Ethereum: The Programmable Blockchain

:::section{mins=3} What Is Ethereum :::

:::lead [[gloss:ethereum|Ethereum]] is a blockchain with programmable [[gloss:smart contract|smart contracts]], launched on July 30, 2015. [[gloss:bitcoin|Bitcoin]] was designed to do one thing — move money. [[gloss:ethereum|Ethereum]] was designed to run programs: a bank, an exchange, a social network, an insurance system — any business logic, written as code and executed by the network itself, without intermediaries. :::

:::p{html} [[gloss:ethereum|Ethereum]] started with 19-year-old Vitalik Buterin, who in 2013 published a [[gloss:whitepaper|whitepaper]] around a single thesis: [[gloss:bitcoin|Bitcoin]]'s scripting language was too limited — what blockchain needed was a full programming language that could express arbitrary logic. We discussed exactly what that limitation was in 2.1 §3: [[gloss:bitcoin|Bitcoin]] Script is deliberately not Turing-complete, with no loops and no recursion. That's reliable, but it can't run complex programs. In 2014 the [[gloss:ethereum|Ethereum]] team raised $18 million in 42 days through an ICO (Initial Coin Offering — a public token sale), one of the first major ICOs in crypto history. Mainnet launched in 2015 with the Frontier release, and the first [[gloss:smart contract|smart contracts]] went live on it. :::

:::p{html} Since then, [[gloss:ethereum|Ethereum]] has grown from an experiment into the world's largest blockchain ecosystem. As of April 2026 the network runs on [[gloss:proof of stake|Proof of Stake]] — the switch from [[gloss:proof of work|Proof of Work]] happened on September 15, 2022, in an event known as [[gloss:the merge|The Merge]]. We covered the [[gloss:proof of stake|PoS]] mechanism in detail in 1.3 §3. A new [[gloss:block|block]] lands roughly every 12 seconds, and L1 throughput sits at 15-30 [[gloss:transaction|transactions]] per second. With [[gloss:layer 2|L2]] solutions (Arbitrum, Base, Optimism, zkSync, and over a hundred others) total throughput pushes past 100,000 TPS. The network is secured by more than 900,000 [[gloss:validator|validators]], [[gloss:ethereum|ETH]]'s market cap is around $280 billion, and the all-time high sits at $4,946, reached in August 2025. :::

:::p{html} Why did [[gloss:ethereum|Ethereum]] become the dominant platform? Because it handed developers a universal tool. On [[gloss:ethereum|Ethereum]] you can deploy a bank like Aave, an exchange like Uniswap, an insurance system, a casino, a social network — all without intermediaries and without anyone's permission. That's what drove the ecosystem's explosive growth: in summer 2020 came [[gloss:defi|DeFi]] Summer, when decentralized finance went mainstream and [[gloss:tvl|TVL]] jumped from $1 billion to $15 billion in three months. In 2021 came the [[gloss:nft|NFT]] boom, when network fees climbed so high a single ordinary [[gloss:transaction|transaction]] could cost tens of dollars. :::

:::p Philosophically, [[gloss:ethereum|Ethereum]] is the opposite of [[gloss:bitcoin|Bitcoin]]. [[gloss:bitcoin|Bitcoin]] is a conservative protocol that changes as little as possible. [[gloss:ethereum|Ethereum]] is progressive: frequent upgrades, new features, a willingness to experiment. Vitalik himself calls [[gloss:ethereum|Ethereum]] "a platform for experiments." That carries real risks — protocol complexity keeps growing, technical debt piles up. But it's exactly this willingness to evolve that has made [[gloss:ethereum|Ethereum]] the only blockchain with real mainstream user adoption. :::

:::p{html}

:::callout{icon=💡} [[gloss:bitcoin|Bitcoin]] is like a banknote in your wallet. It's a carrier of value: it sits there, you can hand it to someone, and that's it. The note itself does nothing. [[gloss:ethereum|Ethereum]] is more like a contract at a notary's office that runs itself. Imagine a rental contract that says: "on the first of every month, $1000 is debited from the tenant's account and credited to the landlord; if late, a penalty kicks in." On [[gloss:ethereum|Ethereum]], contracts like this can be written literally as code, and they execute on their own — no agent, no bank, no lawyer required. So [[gloss:bitcoin|Bitcoin]] lets you send money. [[gloss:ethereum|Ethereum]] lets you launch a full bank, an exchange, or an online casino — running without intermediaries, without registering a company, without a banking license. :::

