:blob|blob]] base fee jumped about 15 million times relative to its pre-[[gloss:fusaka|Fusaka]] level. A typical [[gloss:layer 2|L2]] [[gloss:transaction|transaction]] that had been pricing around a cent climbed back to $0.20–$0.30 in the weeks after [[gloss:fusaka|Fusaka]] — a real line item again, not a rounding error. :::

:::p{html} There is a deeper consequence: [[gloss:ethereum|ETH]] is no longer reliably deflationary. Before [[gloss:dencun|Dencun]], [[gloss:l1|L1]] [[gloss:transaction|transactions]] burned enough base fee through [[gloss:eip|EIP]]-1559 to offset [[gloss:validator|validator]] issuance, and [[gloss:ethereum|ETH]] supply slowly shrank. After [[gloss:dencun|Dencun]], activity moved to [[gloss:layer 2|L2s]], [[gloss:l1|L1]] burn collapsed, and issuance overtook burn — [[gloss:ethereum|ETH]] is now slightly inflationary at roughly +0.5% per year. [[gloss:fusaka|Fusaka]] is supposed to claw some of that burn back through the new [[gloss:blob|blob]] base fee, but [[gloss:blob|blobs]] are still cheap enough that the math doesn't fully close. The "ultrasound money" thesis (see → 2.2 §3) is currently more aspiration than fact. :::

:::p{html} The point. [[gloss:layer 2|L2]] economics aren't a one-time win from [[gloss:dencun|Dencun]] — they're a system that keeps drifting. The fee you see on Base today will be a different fee in a year, and which direction it drifts isn't obvious: every [[gloss:l1|L1]] upgrade reshapes both [[gloss:layer 2|L2]] economics and [[gloss:ethereum|ETH]]'s own monetary policy. When anyone says "[[gloss:layer 2|L2s]] are practically free," it's worth asking — free as of when? ::: :::

:::quiz{id=q-2-5-s1 correct=0 explanation="Correct. An [[gloss:layer 2|L2]] inherits [[gloss:ethereum|Ethereum]]'s security by writing the data and the state commitment back to [[gloss:l1|L1]], not by asking [[gloss:ethereum|Ethereum]] to redo the work. Anyone with the data can either verify the [[gloss:validity proof|validity proof]] ([[gloss:zk rollup|ZK rollups]]) or open a fraud challenge within the window ([[gloss:optimistic rollup|optimistic rollups]]). Option B is the misconception this card is built around: [[gloss:ethereum|Ethereum]] specifically doesn't re-execute [[gloss:layer 2|L2]] [[gloss:transaction|transactions]], which is exactly how [[gloss:layer 2|L2s]] can be 100× cheaper. Option C is wrong — pure [[gloss:layer 2|L2]]-only security is what a sidechain offers, not an [[gloss:layer 2|L2]] [[gloss:rollup|rollup]]. Option D describes the [[gloss:l2beat stage|Stage 0]] reality of many [[gloss:layer 2|L2s]] today, but that's a centralization gap, not the security model."} If you make a [[gloss:transaction|transaction]] on an [[gloss:layer 2|L2]] like Arbitrum, which of the following best describes where its security ultimately comes from?

From the [[gloss:layer 2|L2]] posting a compressed commitment and the underlying data to [[gloss:ethereum|Ethereum]], so anyone can verify the proof or challenge the state — [[gloss:ethereum|Ethereum]] doesn't re-execute the [[gloss:transaction|transactions]], it just stores the data and checks the proof
From [[gloss:ethereum|Ethereum]] re-executing every [[gloss:layer 2|L2]] [[gloss:transaction|transaction]] on [[gloss:l1|L1]] — that's the whole point of inheriting security
From the [[gloss:layer 2|L2]]'s own [[gloss:validator|validators]] alone — [[gloss:ethereum|Ethereum]] is not involved once the [[gloss:layer 2|L2]] confirms the [[gloss:transaction|transaction]]
From a [[gloss:multisig|multisig]] of well-known [[gloss:layer 2|L2]] founders who can override the chain if needed :::

:::section{mins=3} Rollups: The Idea and the Two Types :::

:::lead A [[gloss:rollup|rollup]] batches thousands of [[gloss:layer 2|L2]] [[gloss:transaction|transactions]] and posts one compressed summary to [[gloss:ethereum|Ethereum]]. The two ways of convincing [[gloss:ethereum|Ethereum]] that the summary is honest — [[gloss:fraud proof|fraud proofs]] and [[gloss:validity proof|validity proofs]] — are not minor implementation details. They produce completely different products: one is cheap to build but slow to exit, the other is cryptographically airtight but expensive to operate. Most of what people argue about when they argue about [[gloss:layer 2|L2s]] comes back to this single choice. :::

:::p{html} The mechanical idea behind a [[gloss:rollup|rollup]] is the same in both variants. A [[gloss:sequencer|sequencer]] collects [[gloss:transaction|transactions]] on [[gloss:layer 2|L2]], executes them, and computes a new state root — a short cryptographic fingerprint of who owns what after the batch. It then publishes two things to [[gloss:l1|L1]]: the compressed [[gloss:transaction|transaction]] data (so anyone with that data can reconstruct the [[gloss:layer 2|L2]] state from scratch), and a claim about what the new state root is. The difference is in what backs that claim. :::

:::p{html} [[gloss:optimistic rollup|Optimistic rollups]] publish the claim with no proof attached, then dare the world to challenge it. [[gloss:ethereum|Ethereum]] stores the new state root immediately, but doesn't accept it as final for a window — typically seven days, fixed at about twelve days on Arbitrum after the [[gloss:bold|BOLD]] upgrade in February 2025 (which traded a slightly longer window for a hard upper bound, so delay attacks can no longer drag disputes out indefinitely). During that window, anyone running an [[gloss:layer 2|L2]] [[gloss:node|node]] can rerun the batch, and if their answer differs from the [[gloss:sequencer|sequencer]]'s, they post a [[gloss:fraud proof|fraud proof]] on [[gloss:l1|L1]]. [[gloss:ethereum|Ethereum]] then arbitrates: it doesn't replay the whole batch, it bisects the dispute down to a single contested step and executes only that step on [[gloss:l1|L1]]. If the [[gloss:sequencer|sequencer]] was lying, the batch is rolled back and the challenger is paid out of the [[gloss:sequencer|sequencer]]'s bond. If no one challenges within the window, the batch is final and withdrawals can settle. :::

:::p{html} [[gloss:zk rollup|ZK rollups]] publish the claim with a cryptographic [[gloss:validity proof|validity proof]] attached. The proof is a short certificate that a separate machine (the prover) generates by running the batch through a special circuit. It mathematically attests that the new state root is the unique correct result of applying these [[gloss:transaction|transactions]] to the old state root. [[gloss:ethereum|Ethereum]] verifies the proof in milliseconds; if it passes, the new state is final immediately. There is no challenge window, no honest-watcher assumption, and no waiting period for withdrawals. The cost is that generating the proof is computationally expensive — Scroll, zkSync Era, and Linea paid roughly $21M, $17M, and $12M respectively in proof-related on-chain fees over a recent twelve-month window, and prover infrastructure costs (off-chain) are a multiple of that. :::

:::p{html} The split is consequential. [[gloss:optimistic rollup|Optimistic rollups]] are cheap to build because the "proof" they need from [[gloss:l1|L1]] is just an honest re-execution by anyone, and the [[gloss:l1|L1]] contract is mostly idle. [[gloss:zk rollup|ZK rollups]] front-load all the hard work into the prover and rely on real [[gloss:cryptography|cryptography]], which means they finalize fast but need expensive specialized hardware and complex circuits. Both kinds of [[gloss:rollup|rollup]] still post their data to [[gloss:ethereum|Ethereum]] and therefore inherit its data-availability guarantee equally — the difference is in how they convince [[gloss:l1|L1]] the result is correct, not in how the result gets reconstructed if the [[gloss:layer 2|L2]] disappears. :::

:::callout{icon=💡} Imagine a teacher grading a stack of math homework. An optimistic classroom works on trust: the teacher takes each student's answer at face value and writes it into the gradebook the same day. But for a week afterwards, any classmate is allowed to challenge an answer. If they think Alice got problem twelve wrong, they have to write out the contradiction step by step and present it. The teacher then doesn't re-grade everything — they just look at the exact line in dispute, decide who's right, and if Alice was wrong, the grade is erased and the challenger gets a prize from a bond Alice had to post at the start of the term. After a week with no challenges, the grade is locked in. A validity classroom works differently: every student has to submit, along with the answer, a special form that mathematically proves their work — not a longhand solution, a cryptographic certificate the teacher can check with a quick stamp. The certificate is hard for the student to produce — they need a special calculator that takes a long time — but the teacher's check is instant, and there's no week-long appeal period. Both classrooms end up with the same gradebook. They differ in who does the hard work and when the grade becomes final. :::

:::keypoints{label="Key Points"}

[[gloss:optimistic rollup|Optimistic rollups]] rely on a single honest watcher — for security to hold, at least one person somewhere must be rerunning every batch and willing to post a [[gloss:fraud proof|fraud proof]]. [[gloss:zk rollup|ZK rollups]] make no such assumption: math alone is enough
The seven-day exit window on [[gloss:optimistic rollup|optimistic rollups]] is not a bug to be fixed — it's the exact time honest watchers need to detect and submit a [[gloss:fraud proof|fraud proof]]. Shorten it and you weaken the security model itself
"[[gloss:zk rollup|ZK rollup]]" is the conventional label, though "validity-proof [[gloss:rollup|rollup]]" is technically the more accurate name. The "zero-knowledge" property isn't actually used to hide anything — every [[gloss:transaction|transaction]] is public — it's used for the proof's succinctness, not for privacy
Both types post the same [[gloss:transaction|transaction]] data to [[gloss:ethereum|Ethereum]] — meaning even if every [[gloss:layer 2|L2]] [[gloss:sequencer|sequencer]] disappeared tomorrow, anyone with a copy of the [[gloss:l1|L1]] chain could rebuild every [[gloss:layer 2|L2]] user's balance. That's what "inheriting [[gloss:data availability|data availability]]" buys you :::

:::details{badge=Advanced title="SNARK vs STARK, and why the choice shapes everything else about a rollup"} :::p{html} The "[[gloss:validity proof|validity proof]]" inside a [[gloss:zk rollup|ZK rollup]] is one of two cryptographic families: [[gloss:snark|SNARK]] (Succinct Non-interactive Argument of Knowledge) or [[gloss:stark|STARK]] (Scalable Transparent Argument of Knowledge). Both let a prover convince a verifier that a long computation was carried out correctly, without making the verifier replay the computation. They produce very different [[gloss:rollup|rollups]]. :::

:::p{html} [[gloss:snark|SNARKs]] are small. A typical [[gloss:snark|SNARK]] proof is a few hundred bytes, and verifying it on [[gloss:ethereum|Ethereum]] costs roughly 300,000 [[gloss:gas|gas]] — cheap enough that it doesn't dominate the [[gloss:l1|L1]] fee bill. The price you pay is a trusted setup ceremony: before the [[gloss:rollup|rollup]] goes live, a group of participants has to generate cryptographic parameters together, and if every single one of them colluded and kept their "toxic waste" randomness, they could forge proofs forever after. In practice, modern setups use multi-party ceremonies with hundreds of participants, and the assumption is that at least one of them was honest. [[gloss:snark|SNARKs]] also rely on elliptic-curve pairings, which would be broken by a sufficiently large quantum computer. :::

:::p{html} [[gloss:stark|STARKs]] are larger — proofs are tens to hundreds of kilobytes, and verification costs millions of [[gloss:gas|gas]], multiple times more than a [[gloss:snark|SNARK]] — but they're transparent (no setup ceremony at all, just public randomness) and rely only on [[gloss:hash|hash]] functions, which are believed to remain secure against quantum attacks. Starknet is the largest production [[gloss:stark|STARK]] [[gloss:rollup|rollup]]; the cost shows up in higher per-batch [[gloss:l1|L1]] fees, which it absorbs through proof aggregation and recursion. :::

:::p{html} This split cascades into product decisions. zkSync, Polygon [[gloss:zkevm|zkEVM]] (sunsetting in 2026), Scroll, and Linea all use [[gloss:snark|SNARK]]-based systems and accept the trusted-setup assumption. Starknet picked [[gloss:stark|STARKs]] and the post-quantum guarantee, at the cost of a custom non-[[gloss:evm|EVM]] language ([[gloss:cairo|Cairo]]) optimized for cheap [[gloss:stark|STARK]] proving. zkSync uses a hybrid where [[gloss:stark|STARKs]] handle the heavy computation and then get wrapped in a [[gloss:snark|SNARK]] for cheap on-chain verification — the best of both at the cost of more complex infrastructure. :::

:::p{html} The point. When someone says "[[gloss:zk rollup|ZK rollup]]", a lot of architectural decisions are already implied. Proof system → setup assumptions → quantum security → language → developer experience → fee structure. There's no single "[[gloss:zk rollup|ZK rollup]] design"; there are families, and they trade different risks against different costs. The choice that looks like a low-level [[gloss:cryptography|cryptography]] detail is actually the one that determines what kind of [[gloss:layer 2|L2]] you're using. ::: :::

:::quiz{id=q-2-5-s2 correct=1 explanation="Correct. The window exists for one reason: [[gloss:optimistic rollup|optimistic rollups]] don't include a proof with the batch, so [[gloss:ethereum|Ethereum]] can't verify correctness on its own. Instead, security depends on someone external rerunning the batch and submitting a [[gloss:fraud proof|fraud proof]] if the [[gloss:sequencer|sequencer]] cheated. Seven days is a conservative estimate of how long that detection and submission might take in the worst case — accounting for network downtime, weekends, and time to assemble the proof. Option A is wrong, there's no such regulation. Option C is wrong, [[gloss:ethereum|Ethereum]] finalizes its own [[gloss:block|blocks]] in 12-15 minutes regardless of [[gloss:layer 2|L2]] activity. Option D describes [[gloss:zk rollup|ZK rollups]], not optimistic ones — [[gloss:optimistic rollup|optimistic rollups]] don't generate [[gloss:validity proof|validity proofs]] at all."} Why does an [[gloss:optimistic rollup|optimistic rollup]] need a seven-day window before withdrawals can settle on [[gloss:l1|L1]]?

It's a regulatory requirement imposed by US financial authorities on all crypto bridges
It's the window during which any honest watcher can detect a fraudulent batch and post a [[gloss:fraud proof|fraud proof]] — the security model depends on someone having enough time to do so
It's the time [[gloss:ethereum|Ethereum]] itself needs to fully process and finalize the [[gloss:rollup|rollup]]'s batch on [[gloss:l1|L1]]
It's the time prover hardware needs to generate the [[gloss:validity proof|validity proof]] for a batch of that size :::

:::section{mins=3} Optimistic L2: Arbitrum, Optimism, Base :::

:::lead Three [[gloss:optimistic rollup|optimistic rollups]] handle the vast majority of [[gloss:layer 2|L2]] activity, and they share almost all their technical DNA — but they sell completely different products. Arbitrum is the technical fortress for [[gloss:defi|DeFi]] power users. Optimism is a public-goods experiment with a shared codebase its users help fund. Base is a retail funnel with a hundred million Coinbase users behind it. Same fraud-proof mechanics underneath, three different bets on what an [[gloss:layer 2|L2]] should actually be. :::

:::p{html} Arbitrum One launched in August 2021 and grew up as the serious [[gloss:defi|DeFi]] [[gloss:layer 2|L2]]. Its execution layer is Nitro, an [[gloss:evm|EVM]]-compatible runtime that has been the gold standard for ported [[gloss:ethereum|Ethereum]] contracts for years. By April 2026 Arbitrum One holds roughly $13–17 billion in [[gloss:tvl|TVL]] (depending on which data provider counts), making it the largest [[gloss:layer 2|L2]] by capital actually working in [[gloss:defi|DeFi]] protocols. Daily [[gloss:transaction|transactions]] sit around 4.3 million, fees around 4 cents for a USDC transfer. The standout 2025 event was the activation of [[gloss:bold|BOLD]] on February 12 — the upgrade that turned Arbitrum into the first major [[gloss:layer 2|L2]] with truly permissionless [[gloss:validator|validators]]. Before [[gloss:bold|BOLD]], only an allowlisted set of entities could submit [[gloss:fraud proof|fraud proofs]]. After [[gloss:bold|BOLD]], anyone running a [[gloss:node|node]] can defend the chain, and a single honest [[gloss:validator|validator]] is mathematically sufficient to defeat any number of malicious ones. [[gloss:stylus|Stylus]], generally available since mid-2024, adds Rust, C, and C++ to Arbitrum's developer surface area alongside [[gloss:solidity|Solidity]], with claimed 30%+ [[gloss:gas|gas]] savings for compute-heavy workloads. In late 2025 Robinhood deployed nearly 2,000 tokenized equities on Arbitrum One — the first time a major US broker put traditional securities on an [[gloss:layer 2|L2]] — and announced plans to build its own dedicated chain using the Arbitrum stack. :::

:::p{html} OP [[gloss:mainnet|Mainnet]] (originally just "Optimism") was the first [[gloss:optimistic rollup|optimistic rollup]] in production, going live in December 2021. Its codebase, the [[gloss:op stack|OP Stack]], became something rarer than a chain: a shared toolkit. Today over thirty other [[gloss:rollup|rollups]] run on it, including Base, Unichain (Uniswap), Ink (Kraken), and World. OP [[gloss:mainnet|Mainnet]] itself holds around $5.6 billion in [[gloss:tvl|TVL]] — smaller than Arbitrum or Base — but its strategic weight comes from the [[gloss:superchain|Superchain]], which we'll go through in §5. The Optimism Collective runs a system called Retroactive Public Goods Funding ([[gloss:retropgf|RetroPGF]]): the chain's [[gloss:sequencer|sequencer]] revenue partly flows into rounds that pay teams that have already built useful infrastructure, judged by the impact they delivered rather than the promise of what they will. Through Q1 2026 alone, [[gloss:retropgf|RetroPGF]] distributed roughly $14 million to public-goods projects. It's the most concrete attempt anywhere in crypto to wire prosocial economics into the protocol's own balance sheet. :::

:::p{html} Base launched in August 2023 and rewrote everyone's expectations of what an [[gloss:layer 2|L2]] can grow into. Operated by Coinbase, it inherits direct access to roughly 100 million verified Coinbase users — a distribution channel no other [[gloss:layer 2|L2]] has and no other [[gloss:layer 2|L2]] can replicate at any price (the same kind of structural advantage [[gloss:ton|TON]] gets from Telegram, see → 2.4 §5). By May 2026 Base is the dominant [[gloss:layer 2|L2]] on almost every activity metric: :::

:::p{html}

Bridged [[gloss:tvl|TVL]] — about $13 billion, with $4–4.5 billion of that specifically in [[gloss:defi|DeFi]] protocols.
Daily [[gloss:transaction|transactions]] — around 12.89 million, roughly 3× Arbitrum's count.
Daily active addresses — over one million.
Share of [[gloss:layer 2|L2]] [[gloss:defi|DeFi]] [[gloss:tvl|TVL]] — 46%.

:::

:::p{html} Morpho on Base grew from $48 million to over $2 billion in [[gloss:tvl|TVL]] during 2025 — most of that lending originated inside the Coinbase app itself. Base reached [[gloss:l2beat stage|Stage 1]] in 2024 (fault proofs live), but its [[gloss:sequencer|sequencer]] is a single Coinbase-operated [[gloss:node|node]], and in February 2026 a misconfiguration caused several hours of delayed and dropped [[gloss:transaction|transactions]] before being rolled back. That same month, the team announced Base would be migrating off the [[gloss:op stack|OP Stack]] to its own unified codebase — a [[gloss:move|move]] with consequences for the [[gloss:superchain|Superchain]] (again, §5). :::

:::p{html} Three networks, one mechanism, three different theses on what scaling [[gloss:ethereum|Ethereum]] actually means. Arbitrum bets that depth and credibility win in the long run. Optimism bets on shared infrastructure and an economic model that funds the commons. Base bets that distribution and onboarding eat everything else. None of these bets has obviously lost, but as of May 2026 the activity numbers are tilting hard toward the distribution bet, and that tilt is what makes the next few years interesting to watch. :::

:::callout{icon=💡} Imagine three restaurants run by the same chef school. They all use the same recipes, the same kind of ovens, and the same wholesale supplier — but each is positioned for a completely different customer. The first is a quiet, expensive fine-dining place where serious gourmets come for tasting menus and unusual ingredients. The chef does Rust-based experimental cooking on weekends, and the sommelier just put two thousand tokenized vintage wines on the menu. That's Arbitrum. The second is a neighborhood spot known for a peculiar policy: a portion of every receipt is rerouted to projects that already proved themselves useful to the city — a free library, a fixed park bench, a working community garden — paid in arrears, not in promises. Other restaurants have started copying its recipe book. That's Optimism. The third is a food court inside the largest department store in town. It serves the same recipes, but it sees more diners in a day than the other two see in a month, because everyone walking through the mall is already there. The location does ninety percent of the work; the food just has to be cheap, fast, and good enough. That's Base. Same kitchen, same fraud-proof oven, three completely different businesses. :::

:::keypoints{label="Key Points"}

Arbitrum, OP [[gloss:mainnet|Mainnet]], and Base all use [[gloss:fraud proof|fraud proofs]], but their security postures aren't the same: Arbitrum is the only one with permissionless [[gloss:validator|validators]] (since [[gloss:bold|BOLD]], February 2025), while Base and OP [[gloss:mainnet|Mainnet]] still rely on operator-controlled [[gloss:validator|validator]] sets
Base's growth came from one thing the technical literature rarely treats as serious: a distribution channel. 100 million Coinbase users and one-click bridging beat better tech every time so far
[[gloss:tvl|TVL]] and [[gloss:transaction|transaction]] count measure different things and lead to different "winners": Arbitrum wins on capital depth ($13–17B), Base wins on [[gloss:transaction|transaction]] volume (12.89M/day) and [[gloss:defi|DeFi]] share (46%), OP [[gloss:mainnet|Mainnet]] doesn't lead on either but anchors the [[gloss:superchain|Superchain]]
When the biggest member of a shared codebase walks away, the shared codebase stops being the obvious answer. Base's February 2026 [[gloss:op stack|OP Stack]] departure isn't just an operational decision — it's a vote of confidence (or no confidence) in shared infrastructure as a strategy; see §5 :::

:::details{badge=Advanced title="BOLD's bisection game, force inclusion, and how a single honest watcher can defeat any number of liars"} :::p{html} The mechanics of an [[gloss:optimistic rollup|optimistic rollup]]'s [[gloss:fraud proof|fraud proof]] are more interesting than "someone challenges, then [[gloss:ethereum|Ethereum]] decides." The challenge is structured as a recursive bisection game between the challenger and the [[gloss:sequencer|sequencer]]. When the challenger claims the [[gloss:sequencer|sequencer]]'s state root is wrong, they don't get to argue the whole batch — they have to point to a midpoint, say which side of it they disagree with, and then the dispute zooms in on that half. They repeat this until the disagreement is reduced to a single [[gloss:layer 2|L2]] instruction. Only that one instruction is executed on [[gloss:l1|L1]], where [[gloss:ethereum|Ethereum]] can definitively rule who is right. This means a batch of millions of [[gloss:layer 2|L2]] [[gloss:transaction|transactions]] can be adjudicated using a handful of [[gloss:l1|L1]] [[gloss:transaction|transactions]] — the protocol's killer trick. :::

:::p{html} Before [[gloss:bold|BOLD]], this game had a critical flaw: it was an n-versus-1 protocol, so an attacker could submit many false claims in parallel and force the honest defender to fight each one. With enough false claims, the defender either runs out of bond capital or runs out of time within the dispute window, and one of the attacker's claims gets through. [[gloss:bold|BOLD]] ("Bounded Liquidity Delay") restructures the game so that it's all-versus-all instead of one-versus-one: any number of parties can join the same dispute, every honest party defends the same correct state, and a single honest participant is sufficient to win against any number of attackers. This is the property that makes truly permissionless validation safe — you no longer need a vetted set of [[gloss:validator|validators]] because the math no longer cares how many liars show up. :::