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

• [[gloss:ethereum|Ethereum]] launched in 2015 by Vitalik Buterin as a Turing-complete blockchain platform
• Since 2022 it runs on [[gloss:proof of stake|Proof of Stake]], with a new [[gloss:block|block]] every 12 seconds and 900,000+ [[gloss:validator|validators]]
• The largest blockchain ecosystem in the world: [[gloss:ethereum|ETH]] market cap ~$280B, ATH $4,946 in August 2025
• A philosophy of fast upgrades sets [[gloss:ethereum|Ethereum]] apart from the conservative [[gloss:bitcoin|Bitcoin]] :::

:::details{badge=Timeline title="Key Events in Ethereum's History"} :::p Understanding [[gloss:ethereum|Ethereum]]'s history helps see how the ecosystem formed and which crises forged it. :::

:::p{html}

:::details{badge=Advanced title="The Ethereum Foundation and Protocol Governance"} :::p{html} The [[gloss:ethereum|Ethereum]] Foundation (EF) is a non-profit with a treasury of roughly $1 billion. It coordinates research and development of the protocol but does not control it directly. Active development involves 100-150 core developers distributed across several client teams: Geth (the main execution client, written in Go), Nethermind (.NET), Besu (Java), Erigon (optimized Go), and Reth (new Rust client from Paradigm). :::

:::p{html} This multi-client approach isn't an accident but a deliberate strategy: if a critical bug is found in one client, the network keeps running on the others. Decisions about protocol upgrades are made through the formal EIP ([[gloss:ethereum|Ethereum]] Improvement Proposals) process and regular All Core Devs calls. In 2026 the [[gloss:ethereum|Ethereum]] Foundation also began staking its own [[gloss:ethereum|ETH]] for the first time — on April 16, 2026, it completed staking 70,000 [[gloss:ethereum|ETH]] (around $93M), shifting from periodic [[gloss:ethereum|ETH]] sales to a model funded by staking yield. ::: :::

:::further{label="Further Reading"}

• Ethereum Whitepaper (Vitalik, 2013) | ethereum.org
• ethereum.org — official portal | ethereum.org :::

:::quiz{id=q-2-2-s1 correct=1 explanation="Correct. [[gloss:bitcoin|Bitcoin]] is digital money — primarily a store of value and a means of transfer. [[gloss:ethereum|Ethereum]] is a programmable platform where you can deploy [[gloss:smart contract|smart contracts]]: banks, exchanges, insurance, social networks. [[gloss:bitcoin|Bitcoin]] Script is intentionally limited (no loops, no recursion), while [[gloss:evm|EVM]] is Turing-complete."} What is the fundamental difference between [[gloss:bitcoin|Bitcoin]] and [[gloss:ethereum|Ethereum]]?

• [[gloss:ethereum|Ethereum]] is faster than [[gloss:bitcoin|Bitcoin]]
• [[gloss:ethereum|Ethereum]] is a programmable platform for any decentralized applications, while [[gloss:bitcoin|Bitcoin]] is only digital money
• [[gloss:ethereum|Ethereum]] has no hard cap on supply, [[gloss:bitcoin|Bitcoin]] does
• [[gloss:ethereum|Ethereum]] is centralized, [[gloss:bitcoin|Bitcoin]] is decentralized :::

:::section{mins=2} EVM :::

:::lead The [[gloss:evm|EVM]] ([[gloss:ethereum|Ethereum]] Virtual Machine) is the execution environment for [[gloss:smart contract|smart contracts]]. Every [[gloss:node|node]] on the network runs the same program and is guaranteed to produce the same result. :::

:::p{html} The [[gloss:evm|EVM]] is the heart of [[gloss:ethereum|Ethereum]]. It's the virtual machine that makes [[gloss:ethereum|Ethereum]] a programmable blockchain, not just a ledger of [[gloss:transaction|transactions]] like [[gloss:bitcoin|Bitcoin]]. The flow goes like this: a developer writes code in [[gloss:solidity|Solidity]] (or Vyper), and the compiler turns it into [[gloss:bytecode|bytecode]] — machine code for the [[gloss:evm|EVM]]. That [[gloss:bytecode|bytecode]] is deployed to the blockchain through a special [[gloss:transaction|transaction]] with an empty to field. When a user sends a [[gloss:transaction|transaction]] to the contract, the [[gloss:evm|EVM]] runs the [[gloss:bytecode|bytecode]], and the outcome is a change to network state — either an account balance or data in the contract's storage. :::