:::p{html} The other under-discussed piece is force inclusion. If the [[gloss:sequencer|sequencer]] is down, censoring your [[gloss:transaction|transactions]], or actively malicious, you can submit your [[gloss:transaction|transaction]] directly to a contract on [[gloss:ethereum|Ethereum]] [[gloss:l1|L1]] — the Delayed Inbox — and the [[gloss:rollup|rollup]] is required to execute it after a delay (about 24 hours on Arbitrum One). This is a slow path, not an everyday one, but it's the safety valve that prevents the [[gloss:sequencer|sequencer]] from becoming a king. Without force inclusion, a centralized [[gloss:sequencer|sequencer]] could quietly delete your withdrawal request and you'd have no recourse. With it, you can always escape the [[gloss:layer 2|L2]], just not quickly. Every serious [[gloss:optimistic rollup|optimistic rollup]] has some version of this; Base and OP [[gloss:mainnet|Mainnet]] implement the same [[gloss:op stack|OP Stack]] mechanism. :::

:::p{html} The point. The "security" of an [[gloss:optimistic rollup|optimistic rollup]] is not a single property; it's the interaction of three: a fraud-proof game that's mathematically sound ([[gloss:bold|BOLD]]-style all-versus-all), a participant set that's actually open (permissionless [[gloss:validator|validators]]), and a force-inclusion path that bypasses the [[gloss:sequencer|sequencer]]. Arbitrum has all three. Base and OP [[gloss:mainnet|Mainnet]] have one and three but not two. That gap is the practical difference between "[[gloss:l2beat stage|Stage 1]]" and what [[gloss:l2beat stage|Stage 1]] was supposed to mean. ::: :::

:::quiz{id=q-2-5-s3 correct=3 explanation="Correct. Base's defining advantage is the distribution funnel, not any technical edge. Coinbase already had ~100 million verified users when Base launched, and they could [[gloss:move|move]] into the chain inside the app they were already using. Morpho's [[gloss:tvl|TVL]] on Base grew from $48M to over $2B in 2025 mostly because Coinbase's app routed lending demand there directly. Option A is wrong — Base's fraud-proof system is the same [[gloss:op stack|OP Stack]] code used by OP [[gloss:mainnet|Mainnet]] (and Base hasn't yet matched Arbitrum's permissionless [[gloss:bold|BOLD]] model). Option B is wrong — Base notably does not have an airdrop or native token as of early 2026, though one has been discussed. Option C is wrong for the same reason: no native token, [[gloss:gas|gas]] is paid in [[gloss:ethereum|ETH]]."} Base has grown to become the largest [[gloss:layer 2|L2]] by [[gloss:transaction|transaction]] volume and [[gloss:defi|DeFi]] market share. What's the single most important driver of that growth?

A technically superior fraud-proof system that gives Base faster [[gloss:finality|finality]] than Arbitrum or OP [[gloss:mainnet|Mainnet]]
Generous airdrop and incentive programs that paid users directly to use the chain
A native token with strong staking rewards that attracted long-term holders
Direct distribution through the Coinbase app, which already had roughly 100 million verified users with funded accounts and one-click access to bridging :::

:::section{mins=2} ZK Rollups and zkEVMs :::

:::lead Validity-proof [[gloss:rollup|rollups]] were the long-promised "real" scaling for [[gloss:ethereum|Ethereum]]: no challenge window, no honest-watcher assumption, mathematically final the moment the proof lands on [[gloss:l1|L1]]. By 2026 the technology works, every major [[gloss:zkevm|zkEVM]] is in production, and yet none of them has overtaken the [[gloss:optimistic rollup|optimistic rollups]] in capital or activity. The story of why is the actual story of [[gloss:zk rollup|ZK rollups]] today — not what they can do, but what it costs to do it and whether the market cares. :::

:::p{html} The current [[gloss:zk rollup|ZK rollup]] field has four serious production chains and one notable shutdown. Starknet, built by StarkWare, uses [[gloss:stark|STARK]] proofs and a custom non-[[gloss:evm|EVM]] language called [[gloss:cairo|Cairo]]. Its [[gloss:tvl|TVL]] sits around $1–1.5 billion in [[gloss:defi|DeFi]] terms, but its technical position is unique: it's the first [[gloss:layer 2|L2]] in production with a distributed [[gloss:sequencer|sequencer]] (since the Grinta upgrade in September 2025), it reached [[gloss:l2beat stage|L2BEAT Stage]] 1 in May 2025, and its [[gloss:block|block]] time dropped from 30 seconds to 4 seconds with pre-confirmations under half a second. [[gloss:cairo|Cairo]] is a real cost — developers have to learn a new language instead of porting [[gloss:solidity|Solidity]] contracts — but it lets the prover be dramatically more efficient, which is part of why Starknet has been able to push [[gloss:decentralization|decentralization]] further than peers. :::

:::p{html} zkSync Era (by Matter Labs) and Linea (by Consensys, the team behind MetaMask) take the opposite bet: maximize compatibility with existing [[gloss:ethereum|Ethereum]] tooling. Linea sits at about $3–4 billion in [[gloss:defi|DeFi]] [[gloss:tvl|TVL]], helped enormously by direct integration into MetaMask's ~30 million monthly users, and positions itself as "the first battle-tested [[gloss:native rollup|Native Rollup]] stack" — a reference to where Vitalik thinks [[gloss:zk rollup|ZK rollups]] should be headed (see §8). zkSync Era is in the $4–5 billion range, has native [[gloss:account abstraction|account abstraction]] baked in at the protocol level, and pivoted in 2025 toward institutional and privacy-focused settlement with a product called Prividium. Scroll, with roughly $2 billion [[gloss:tvl|TVL]], is the purest [[gloss:evm|EVM]]-equivalent [[gloss:zkevm|zkEVM]] — [[gloss:solidity|Solidity]] contracts deploy with literally zero changes — and as of 2025 it joined Starknet at [[gloss:l2beat stage|L2BEAT Stage]] 1, the only two [[gloss:zk rollup|ZK rollups]] to have done so. zkSync and Linea remain at [[gloss:l2beat stage|Stage 0]]. :::

:::p{html} Then there's Polygon [[gloss:zkevm|zkEVM]], which is the cautionary tale. Polygon bought Hermez in 2021 for $250 million and turned it into Polygon [[gloss:zkevm|zkEVM]], one of the first production [[gloss:zkevm|zkEVMs]]. By 2025 the chain was running at over $1 million per year in losses, had never been upgraded to use [[gloss:blob|blobs]] after [[gloss:eip-4844|EIP-4844]], and had quietly stopped active development. In June 2025 Polygon's leadership announced the chain would be sunset; the [[gloss:sequencer|sequencer]] was permanently shut down on July 1, 2026. Polygon redirected its entire ZK research effort into AggLayer — a cross-chain settlement protocol — and moved its lead [[gloss:zkevm|zkEVM]] engineer to a separate spin-off company called ZisK. The lesson isn't that [[gloss:zk rollup|ZK rollups]] are hard (everyone knew that). The lesson is that even a well-funded, technically credible chain can lose to market positioning, and [[gloss:zk rollup|ZK rollups]] in particular have to keep pace with [[gloss:l1|L1]] economics (specifically [[gloss:blob|blob]] support) or quietly die. :::

:::p{html} Stepping back: [[gloss:zk rollup|ZK rollups]] now collectively settle more [[gloss:stablecoin|stablecoin]] volume than every [[gloss:optimistic rollup|optimistic rollup]] combined, but they trail by a wide margin on raw [[gloss:tvl|TVL]] and consumer activity. The remaining engineering work — [[gloss:sequencer|sequencer]] [[gloss:decentralization|decentralization]], cross-[[gloss:rollup|rollup]] composability, proof aggregation to bring per-batch [[gloss:l1|L1]] cost down — is real but well-defined. There are no open research questions. What [[gloss:zk rollup|ZK rollups]] still need to figure out is something dirtier: how to attract liquidity and developers when the optimistic side has a four-year head start, an easier development path, and lower operating costs. :::

:::callout{icon=💡} Picture two auction houses selling antique paintings. At the first, when a bidder wins a lot, the house records the sale, hands over the painting, and pockets the commission. If anyone later believes the work is a forgery, they have a window — say, seven days — to commission an independent expert, prove the case, and have the sale reversed. That's an [[gloss:optimistic rollup|optimistic rollup]]. At the second, every painting arrives at the floor with a technical dossier: spectral analysis of the pigments, isotope dating of the canvas, ultra-high-resolution scans of brush strokes cross-referenced against a master database. Producing each dossier takes a forensic lab weeks and costs real money. But once it's in front of you, the in-house verifier scans the certificate, checks the cryptographic signature, and confirms authenticity in under a minute — no week-long appeal, no expert callbacks. Both auction houses sell the same paintings to the same collectors at similar prices. They differ in who carries the cost of producing certainty and when the buyer can walk away truly sure. :::

:::keypoints{label="Key Points"}

"[[gloss:zk rollup|ZK rollup]]" is one mechanism; "[[gloss:zkevm|zkEVM]]" is a product category. A [[gloss:zkevm|zkEVM]] is a [[gloss:zk rollup|ZK rollup]] that runs [[gloss:solidity|Solidity]] contracts — Starknet, which uses [[gloss:cairo|Cairo]], is a [[gloss:zk rollup|ZK rollup]] but not a [[gloss:zkevm|zkEVM]]
Of all the ZK chains, only Starknet and Scroll have reached [[gloss:l2beat stage|L2BEAT Stage]] 1. zkSync Era and Linea — much larger by [[gloss:tvl|TVL]] — are still [[gloss:l2beat stage|Stage 0]]. Size and [[gloss:decentralization|decentralization]] don't [[gloss:move|move]] together
Polygon [[gloss:zkevm|zkEVM]]'s shutdown wasn't a technology failure — it was a positioning failure. The chain technically worked but couldn't afford its own proof costs after the team failed to adopt [[gloss:blob|blobs]], and the broader Polygon strategy moved past it
Choosing between [[gloss:zk rollup|ZK rollups]] in 2026 isn't a binary "which is best" — it's a tradeoff matrix: [[gloss:cairo|Cairo]]-language depth (Starknet), [[gloss:solidity|Solidity]] equivalence (Scroll), MetaMask distribution (Linea), or [[gloss:account abstraction|account abstraction]] and privacy positioning (zkSync). The "best [[gloss:zkevm|zkEVM]]" depends entirely on what you're building :::

:::details{badge=Advanced title="Vitalik's zkEVM type spectrum, and what each rollup actually picked"} :::p{html} Every [[gloss:zk rollup|ZK rollup]] that wants to run [[gloss:ethereum|Ethereum]]-style [[gloss:smart contract|smart contracts]] has to make one foundational tradeoff: how identical its execution should be to [[gloss:ethereum|Ethereum]]'s. The choice has consequences in both directions. Stay close to [[gloss:ethereum|Ethereum]] and existing tools, contracts, and audits all work unchanged — but proving every [[gloss:block|block]] becomes punishingly slow and expensive, because [[gloss:ethereum|Ethereum]] has dozens of historical design decisions that are simply hostile to ZK proving (the Keccak [[gloss:hash|hash]] function, Merkle Patricia tries, RLP encoding). [[gloss:move|Move]] away from [[gloss:ethereum|Ethereum]]'s exact rules and the prover becomes orders of magnitude cheaper — but [[gloss:solidity|Solidity]] contracts start to misbehave in subtle ways, [[gloss:evm|EVM]]-specific tools no longer work, and developers may have to learn a different language entirely. :::

:::p{html} In August 2022 Vitalik Buterin wrote an essay laying out this spectrum and giving the points on it numbers, from Type 1 (literally identical to [[gloss:ethereum|Ethereum]]) through Type 4 ([[gloss:solidity|Solidity]]-source compatible but compiled to a different VM). The numbers themselves don't matter much — what matters is that they let a developer ask one specific question about any [[gloss:zkevm|zkEVM]]: where on this spectrum does it sit, and what does that mean for my code? :::

:::p{html} The chains closest to [[gloss:ethereum|Ethereum]] live on the expensive end. Taiko is the public attempt at full Type-1 equivalence: it can theoretically prove any historical [[gloss:ethereum|Ethereum]] [[gloss:block|block]], which means existing infrastructure works without any adaptation. Nobody is running this at meaningful production scale because the proving cost is currently brutal. Scroll takes one step away — its contracts deploy byte-for-byte from [[gloss:ethereum|Ethereum]], but it uses ZK-friendlier [[gloss:hash|hash]] functions under the hood (Type 2). For application developers this is invisible; for tooling like [[gloss:block|block]] explorers and provers, the difference is everything. Linea sits a little further out (sometimes called Type 2.5), trading a few more [[gloss:ethereum|Ethereum]]-internals shortcuts for faster proving. :::

:::p{html} At the far end, zkSync Era and Starknet made the opposite call. [[gloss:solidity|Solidity]] code still runs on zkSync, but it gets compiled to a custom [[gloss:zkevm|zkEVM]]-style VM rather than to [[gloss:ethereum|Ethereum]]'s [[gloss:evm|EVM]] directly — most contracts work, a few low-level patterns don't, and developers do need to test more carefully. Starknet went furthest: it doesn't run [[gloss:solidity|Solidity]] at all, it runs a language called [[gloss:cairo|Cairo]] that was designed from scratch to be cheap to prove. [[gloss:cairo|Cairo]] costs the team an ecosystem ([[gloss:solidity|Solidity]] has hundreds of thousands of developers; [[gloss:cairo|Cairo]] has thousands), but it gives Starknet a proving advantage that's part of why the chain could afford to decentralize its [[gloss:sequencer|sequencer]] first. :::

:::p{html} The point. When a chain says "[[gloss:solidity|Solidity]] contracts work here," the same sentence means three different things. On Scroll it means: deploy your audited contract, change nothing, it behaves identically. On zkSync it means: deploy it, test it carefully, watch the low-level patterns. On Starknet it means: rewrite it in [[gloss:cairo|Cairo]], or transpile and live with the risk. The labels look interchangeable from a marketing page; in practice, choosing the wrong type for your application can cost weeks of unexpected porting or, worse, subtle bugs that only show up in production. Every team that builds on a [[gloss:zkevm|zkEVM]] eventually meets this spectrum, and the sooner you know where your target sits, the cheaper the project gets. The whole geometry of the trade between compatibility and efficiency is a specific instance of the [[gloss:blockchain|blockchain]] trilemma — see → 2.4 §7 for the general pattern. ::: :::

:::quiz{id=q-2-5-s4 correct=1 explanation="Correct. The chain wasn't broken in any technical sense — [[gloss:solidity|Solidity]] contracts deployed, proofs were generated, settlements landed on [[gloss:ethereum|Ethereum]]. What killed it was economics. After [[gloss:eip-4844|EIP-4844]] introduced [[gloss:blob|blobs]] in March 2024, every competitive [[gloss:layer 2|L2]] migrated to them to slash data-availability costs by 90% or more. Polygon [[gloss:zkevm|zkEVM]] never did, so its per-[[gloss:transaction|transaction]] costs stayed high relative to peers. Combined with a strategic pivot at Polygon toward AggLayer and the [[gloss:proof of stake|PoS]] chain, the [[gloss:zkevm|zkEVM]] team was deprioritized and eventually wound down. Option A is wrong, the [[gloss:cryptography|cryptography]] works fine. Option C is wrong, the chain ran on [[gloss:mainnet|mainnet]] from March 2023 to July 2026. Option D is wrong, Polygon [[gloss:zkevm|zkEVM]] was always a proper [[gloss:zk rollup|ZK rollup]] that posted data and proofs to [[gloss:ethereum|Ethereum]]."} Why does Polygon [[gloss:zkevm|zkEVM]]'s shutdown count as a cautionary tale rather than a vindication of the ZK approach?

Because the proving math turned out to be broken and was quietly being subsidized by Polygon
Because the chain technically worked but couldn't pay for its own operations — most directly because the team never upgraded it to use [[gloss:blob|blobs]] after [[gloss:eip-4844|EIP-4844]], leaving its data-availability costs uncompetitive while Polygon's strategy shifted elsewhere
Because Polygon [[gloss:zkevm|zkEVM]] was never actually running on [[gloss:mainnet|mainnet]], just on a long-running testnet
Because Polygon [[gloss:zkevm|zkEVM]] was actually a sidechain, not a [[gloss:zk rollup|ZK rollup]], and [[gloss:l2beat|L2Beat]] reclassified it :::

:::section{mins=2} OP Stack and the Superchain :::

:::lead Optimism made a bet that the future of [[gloss:ethereum|Ethereum]] scaling wouldn't be a single chain but a federation of compatible chains sharing one codebase, one governance, and one funding pool. The [[gloss:superchain|Superchain]] is what that bet looks like in production: more than thirty chains running on the [[gloss:op stack|OP Stack]], a revenue share that funds public infrastructure, and a long-promised interop layer that would let them feel like one chain. In May 2026 the federation works, the public goods funding works, and the interop is still on devnet — and the most strategically important member just announced it's leaving. :::

:::p{html} The [[gloss:op stack|OP Stack]] is the open-source software for running an [[gloss:ethereum|Ethereum]] [[gloss:layer 2|L2]]. It came out of Optimism's own development of OP [[gloss:mainnet|Mainnet]] and was released as a public good in 2023. Any team can [[gloss:fork|fork]] it and launch a chain with a few engineers and a few weeks of work instead of a few years and a research department. The stack handles all the unglamorous things a [[gloss:rollup|rollup]] needs: :::

:::p{html}

Sequencing — ordering [[gloss:transaction|transactions]] as they arrive on [[gloss:layer 2|L2]].
Batch posting — publishing those [[gloss:transaction|transactions]] to [[gloss:l1|L1]] as [[gloss:blob|blobs]].
Fault proofs — the on-chain machinery for challenging dishonest batches.
The [[gloss:l1|L1]]↔[[gloss:layer 2|L2]] bridge — moving assets in and out of the [[gloss:rollup|rollup]].

:::

:::p{html} Each chain that runs the [[gloss:op stack|OP Stack]] is still a separate [[gloss:layer 2|L2]] — its own state, its own [[gloss:sequencer|sequencer]], its own [[gloss:tvl|TVL]] — but the underlying code is shared upstream, so when Optimism's engineers ship a fault-proof improvement, every [[gloss:superchain|Superchain]] chain inherits it. :::

:::p{html} The [[gloss:superchain|Superchain]] is the federation of [[gloss:op stack|OP Stack]] chains that have additionally opted into a common governance and economic agreement called the [[gloss:law of chains|Law of Chains]]. As of mid-2026 it includes Base (until its announced departure), OP [[gloss:mainnet|Mainnet]], Unichain (Uniswap), Ink (Kraken), Soneium (Sony), World, Celo, Lisk, Zora, Mode, and twenty-odd others. The [[gloss:law of chains|Law of Chains]] commits members to three things: :::

:::p{html}

Shared upgrades — managed by a [[gloss:security council|Security Council]] with representatives from OP Labs, Coinbase, and external members.
Revenue share — every member pays the Optimism Collective the greater of 2.5% of net [[gloss:sequencer|sequencer]] revenue or 15% of on-chain profit.
Neutrality and free exit — a commitment that users can always exit the chain and that [[gloss:block|block]] ordering stays open.

:::

:::p{html} The revenue share funds the next round of [[gloss:retropgf|RetroPGF]], paying teams that have already shipped public goods. In Q1 2026 alone, Base's contribution was roughly $1.4M and the Collective distributed $14M in total to public-goods projects. :::

:::p{html} The missing piece is [[gloss:native interop|native interop]] — the part that would actually make the [[gloss:superchain|Superchain]] feel like one chain instead of a federation. The goal is three properties working together: :::

:::p{html}

Native asset [[gloss:movement|movement]] — [[gloss:ethereum|ETH]] and ERC-20 tokens [[gloss:move|move]] between [[gloss:superchain|Superchain]] members as mint-and-burn, not as wrapped tokens.
Cross-chain composition — apps on one chain can read and write state on another inside a single [[gloss:transaction|transaction]], in under two seconds.
Shared fault proofs — a fraud challenge resolved on one chain settles consistently across the whole federation.

:::

:::p{html} The roadmap has been in progress since 2024. Upgrade 16a landed on [[gloss:mainnet|mainnet]] in October 2025 as the foundation. [[gloss:native interop|Native interop]] went live on devnet in April 2026. [[gloss:mainnet|Mainnet]] deployment is targeted for a [[gloss:pectra|Pectra]]-aligned upgrade later in 2026. Until then, moving between [[gloss:superchain|Superchain]] chains still requires explicit bridging through third-party protocols, which is the same friction users would face between any two unrelated [[gloss:layer 2|L2s]]. :::

:::p{html} And then there's the Base announcement. In February 2026, the largest member of the [[gloss:superchain|Superchain]] — and a co-founder of the whole vision — said it would migrate off the [[gloss:op stack|OP Stack]] onto its own self-managed codebase. Base will continue contributing the revenue share for now, and the technical drift will happen gradually rather than as a [[gloss:hard fork|hard fork]]. But the message to other [[gloss:superchain|Superchain]] members is unambiguous: at sufficient scale, the shared codebase stops being a benefit and starts being a constraint. Whether smaller members follow, whether the [[gloss:op stack|OP Stack]] reabsorbs Base's improvements over time, and whether interop arrives in time to make staying worthwhile — these are the questions that decide what the [[gloss:superchain|Superchain]] actually is by the end of 2026. :::

:::callout{icon=💡} Think of a professional sports league. Each team is independently owned — different cities, different colors, different ownership groups, different home arenas — but they all play by the same rulebook, follow the same calendar, and are bound by the same player conduct standards. There's a central league office that sets the rules and arbitrates disputes. A slice of every ticket sale and broadcasting deal flows into a league-wide pool, which is then redistributed: some to revenue sharing among smaller-market teams, some to youth development, some to officiating and infrastructure. The teams compete fiercely against each other on the court — but they need each other off it, because the league brand is the thing that makes any individual team valuable. Now imagine the biggest, most valuable franchise in the league announces it's starting its own competing league next season. It will keep paying its current obligations, the divorce is amicable, the players still play. But every other owner has to look at the situation and ask: if the headline team thinks it's better off alone, what does that say about the rest of us? Some will follow. Some will double down on the league brand and try to take the vacated spotlight. The league will be a different league afterwards either way. :::

:::keypoints{label="Key Points"}

[[gloss:op stack|OP Stack]] and [[gloss:superchain|Superchain]] aren't the same thing. [[gloss:op stack|OP Stack]] is the software (anyone can use it, no strings); [[gloss:superchain|Superchain]] is the federation (governance plus revenue share). Forking the [[gloss:op stack|OP Stack]] without joining the [[gloss:superchain|Superchain]] is allowed and common
The economic model rewards public goods retroactively, not in advance. Teams build first, the Collective looks at what was useful, and pays for it later — a counterintuitive structure that has so far paid out tens of millions of dollars
"Interop" has been the promise behind the [[gloss:superchain|Superchain]] since 2023, but as of May 2026 it's still on devnet. Until it lands on [[gloss:mainnet|mainnet]], the [[gloss:superchain|Superchain]] is more of an upgrade-sharing club than a single user-facing chain
Base's departure isn't a technical breakup — Base will keep paying the revenue share and the chains stay compatible for now. It's a statement about scale: the biggest chain in the federation decided it's outgrown the shared codebase, and that signal travels :::

:::details{badge=Advanced title="The Security Council, "Stage 1," and the centralization that funds decentralization"} :::p{html} Every [[gloss:superchain|Superchain]] member has reached [[gloss:l2beat stage|L2BEAT Stage]] 1, which sounds reassuring until you read the fine print. [[gloss:l2beat stage|Stage 1]] means fault proofs work end-to-end and a public set of challengers can submit them — but it explicitly allows a [[gloss:security council|security council]] to pause or upgrade contracts unilaterally. [[gloss:l2beat stage|Stage 2]] requires that the council either be removed or be limited to genuinely emergency-only intervention. As of mid-2026, no major [[gloss:layer 2|L2]] has reached [[gloss:l2beat stage|Stage 2]], and the [[gloss:superchain|Superchain]] explicitly relies on its council for ongoing operations. :::

:::p{html} The [[gloss:superchain|Superchain]] [[gloss:security council|Security Council]] has representatives from OP Labs, Coinbase, and several external members. It holds upgrade authority across all member chains — meaning, in practice, that any chain's bridge contracts can be paused or upgraded if the council [[gloss:multisig|multisig]] signs. This is the explicit centralization tradeoff: chains get shared security audits, shared fault-proof upgrades, and a fast response path if something goes wrong on any member chain. They give up the property that no small group can intervene in their contracts. For Base, this was acceptable because Coinbase was on the council and the council acted in their interest. For Base post-departure, the calculus changes, which is part of why the departure makes sense beyond just engineering autonomy. :::

:::p{html} The economic argument for accepting this centralization is honest, if uncomfortable. Truly decentralized [[gloss:layer 2|L2]] development is extremely expensive: someone has to maintain the fault proof code, respond to vulnerabilities, ship upgrades that keep pace with [[gloss:l1|L1]] ([[gloss:pectra|Pectra]], [[gloss:fusaka|Fusaka]], future hard forks). If every [[gloss:layer 2|L2]] has to do this work independently, only the very largest can afford it. The [[gloss:op stack|OP Stack]] model concentrates that work in OP Labs and the council, splits the costs across the federation via the revenue share, and lets smaller chains exist that otherwise wouldn't. The result is that [[gloss:superchain|Superchain]] is decentralized at the federation level — chains can join and leave — while remaining moderately centralized at the upgrade level. [[gloss:retropgf|RetroPGF]] is the moral compensation: the centralized upgrade authority funds itself by paying public-goods builders retroactively, so the value created by the federation flows back to the commons. :::

:::p{html} The point. When you see "[[gloss:l2beat stage|Stage 1]]" next to a [[gloss:superchain|Superchain]] member on [[gloss:l2beat|L2BEAT]], that label is doing more work than it looks like. It says the [[gloss:cryptography|cryptography]] is sound, the challengers are public, the path to [[gloss:fraud proof|fraud proofs]] is open. It does not say the chain is operationally independent of a small [[gloss:multisig|multisig]]. That's not a hidden flaw — it's the explicit deal, and it's part of why the federation can ship at the speed it does. But the moment you stop trusting the council, or the moment the council's composition changes in a way you didn't expect, the security model changes underneath you. ::: :::

:::quiz{id=q-2-5-s5 correct=2 explanation="Correct. [[gloss:op stack|OP Stack]] is MIT-licensed open source — a team can [[gloss:fork|fork]] it tomorrow, run a private [[gloss:layer 2|L2]], and never speak to Optimism again. The [[gloss:superchain|Superchain]] is what you opt into on top: you sign the [[gloss:law of chains|Law of Chains]], you accept that the [[gloss:security council|Security Council]] can pause your contracts, and you commit to the revenue share. The benefits are shared upgrades, shared audits, eventually shared interop, and the legitimacy of the OP brand. The costs are governance constraints and the fee. Option A is wrong — many [[gloss:op stack|OP Stack]] chains run completely outside the [[gloss:superchain|Superchain]]. Option B is wrong — there's no fee for the [[gloss:op stack|OP Stack]] code itself, only for the [[gloss:superchain|Superchain]] opt-in. Option D is wrong — both [[gloss:superchain|Superchain]] and standalone [[gloss:op stack|OP Stack]] chains use the same optimistic-[[gloss:rollup|rollup]] fault proofs; no ZK upgrade is part of the [[gloss:superchain|Superchain]] agreement (yet)."} What's the practical difference between forking the [[gloss:op stack|OP Stack]] and joining the [[gloss:superchain|Superchain]]?

There's no difference — using [[gloss:op stack|OP Stack]] code automatically makes a chain part of the [[gloss:superchain|Superchain]]
Forking the [[gloss:op stack|OP Stack]] requires a license fee paid to OP Labs; joining the [[gloss:superchain|Superchain]] waives that fee
The [[gloss:op stack|OP Stack]] is open-source software that anyone can [[gloss:fork|fork]] freely. Joining the [[gloss:superchain|Superchain]] is an additional opt-in agreement that adds shared governance, a [[gloss:security council|Security Council]] with upgrade authority, and a revenue share to the Optimism Collective (2.5% of net [[gloss:sequencer|sequencer]] revenue or 15% of profit)
Joining the [[gloss:superchain|Superchain]] gives a chain access to ZK proofs; forking the [[gloss:op stack|OP Stack]] alone keeps it on optimistic proofs only :::

:::section{mins=2} Arbitrum Orbit, Stylus, and L3 :::

:::lead If Optimism's bet was a federation of equal chains under a shared codebase, Arbitrum's bet was the opposite: a single deep flagship and a permissioned framework for derivative chains that settle back to it. [[gloss:arbitrum orbit|Orbit]] chains are dedicated [[gloss:layer 2|L2s]] and L3s on Arbitrum's Nitro stack, with a commercial license and a clear economic tie to the parent; alongside that, [[gloss:stylus|Stylus]] added Rust, C, and C++ as first-class languages next to the [[gloss:evm|EVM]]. The contrast with [[gloss:op stack|OP Stack]] tells you most of what you need to know about how the two ecosystems think. :::

:::p{html} [[gloss:arbitrum orbit|Arbitrum Orbit]] is a framework for launching your own chain on top of Arbitrum's Nitro stack. The team that runs it gets a fully customizable [[gloss:rollup|rollup]] — own [[gloss:gas|gas]] token, own throughput targets, own governance, own privacy posture — that settles either to Arbitrum One, [[gloss:arbitrum nova|Arbitrum Nova]], or directly to [[gloss:ethereum|Ethereum]] [[gloss:l1|L1]]. [[gloss:arbitrum nova|Arbitrum Nova]] is worth introducing separately: it's Offchain Labs' second chain, launched in August 2022 and aimed at gaming and social applications. Nova is cheaper than Arbitrum One because it stores data not as [[gloss:blob|blobs]] on [[gloss:l1|L1]] but in a [[gloss:data availability|Data Availability]] Committee (the advanced [[gloss:block|block]] below covers what that means). Arbitrum One holds the deep capital and serious [[gloss:defi|DeFi]]; Nova trades some [[gloss:ethereum|Ethereum]]-grade [[gloss:data availability|data availability]] for very low fees and very high throughput. By May 2026 there are over fifty [[gloss:arbitrum orbit|Orbit]] chains in development and roughly twenty live on [[gloss:mainnet|mainnet]], with two of them — XAI and Sanko — accounting for over 99% of [[gloss:arbitrum orbit|Orbit]] ecosystem [[gloss:tvl|TVL]]. :::

:::p{html} The most important nuance about [[gloss:arbitrum orbit|Orbit]] is its license. Nitro, the underlying codebase, is released under the Business Source License, not MIT. In practice this means a team can launch an [[gloss:arbitrum orbit|Orbit]] chain permissionlessly only if it settles to Arbitrum One or [[gloss:arbitrum nova|Arbitrum Nova]]. If a team wants to take the Nitro code and run a chain that settles directly to [[gloss:ethereum|Ethereum]] [[gloss:l1|L1]] (bypassing Arbitrum), or that settles to another base chain entirely, they need to sign the Arbitrum Expansion Program — an explicit agreement with Offchain Labs that includes commercial terms. This is the opposite philosophy from the MIT-licensed [[gloss:op stack|OP Stack]] we covered in → §5. Optimism gave its codebase away and built revenue capture into a separate federation contract. Arbitrum kept its codebase commercial and built revenue capture into the license itself. :::

:::p{html} [[gloss:l3|L3]] in Arbitrum's vocabulary means a chain that settles to Arbitrum One (an [[gloss:layer 2|L2]]) rather than to [[gloss:ethereum|Ethereum]] (the [[gloss:l1|L1]]). The motivation is specialization: a high-throughput gaming chain doesn't need to share blockspace with a [[gloss:stablecoin|stablecoin]] DEX, and putting it on its own dedicated chain that settles to Arbitrum One gives it independent throughput while still inheriting some of Arbitrum's security guarantees. XAI is the canonical [[gloss:l3|L3]] — a gaming chain co-developed by Offchain Labs and Ex Populus, with its own token, its own [[gloss:validator|validator]] set, and [[gloss:block|block]] times under 100 milliseconds. ApeChain, launched October 2024, is an [[gloss:l3|L3]] dedicated to the ApeCoin ecosystem with APE as its [[gloss:gas|gas]] token. Pirate Nation, a fully on-chain strategy game from Proof of Play, runs on its own [[gloss:l3|L3]] with 250-millisecond [[gloss:block|blocks]] and over 2.5 million players. The [[gloss:l3|L3]] pattern works when you're optimizing for a single application; it stops working when you're trying to be a generic [[gloss:layer 2|L2]]. :::

:::p{html} [[gloss:stylus|Stylus]] is the other half of Arbitrum's horizontal-scaling story. Available on Arbitrum One since June 2024, [[gloss:stylus|Stylus]] lets developers write [[gloss:smart contract|smart contracts]] in Rust, C, C++, or any other language that compiles to WebAssembly, and runs them in a second virtual machine alongside the [[gloss:evm|EVM]]. [[gloss:stylus|Stylus]] contracts and [[gloss:solidity|Solidity]] contracts can call each other directly, share state, and live in the same [[gloss:wallet|wallet]]. The pitch is twofold: existing Web2 engineers can use languages they already know instead of learning [[gloss:solidity|Solidity]], and computationally heavy operations ([[gloss:cryptography|cryptography]], signal processing, game logic) run 10–100× cheaper in [[gloss:wasm|WASM]] than in the [[gloss:evm|EVM]]. As of 2026 several production protocols use [[gloss:stylus|Stylus]] for the hot paths in their critical math while leaving the asset-handling layer in [[gloss:solidity|Solidity]] — a mixed-language pattern that didn't exist anywhere in [[gloss:evm|EVM]]-land a year earlier. :::

:::callout{icon=💡} Picture an established watchmaking workshop with a master craftsman who has spent decades developing a proprietary set of tools and techniques. The master takes on apprentices, but on specific terms. An apprentice gets access to the full toolset and the workshop's brand on her own pieces, as long as every watch she makes still passes through the master's bench for final inspection and stamping — that's how the apprentice can charge premium prices, because the master's stamp is what collectors trust. If an apprentice wants to set up her own shop in another city and use the same techniques there, she has to negotiate a separate license with the master, with explicit terms about which trademarks she can use and what royalty she pays. Meanwhile, the master has also added a Rust-trained machinist to the workshop, who specializes in the small intricate gears that the traditional master craftsmen find tedious — and the apprentices can now hire that machinist for their hot precision parts while doing the rest of the watch in the traditional way. The workshop scales by adding apprentices and specialists, not by handing out its toolset for free. :::

:::keypoints{label="Key Points"}

The same engineering problem — how to monetize a [[gloss:rollup|rollup]] framework — produced two opposite legal answers: MIT-licensed open source plus a separate revenue-sharing federation (Optimism), or commercial source-available licensing where the revenue capture is the license itself (Arbitrum). Same code surface, different control structures
"[[gloss:l3|L3]]" isn't a deeper layer in some Russian-doll sense. It's a marketing word for one specific architecture choice: where the chain posts its data and proofs. An [[gloss:l3|L3]] still anchors to [[gloss:ethereum|Ethereum]]'s [[gloss:data availability|data availability]] — it just routes through an [[gloss:layer 2|L2]] on the way there, adding one extra trust hop in exchange for an isolated execution lane
The [[gloss:l3|L3]] pattern works for application-specific chains (gaming, [[gloss:nft|NFT]] communities, a specific ecosystem like ApeCoin) and doesn't work for general-purpose [[gloss:layer 2|L2s]]. Pick this path when you know exactly what your chain is optimizing for
[[gloss:stylus|Stylus]] is not a replacement for [[gloss:solidity|Solidity]], it's a coprocessor. The realistic pattern is to keep asset-handling logic in audited [[gloss:solidity|Solidity]] and [[gloss:move|move]] only the computationally expensive parts (math, [[gloss:cryptography|cryptography]], game logic) to Rust or C — and gain 10–100× efficiency on those hot paths :::

:::details{badge=Advanced title="AnyTrust, the DAC, and why most "L3s" are actually validiums in everything but name"} :::p{html} The reason an [[gloss:l3|L3]] like XAI can offer 100-millisecond [[gloss:block|blocks]] while a regular [[gloss:rollup|rollup]] runs at 250ms or slower is that it usually isn't a [[gloss:rollup|rollup]] in the strict sense — it's running a different data-availability model called [[gloss:anytrust|AnyTrust]]. Understanding why matters for understanding how much of [[gloss:ethereum|Ethereum]]'s security an [[gloss:l3|L3]] actually inherits. :::

:::p{html} Recall that a true [[gloss:rollup|rollup]] posts all of its [[gloss:transaction|transaction]] data to [[gloss:l1|L1]] (as [[gloss:blob|blobs]] since [[gloss:dencun|Dencun]]), so anyone with the [[gloss:l1|L1]] chain can rebuild the [[gloss:layer 2|L2]] state from scratch. This is what makes a [[gloss:rollup|rollup]] "secure": even if every [[gloss:layer 2|L2]] [[gloss:sequencer|sequencer]] disappears, the data is still on [[gloss:ethereum|Ethereum]]. The cost is the [[gloss:l1|L1]] fee for those [[gloss:blob|blobs]], which for a high-throughput chain becomes the dominant operational expense. :::

:::p{html} [[gloss:anytrust|AnyTrust]] changes this. Instead of posting all data to [[gloss:l1|L1]], the chain posts data to a [[gloss:data availability|Data Availability]] Committee (DAC) — a small fixed group of trusted parties (typically 5-10) who promise to store the data and serve it on request. Only a [[gloss:hash|hash]] of the data goes to [[gloss:l1|L1]], plus a signed attestation from the DAC that the data is available. As long as at least one DAC member is honest and serves the data when asked, the chain's state can be reconstructed. If every single DAC member colludes or goes offline, the data is lost and users may not be able to withdraw. [[gloss:arbitrum nova|Arbitrum Nova]] uses [[gloss:anytrust|AnyTrust]], XAI uses [[gloss:anytrust|AnyTrust]], and most application-specific L3s use [[gloss:anytrust|AnyTrust]] because it's dramatically cheaper than posting [[gloss:blob|blobs]]. :::

:::p{html} This is the same architectural choice that powers validiums: a [[gloss:zk rollup|ZK rollup]] that posts proofs to [[gloss:l1|L1]] but not [[gloss:transaction|transaction]] data is called a [[gloss:validium|validium]]. The DAC is the optimistic-[[gloss:rollup|rollup]] version of that choice. Both buy lower fees by giving up [[gloss:ethereum|Ethereum]]-grade [[gloss:data availability|data availability]], and both ask the user to trust a small committee rather than trusting only math and [[gloss:blob|blob]] storage. [[gloss:l2beat|L2BEAT]] rates [[gloss:anytrust|AnyTrust]] chains and validiums lower than true [[gloss:rollup|rollups]] for exactly this reason — the trust assumption is real, even if the committee is reputable. :::

:::p{html} The point. When marketing materials describe an "[[gloss:l3|L3]] settling to Arbitrum One," that phrase usually elides the question of where the data is. If the answer is "DAC of five companies," then the chain is fast and cheap, but it's also asking the user to trust those five companies in a way a true [[gloss:rollup|rollup]] doesn't. This isn't necessarily wrong — for a high-frequency game with assets that aren't bank-account-sized, the tradeoff is sensible. But it's a tradeoff, and "[[gloss:l3|L3]] on Arbitrum" doesn't tell you which side of it the chain landed on. ::: :::

:::quiz{id=q-2-5-s6 correct=0 explanation="Correct. Nitro's Business Source License is a deliberate design choice that distinguishes Arbitrum from Optimism. Permissionless usage of the Nitro code is granted only when the resulting chain settles to Arbitrum One or [[gloss:arbitrum nova|Arbitrum Nova]] — meaning the parent ecosystem captures the relationship. Anyone wanting to use Nitro outside that relationship needs to sign the Arbitrum Expansion Program, an explicit commercial license with Offchain Labs. Option B is wrong — the technical interaction isn't the issue, it's the legal license. Option C is the [[gloss:op stack|OP Stack]] situation, not Arbitrum's. Option D is a confusion between technical capability and licensing rights; Nitro can technically settle anywhere, but the license restricts where it's allowed to."} If you [[gloss:fork|fork]] the Nitro codebase to launch a new chain that settles directly to [[gloss:ethereum|Ethereum]] [[gloss:l1|L1]] instead of Arbitrum One, what happens?

Nitro is released under the Business Source License, which makes permissionless deployment available only when the new chain settles to Arbitrum One or Nova. Settling elsewhere requires signing the Arbitrum Expansion Program with Offchain Labs — an explicit commercial agreement with terms and fees
The chain will technically run but won't be able to interact with any other Arbitrum infrastructure
Nothing special — Nitro is MIT-licensed and anyone can use it however they want
Settling directly to [[gloss:l1|L1]] is technically impossible without using [[gloss:op stack|OP Stack]], since only [[gloss:op stack|OP Stack]] has the necessary contracts :::

:::section{mins=2} Bridges, Withdrawal Risk, and Cross-L2 :::

:::lead An [[gloss:layer 2|L2]] inherits [[gloss:ethereum|Ethereum]]'s security only as far as its [[gloss:canonical bridge|canonical bridge]] — the contract on [[gloss:l1|L1]] that holds user deposits and processes withdrawals. Everything else is third-party infrastructure with its own trust assumptions. Bridges have been the single most damaging attack surface in crypto: roughly $2.8 billion in cumulative losses, about 40% of all Web3 exploits. The category has gotten safer since 2022, but understanding why bridges fail — and what the seven-day withdrawal window on [[gloss:optimistic rollup|optimistic rollups]] actually protects you from — is unavoidable if you're moving anything serious between chains. :::

:::p{html} There are four common bridge architectures, and the security and UX they offer differ enormously: :::

:::p{html}

Canonical (lock-and-mint) — the [[gloss:rollup|rollup]]'s own bridge contract holds your asset on [[gloss:l1|L1]] and issues a 1:1 representation on [[gloss:layer 2|L2]] (or vice versa for withdrawal). Security inherits from [[gloss:l1|L1]] — if the [[gloss:rollup|rollup]]'s fraud or [[gloss:validity proof|validity proof]] is sound, your assets are recoverable. Cost: time. [[gloss:optimistic rollup|Optimistic rollups]] force a seven-day withdrawal window; [[gloss:zk rollup|ZK rollups]] settle within hours.
Burn-and-mint native — protocols like Circle's [[gloss:cctp|CCTP]] and LayerZero's USDT0 burn the asset on the source chain and mint a fresh native version on the destination, with no wrapped IOU in between. Output is the same canonical contract Tether or Circle issues directly. [[gloss:cctp|CCTP]] V2 clears in under 15 seconds; trust is in the issuer's verification network.
Liquidity pool — Stargate, Synapse, Hop. Two chains each hold a pool of the same asset; your [[gloss:transaction|transaction]] takes from one and gives from the other. No asset actually moves cross-chain. Fast and cheap, but constrained by pool depth — large transfers slip, very large transfers can't fill.
[[gloss:intent-based|Intent-based]] — Across, Eco Portal, deBridge. You declare an outcome ("I want 1,000 USDC on Base"); a third-party solver fronts that liquidity instantly from its own balance and gets reimbursed from your source chain a minute later through an optimistic-oracle settlement. 2–30 second clears, lowest fees on most routes, but the trust assumption shifts to the solver network.

:::

:::p{html} Why bridges keep getting hacked. Almost every major incident traces to one of two failure modes: a small set of signers compromised (Ronin's five-of-nine [[gloss:validator|validator]] keys, [[gloss:arbitrum orbit|Orbit]] Chain's seven-of-ten [[gloss:multisig|multisig]]), or a flaw in the cross-chain verification logic that lets an attacker fabricate proofs (Wormhole's signature bypass, Nomad's Merkle root misconfiguration, BNB Bridge's IAVL proof forgery). Both classes share a root cause: a bridge is the only place in crypto where assets on one chain depend on attestations from a system that the destination chain cannot independently verify. Any weakness in the attestation pipeline becomes a way to print free money. The 2022 wave hit $2 billion in bridge losses. By 2025 that number was down toward $300 million, mostly because MPC migrated to TEE-backed signing, light-client bridges matured, and critical contracts adopted formal verification — but it hasn't gone to zero, and it likely won't. :::

:::p{html} The seven-day window, in practice. When you withdraw from an optimistic [[gloss:layer 2|L2]] through its [[gloss:canonical bridge|canonical bridge]], you wait approximately seven days before the funds settle on [[gloss:l1|L1]] (about twelve on Arbitrum after [[gloss:bold|BOLD]], see → §2). The window isn't bureaucratic — it's the time during which honest watchers can detect a fraudulent withdrawal and stop it. If you don't want to wait, you can use a liquidity-pool or [[gloss:intent-based bridge|intent-based bridge]] that gives you the asset on [[gloss:l1|L1]] immediately, but the protocol fronting that liquidity takes the seven-day risk on your behalf — and charges 0.05–0.3% for it. The window doesn't disappear; somebody is always carrying it. :::

:::p{html} Cross-[[gloss:layer 2|L2]] [[gloss:movement|movement]] is still bridging. Moving from Arbitrum to Base isn't "[[gloss:layer 2|L2]] interop" — it's two bridge operations in sequence (Arbitrum [[gloss:layer 2|L2]] → [[gloss:ethereum|Ethereum]] [[gloss:l1|L1]] → Base [[gloss:layer 2|L2]]), or a third-party bridge that batches them into a single user-facing [[gloss:transaction|transaction]]. Until [[gloss:native interop|native interop]] ships on the [[gloss:superchain|Superchain]] (devnet as of April 2026, see → §5) and similar work on other ecosystems, every cross-[[gloss:layer 2|L2]] hop carries the trust assumptions of whichever bridge you used. The "multi-chain world" most users experience is held together by [[gloss:intent-based bridge|intent-based bridges]] and [[gloss:cctp|CCTP]], not by anything that would honestly call itself unified. :::

:::callout{icon=💡} Think of a busy international port where ships from different shipping lines unload at different terminals. None of the cranes from terminal A can directly reach terminal B — the terminals have separate operators, separate manifests, separate insurance. Moving a container from one to the other can happen four ways. You can leave the original container locked in a warehouse at terminal A and walk out of terminal B with a paper receipt that says "we have your container next door"; that's lock-and-mint, and your goods are only as safe as that warehouse and that receipt. You can have terminal A's customs officially destroy the container's papers and have terminal B's customs issue brand-new papers for the same goods, with both sides talking through a verified inter-port radio; that's burn-and-mint, the way Circle's [[gloss:cctp|CCTP]] works. You can ignore your own container entirely and just take an identical one out of terminal B's pre-stocked pool, paying terminal A to top its own pool back up later; that's a liquidity bridge, fast but only as deep as the pool. Or you can tell a freight broker "I need a container of these exact goods at terminal B by tonight" and let the broker hand you one out of their own stock immediately, then settle with terminal A behind the scenes — that's [[gloss:intent-based|intent-based]] bridging, where the trust is in the broker, not the terminals. Most of the dramatic bridge thefts in crypto have been the equivalent of forged warehouse receipts in option one. :::

:::keypoints{label="Key Points"}

The [[gloss:canonical bridge|canonical bridge]] of a [[gloss:rollup|rollup]] and a third-party bridge are not the same animal. Canonical inherits [[gloss:l1|L1]] security; third-party introduces its own trust assumptions on top. Mixing them up is how non-technical users have lost the most money
The seven-day withdrawal window can't be removed without weakening security — but it can be moved. Fast-exit bridges shift the risk onto a third party and charge 0.05–0.3% for absorbing it; that fee is the price of the wait, not of the transfer
Bridge hack losses fell from ~$2B in 2022 to ~$300M in 2025 not because attackers gave up, but because the industry adopted TEE-signed MPC, light clients, and formal verification on critical contracts. The category is safer; it isn't safe
"Cross-[[gloss:layer 2|L2]]" without [[gloss:native interop|native interop]] is just bridging dressed up: every hop between [[gloss:layer 2|L2s]] still passes through [[gloss:ethereum|Ethereum]] [[gloss:l1|L1]] or a third-party intermediary. When the UI feels like one unified chain, what you're trusting is the aggregator, not any underlying protocol-level unity :::

:::details{badge=Advanced title="Anatomy of the Ronin hack, and what TEE-signed MPC actually fixed"} :::p{html} The Ronin bridge hack of March 2022 lost $625 million in a single [[gloss:transaction|transaction]]. The mechanism was almost embarrassingly simple, and worth understanding because every later improvement to bridge architecture is a response to it. Ronin was the [[gloss:canonical bridge|canonical bridge]] for Axie Infinity, a sidechain operated by Sky Mavis. Its security model was a [[gloss:multisig|multisig]]: nine [[gloss:validator|validators]], of which five had to sign for any withdrawal from the [[gloss:l1|L1]] bridge contract. Four of those nine [[gloss:validator|validators]] were operated by Sky Mavis directly. A fifth was operated by Axie DAO. In November 2021, to handle a temporary load spike, Axie DAO had delegated its signing key to Sky Mavis's infrastructure on a "just for now" basis. That delegation was never revoked. :::

:::p{html} By March 2022, five of nine [[gloss:validator|validator]] keys were therefore controlled by a single operator. An attacker compromised Sky Mavis's IT infrastructure (the leading theory is a social-engineered job offer to an employee with the relevant access), retrieved the keys, and signed a withdrawal of $625 million in [[gloss:ethereum|ETH]] and USDC to their own [[gloss:l1|L1]] address. The bridge contract did exactly what it was designed to do: it saw five valid signatures and released the funds. Six days passed before anyone at Sky Mavis noticed. :::

:::p{html} Two things in this story matter for understanding modern bridge design. First: the threshold (5 of 9) was both the cryptographic guarantee and the entire security model. Once that threshold falls, the bridge has no other defense. Second: the keys were stored in a way that a corporate attacker could reach. They were on regular servers, accessible to Sky Mavis engineers, encrypted at rest but decryptable in process memory at signing time. The fix that emerged across the industry over the next three years addresses both. :::

:::p{html} TEE-signed MPC (Trusted Execution Environment, e.g., Intel SGX or AWS Nitro Enclaves) means each [[gloss:validator|validator]]'s signing key never exists in plaintext outside a hardware-isolated enclave. Even an attacker with root access to the [[gloss:validator|validator]]'s server cannot extract the key — they can only request signatures, which the TEE refuses unless its attached policy allows them (rate limits, destination allowlists, anomaly detection). Combined with threshold MPC, where no [[gloss:validator|validator]] ever holds a complete key in the first place, the attack surface shrinks from "compromise one machine" to "compromise enough enclaves simultaneously to reconstruct a key, against active policy enforcement." That's an order of magnitude harder. The remaining attack surfaces are now in the cross-chain verification logic itself — which is why bridges that adopted formal verification (Mathematical proofs that the bridge code does exactly what its spec says, no more) have become the safer category in 2026. :::

:::p{html} The point. Ronin wasn't a sophisticated zero-day. It was a key-management mistake amplified by a [[gloss:multisig|multisig]] with too few keys, all in one operator's reach. Modern bridges aren't safe because attackers got worse; they're safer because the keys themselves became harder to steal, and the verification logic became harder to bypass. When a bridge advertises "decentralized validation," what matters isn't the number of signers on the diagram — it's three practical questions: how the keys are stored, what guardrails fire before each signature (rate limits, destination allowlists, anomaly detection), and whether the cross-chain verification logic has been formally proven. ::: :::

:::quiz{id=q-2-5-s7 correct=3 explanation="Correct. [[gloss:cctp|CCTP]] is the cleanest answer because it doesn't add a new trust assumption on top of the one you already have: if you hold USDC, you already trust Circle. Native burn/mint avoids wrapped tokens, third-party liquidity, and solver networks entirely. Settlement is sub-15 seconds on most routes. Option A is correct on security but wrong on speed — canonical [[gloss:l1|L1]]↔[[gloss:layer 2|L2]] bridging through both chains takes far longer than a minute. Option B is wrong on the security claim — Stargate adds LayerZero's [[gloss:validator|validator]]/oracle trust assumption on top of Circle's, which is a strict expansion of the trust surface. Option C is wrong because [[gloss:intent-based bridge|intent-based bridges]] shift the trust assumption to the solver network and the optimistic oracle — fast and cheap, but not identical to canonical security."} You want to [[gloss:move|move]] 5,000 USDC from Arbitrum to Base in under a minute. Which option preserves the security guarantees closest to using the [[gloss:canonical bridge|canonical bridge]]?

Use Arbitrum's [[gloss:canonical bridge|canonical bridge]] to [[gloss:l1|L1]] and Base's [[gloss:canonical bridge|canonical bridge]] from [[gloss:l1|L1]] — you'll get there in about an hour with full canonical security
Use a high-volume liquidity-pool bridge like Stargate — pool depth eliminates the trust assumption entirely
Use any [[gloss:intent-based bridge|intent-based bridge]] — the optimistic-oracle settlement makes the trust model identical to a [[gloss:canonical bridge|canonical bridge]]
Use Circle's [[gloss:cctp|CCTP]] — it burns native USDC on Arbitrum and mints native USDC on Base through Circle's issuer-verified attestation network, without any wrapped representation or third-party liquidity in the middle. Your trust assumption is Circle itself, the same as anyone holding USDC :::

:::section{mins=3} L2 in 2026: Stages, Sequencer Decentralization, and the Strategy Pivot :::

:::lead Six years into the [[gloss:rollup|rollup]]-centric roadmap, the scoreboard reads: five [[gloss:layer 2|L2s]] at [[gloss:l2beat stage|Stage 1]], zero at [[gloss:l2beat stage|Stage 2]], one production chain with a distributed [[gloss:sequencer|sequencer]], one with permissionless [[gloss:fraud proof|fraud proofs]], and an [[gloss:l1|L1]] quietly scaling on its own. On February 3, 2026, Vitalik posted that the original [[gloss:layer 2|L2]] vision "no longer makes sense": today's [[gloss:rollup|rollups]] can't or won't meet the security bar implied by being a "branded shard," and the way forward is specialization, not more generic [[gloss:evm|EVM]] clones. The two days of industry argument that followed are the most useful snapshot we have of what [[gloss:layer 2|L2]] actually means now. :::

:::p{html} Where everyone actually stands. [[gloss:l2beat|L2BEAT]]'s Stage framework is the unflattering mirror the industry has been avoiding. [[gloss:l2beat stage|Stage 0]] means a [[gloss:security council|security council]] can override the chain unilaterally — most live [[gloss:layer 2|L2s]] are still here. [[gloss:l2beat stage|Stage 1]] means [[gloss:fraud proof|fraud proofs]] work end-to-end and challengers can be permissionless, but the council still has emergency upgrade powers. [[gloss:l2beat stage|Stage 2]] means the council is removed or limited to genuine emergencies, and the chain runs on math alone. As of May 2026: :::

:::p{html}

[[gloss:l2beat stage|Stage 1]] — Arbitrum One, Base, OP [[gloss:mainnet|Mainnet]], Starknet, Scroll. Five chains total.
[[gloss:l2beat stage|Stage 0]] — zkSync Era, Linea, Mode, Lisk, Taiko, Zora, Kroma, Boba, and most others, including Polygon [[gloss:zkevm|zkEVM]] until its July 1 shutdown.
[[gloss:l2beat stage|Stage 2]] — no major [[gloss:layer 2|L2]] has reached it. None.

:::

:::p{html} [[gloss:sequencer|Sequencer]] [[gloss:decentralization|decentralization]] is even thinner. Starknet shipped the first production distributed [[gloss:sequencer|sequencer]] with the Grinta upgrade in September 2025 — three [[gloss:node|nodes]], all run by StarkWare, but architecturally a real [[gloss:consensus|consensus]] protocol rather than a single operator. Arbitrum is the only [[gloss:layer 2|L2]] with both permissionless [[gloss:validator|validators]] (since [[gloss:bold|BOLD]], February 2025) and an open [[gloss:mev|MEV]] auction (Timeboost). Everyone else — Base, OP [[gloss:mainnet|Mainnet]], zkSync, Linea, Scroll, every [[gloss:arbitrum orbit|Orbit]] chain, every [[gloss:superchain|Superchain]] member — runs a single [[gloss:sequencer|sequencer]] operated by the team. Espresso, the shared-[[gloss:sequencer|sequencer]] protocol that the [[gloss:superchain|Superchain]] has targeted for some time, is scheduled for production rollout in 2026 but isn't live yet. Astria, the most public alternative shared-[[gloss:sequencer|sequencer]] project, shut down in 2025. The "decentralized [[gloss:sequencer|sequencer]]" line on most [[gloss:layer 2|L2]] roadmaps is real engineering work. It just hasn't reached a production chain with meaningful user activity yet. :::

:::p{html} Vitalik's February 3 post. Buterin's argument has two halves. First, the diagnosis: [[gloss:layer 2|L2]] progress toward [[gloss:l2beat stage|Stage 2]] has been "far slower and more difficult than originally expected," and at the same time [[gloss:l1|L1]] itself is now scaling, with [[gloss:pectra|Pectra]] having already doubled the [[gloss:gas|gas]] limit and further increases scheduled through 2026. Second, the prescription: "We should stop thinking about [[gloss:layer 2|L2s]] as literally being 'branded shards' of [[gloss:ethereum|Ethereum]], with the social status and responsibilities that this entails". Instead, [[gloss:layer 2|L2s]] should pick a value-add other than scaling — something [[gloss:ethereum|Ethereum]] [[gloss:l1|L1]] won't do at any [[gloss:gas|gas]] limit: :::

:::p{html}

Privacy — a VM with built-in zero-knowledge primitives for a specific class of operations.
Application-specific environments — execution optimized for one major application (a game, a marketplace, an oracle).
Ultra-low latency — sub-100ms [[gloss:block|blocks]] and advanced sequencing properties that [[gloss:l1|L1]] cannot offer in principle.
Non-financial use cases — social, identity, AI infrastructure, where the guarantees needed are different from those for handling money.

:::

:::p{html} Two days later he sharpened the point: "We don't need more copypasta [[gloss:evm|EVM]] chains". His hard minimum bar: "Be [[gloss:l2beat stage|stage 1]] at the minimum (otherwise you really are just a separate [[gloss:l1|L1]] with a bridge, and you should just call yourself that)". :::

:::p{html} The reactions, by team. Steven Goldfeder of Offchain Labs (Arbitrum) said Arbitrum is "not [[gloss:ethereum|Ethereum]] but remains tightly linked" to it, citing Offchain Labs' funding of the Prysm [[gloss:consensus|consensus]] client and Arbitrum's record blockspace demand. Declan Fox of Linea/ConsenSys agreed with the direction and described Linea as "the first battle-tested [[gloss:native rollup|Native Rollup]] technology stack", positioning Linea for the post-pivot world. Jesse Pollak of Base reframed differentiation as the long-term role of [[gloss:layer 2|L2s]]: a chain has to be more than "[[gloss:ethereum|Ethereum]] but cheaper" to justify existing. Polygon leadership treated the critique as a call for sharper positioning rather than an existential threat. The most pointed external reaction came from Max Resnick, lead economist at Anza in the [[gloss:solana|Solana]] ecosystem, who called [[gloss:ethereum|Ethereum]]'s [[gloss:layer 2|L2]] roadmap a "catastrophic failure," arguing it ceded revenue to corporate [[gloss:layer 2|L2s]] while [[gloss:solana|Solana]] moved faster and cheaper. You can disagree with his conclusions, but the question underneath them is well-posed: did the strategy of a fragmented [[gloss:layer 2|L2]] market actually justify the price it cost? By 2026 that question is being asked seriously, including inside the [[gloss:ethereum|Ethereum]] community itself. The framing pulls directly on the trilemma we covered in → 2.4 §7 — different ecosystems have made different bets on which axis to compromise on, and the post-2026 debate is partly a re-litigation of those bets. :::

:::p{html} [[gloss:native rollup|Native rollups]], the technical exit. The path Vitalik is pushing has a name now: a [[gloss:native rollup|native rollup]] [[gloss:precompile|precompile]] in [[gloss:ethereum|Ethereum]]'s protocol that verifies ZK-[[gloss:evm|EVM]] proofs directly. What it removes from a [[gloss:rollup|rollup]]: :::

:::p{html}

Separate [[gloss:security council|security council]] — no longer needed, because proof verification is done by the protocol itself.
Per-chain bridge contract — replaced by a single shared [[gloss:precompile|precompile]] on [[gloss:l1|L1]].
Separate upgrade governance — upgrades come with [[gloss:ethereum|Ethereum]]'s hard forks, like any other piece of the protocol.

:::

:::p{html} A [[gloss:precompile|precompile]] is part of [[gloss:ethereum|Ethereum]] itself, so a [[gloss:rollup|rollup]] using it inherits the [[gloss:l1|L1]]'s audit, the [[gloss:l1|L1]]'s hard-[[gloss:fork|fork]] process, and [[gloss:synchronous composability|synchronous composability]] with [[gloss:l1|L1]] state. A proof-of-concept landed in March 2026 using the Ethrex execution client. If [[gloss:native rollup|native rollups]] ship, the architectural difference between "[[gloss:l1|L1]]" and "[[gloss:layer 2|L2]] on a [[gloss:native rollup|native rollup]]" gets very small — and the entire branded-shard idea finally has a way to actually work, just not through the chains that have been calling themselves shards for the past five years. :::

:::callout{icon=💡} Imagine a great national library that, decades ago, decided it couldn't possibly hold every book itself and launched an ambitious program of partner libraries across the country. Each partner would carry a "Branch of the National Library" sign, follow the same cataloguing standards, and serve as a trusted extension where citizens could read the same texts with the same guarantees of provenance. Years pass. The partner libraries grew. They became indispensable. But none of them actually adopted the full standards — they kept their own cataloguing shortcuts, their own access controls, their own decisions about which donations to accept. Meanwhile the central library itself, quietly, kept improving: its catalogue got faster, its reading rooms got bigger, its archival systems got cheaper. And then one day the head librarian writes a public letter saying: the original plan no longer fits. The central library can handle far more than we thought. The partners haven't met the standards we hoped for and several have explicitly told us they never will. We were wrong to call them all "branches." From now on, partner libraries should each pick a real specialty — a children's collection, a cartographic archive, a rare manuscripts wing — and stop pretending they're just smaller copies of the central one. The shock isn't that the partners did badly. It's that the central library no longer needs them to do what it once asked them to do. :::

:::keypoints{label="Key Points"}

Vitalik's February 2026 post isn't the death of [[gloss:layer 2|L2s]] — it's a redefinition. The new bar for being "part of [[gloss:ethereum|Ethereum]]" is [[gloss:l2beat stage|Stage 1]] minimum and a real differentiator beyond cheap blockspace. Chains that stay at [[gloss:l2beat stage|Stage 0]] with a [[gloss:multisig|multisig]] bridge get reclassified, by his framework, as separate [[gloss:l1|L1s]]
The fact that [[gloss:l1|L1]] is scaling on its own changes the bargain. When [[gloss:ethereum|Ethereum]] [[gloss:mainnet|mainnet]] can handle far more [[gloss:transaction|transactions]] than it could in 2020, the original "[[gloss:layer 2|L2s]] exist to scale [[gloss:ethereum|Ethereum]]" justification weakens — chains that exist mainly to be [[gloss:ethereum|Ethereum]]-but-cheaper are competing against an [[gloss:ethereum|Ethereum]] that's increasingly fast enough
[[gloss:sequencer|Sequencer]] [[gloss:decentralization|decentralization]] is the slowest-moving variable in the whole [[gloss:layer 2|L2]] story. Six years in, exactly one chain has a distributed [[gloss:sequencer|sequencer]] in production (Starknet), and one has permissionless [[gloss:validator|validators]] (Arbitrum). For most users, the [[gloss:layer 2|L2]] they're using is operationally one company
[[gloss:native rollup|Native rollups]], if they ship, change the architectural meaning of "[[gloss:layer 2|L2]]." A [[gloss:native rollup|native rollup]] chain inherits [[gloss:ethereum|Ethereum]]'s upgrade process and audit because the verification logic lives inside [[gloss:ethereum|Ethereum]] itself. The chains that pivot to that model will look very different from today's optimistic and [[gloss:zk rollup|ZK rollups]] :::

:::details{badge=Advanced title="EIP-8079, the EXECUTE precompile, and what synchronous composability actually buys you"} :::p{html} The technical core of the native-[[gloss:rollup|rollup]] idea is a single new [[gloss:opcode|opcode]] (or [[gloss:precompile|precompile]]) on [[gloss:ethereum|Ethereum]] [[gloss:l1|L1]] that takes a ZK-[[gloss:evm|EVM]] proof as input and verifies it inline. Call it EXECUTE. A [[gloss:rollup|rollup]] that uses this [[gloss:precompile|precompile]] no longer needs its own [[gloss:l1|L1]] contract to verify proofs — it just emits proofs into [[gloss:ethereum|Ethereum]], and [[gloss:ethereum|Ethereum]]'s protocol itself does the verification with the same hard-[[gloss:fork|fork]] process and the same security guarantees that protect every other [[gloss:ethereum|Ethereum]] operation. The current [[gloss:eip|EIP]] draft is [[gloss:eip|EIP]]-8079, and the [[gloss:ethereum|Ethereum]] Foundation has a working proof-of-concept built on the Ethrex execution client as of March 2026. :::

:::p{html} Why this matters is subtle. Today, every [[gloss:layer 2|L2]] has its own bridge contract on [[gloss:l1|L1]], with its own audit history, its own [[gloss:security council|security council]], and its own emergency-pause logic. If the bridge contract has a bug, only that [[gloss:layer 2|L2]]'s users are affected; but the price of that isolation is that every [[gloss:layer 2|L2]] reinvents the wheel, audits it separately, and asks for its own dose of user trust. A [[gloss:native rollup|native rollup]] using EXECUTE moves verification into [[gloss:ethereum|Ethereum]] itself. If there's a bug in EXECUTE, [[gloss:ethereum|Ethereum]] hard-forks to fix it, the same way it would fix any other [[gloss:consensus|consensus]] bug — and every [[gloss:layer 2|L2]] using the [[gloss:precompile|precompile]] inherits the fix automatically. The audit surface compresses from "audit every [[gloss:layer 2|L2]] bridge contract" to "audit one [[gloss:precompile|precompile]] in [[gloss:ethereum|Ethereum]]." :::

:::p{html} The second consequence is [[gloss:synchronous composability|synchronous composability]]. Today, even on the [[gloss:superchain|Superchain]]'s planned [[gloss:native interop|native interop]], cross-chain calls are eventual: you send a message, it propagates, the destination chain acts on it eventually. [[gloss:synchronous composability|Synchronous composability]] would mean a single [[gloss:l1|L1]] [[gloss:transaction|transaction]] can call into an [[gloss:layer 2|L2]] and back into [[gloss:l1|L1]] within the same [[gloss:block|block]], with the [[gloss:layer 2|L2]] execution proven on the spot. This is what makes "[[gloss:layer 2|L2]] as a true extension of [[gloss:ethereum|Ethereum]]" actually possible at the application layer — a [[gloss:defi|DeFi]] protocol could route part of a single trade through a high-throughput [[gloss:layer 2|L2]] and have it all settle atomically. The research is real (the references Vitalik points to involve preconfirmations on based [[gloss:rollup|rollups]] plus real-time proving) but it requires both [[gloss:native rollup|native rollups]] and the proving stack getting much faster than today. :::

:::p{html} The third consequence is political. A [[gloss:native rollup|native rollup]] using EXECUTE no longer needs a [[gloss:security council|security council]] with upgrade powers, because the [[gloss:security council|security council]] is [[gloss:ethereum|Ethereum]]'s social layer. This is the property Vitalik keeps pointing to: it solves the [[gloss:l2beat stage|Stage 2]] problem by making the [[gloss:security council|security council]] unnecessary in the first place. Linea's positioning as "the first battle-tested [[gloss:native rollup|Native Rollup]] technology stack" is a bet on this future arriving — that a chain already running an [[gloss:evm|EVM]]-equivalent [[gloss:zkevm|zkEVM]] with mainline [[gloss:evm|EVM]] specifications will need fewer changes to slot into a native-[[gloss:rollup|rollup]] [[gloss:precompile|precompile]] when it ships than a chain that took its own shortcuts. :::

:::p{html} The point. [[gloss:native rollup|Native rollups]] aren't a marketing rebrand of existing [[gloss:zk rollup|ZK rollups]]. They are a deeper architectural change that moves the trust boundary inside [[gloss:ethereum|Ethereum]] itself rather than at a bridge contract. If the [[gloss:precompile|precompile]] ships and proving gets fast enough, the chains that adapt to it become harder to distinguish from [[gloss:l1|L1]] from a security perspective — which is the original branded-shard promise, finally honored, just five years later than planned and with a smaller cast than expected. ::: :::

:::quiz{id=q-2-5-s8 correct=2 explanation="Correct. Vitalik's explicit minimum is [[gloss:l2beat stage|Stage 1]]: fraud or [[gloss:validity proof|validity proofs]] working end-to-end and a permissionless challenger set. Anything below that — [[gloss:l2beat stage|Stage 0]] with a [[gloss:security council|security council]] that can override [[gloss:transaction|transactions]] — gets reclassified, in his framing, as a separate [[gloss:l1|L1]] with a bridge. He's deliberately not requiring [[gloss:l2beat stage|Stage 2]] as the minimum because almost nobody is there yet; [[gloss:l2beat stage|Stage 2]] is the long-term destination, [[gloss:l2beat stage|Stage 1]] is the floor for being part of the [[gloss:ethereum|Ethereum]] conversation at all. Option A is wrong — [[gloss:tvl|TVL]] and exchange listings aren't part of the [[gloss:decentralization|decentralization]] framework. Option B is wrong — codebase choice and [[gloss:superchain|Superchain]] membership are independent of stage. Option D would be the ideal bar, but Vitalik's actual minimum is one step lower; he's setting a threshold that's hard but achievable, not aspirational."} In Vitalik's February 2026 framing, what's the minimum threshold a chain has to clear before he considers it meaningfully part of [[gloss:ethereum|Ethereum]] rather than "a separate [[gloss:l1|L1]] with a bridge"?

Have a [[gloss:tvl|TVL]] above $1 billion and be on at least two centralized exchanges
Use the [[gloss:op stack|OP Stack]] codebase and be a [[gloss:superchain|Superchain]] member
Be at [[gloss:l2beat stage|L2BEAT Stage]] 1 at minimum (fraud or [[gloss:validity proof|validity proofs]] working end-to-end, permissionless challengers) if it handles [[gloss:ethereum|ETH]] or [[gloss:ethereum|Ethereum]]-issued assets — anything below that, in his framing, is "a separate [[gloss:l1|L1]] with a bridge"
Have reached [[gloss:l2beat stage|L2BEAT Stage]] 2, meaning the [[gloss:security council|security council]] has been removed entirely :::