:::p{html} Architecturally, the [[gloss:evm|EVM]] is a stack machine with 256-bit words. The word size was chosen to work cleanly with cryptographic hashes. The instruction set runs to about 140 opcodes. The [[gloss:evm|EVM]]'s defining property is full determinism: no randomness, no dependence on time (except [[gloss:block|block]] number). That means 2+2 always equals 4 on any [[gloss:node|node]], in any country, at any moment. The [[gloss:evm|EVM]] is Turing-complete — any computable program can be written for it. With one critical constraint: every operation costs [[gloss:gas|gas]], which makes infinite loops impossible. That's exactly what sets the [[gloss:evm|EVM]] apart from [[gloss:bitcoin|Bitcoin]] Script, which we covered in 2.1 §3: in [[gloss:bitcoin|Bitcoin]], loops are banned outright at the protocol level. :::

:::p The [[gloss:evm|EVM]] works with two kinds of data storage. Memory is temporary, cheap, and resets after every [[gloss:transaction|transaction]]. Storage is permanent, expensive, and anything written to it lives forever. Writing to storage (the SSTORE opcode) is the most expensive operation in the [[gloss:evm|EVM]]: 20,000 [[gloss:gas|gas]] for a new slot, 5,000 [[gloss:gas|gas]] to update an existing one. For comparison: a simple addition (ADD opcode) costs just 3 [[gloss:gas|gas]]. The price difference reflects real cost to the network — storing data forever genuinely is expensive, because every [[gloss:node|node]] has to keep a copy. :::

:::p The most important consequence of the [[gloss:evm|EVM]]'s existence is [[gloss:evm|EVM]] compatibility. Most [[gloss:layer 2|L2]] solutions (Arbitrum, Optimism, Base) are [[gloss:evm|EVM]]-compatible: code written for [[gloss:ethereum|Ethereum]] runs on them unchanged. This creates a massive ecosystem effect — a developer writes a contract once and can deploy it on dozens of networks. :::

:::callout{icon=💡} The [[gloss:evm|EVM]] is like a coffee vending machine that simultaneously stands at thousands of [[gloss:gas|gas]] stations across the country. Drop in a coin, press the "cappuccino" button, and you get a cappuccino — at any station, in any city, at any time. No station owner can swap your drink or pocket your coin without filling the cup. The machine's rules are set once, and all thousands of machines run them identically. :::

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

• The [[gloss:evm|EVM]] is a stack machine with 256-bit words and ~140 opcodes, delivering 100% deterministic execution
• [[gloss:solidity|Solidity]] code is compiled to [[gloss:bytecode|bytecode]], deployed on-chain, and executed identically on every [[gloss:node|node]]
• Storage (permanent) costs thousands of times more than memory (temporary) — 20,000 [[gloss:gas|gas]] vs 3 [[gloss:gas|gas]]
• [[gloss:evm|EVM]] compatibility lets the same code deploy on dozens of [[gloss:layer 2|L2]] networks without rewriting :::

:::details{badge=Practical title="How the EVM Executes Smart Contracts"} :::p{html} Data in the [[gloss:evm|EVM]] is processed through a stack on the LIFO principle (last in, first out). The main opcodes are PUSH and POP for stack management, ADD and MUL for arithmetic, SLOAD for reading from storage, SSTORE for writing to storage, and CALL for invoking another contract. Each opcode has a fixed [[gloss:gas|gas]] cost, known in advance and unchanging. :::

:::p{html} When a contract is deployed, its [[gloss:bytecode|bytecode]] is saved at the contract's address on-chain. When a [[gloss:transaction|transaction]] arrives, the [[gloss:evm|EVM]] loads the [[gloss:bytecode|bytecode]], creates a fresh execution context (a clean stack, fresh memory), and sequentially executes the instructions. If [[gloss:gas|gas]] runs out before execution completes, the [[gloss:transaction|transaction]] reverts, but the [[gloss:gas|gas]] spent is not refunded. This is protection against DoS attacks: even a failed [[gloss:transaction|transaction]] costs the attacker money. ::: :::

:::details{badge=Advanced title="EOF and the Future of the EVM"} :::p{html} The current [[gloss:evm|EVM]] architecture has significant limitations: no clear separation between code and data, which creates security risks; dynamic jumps that complicate static analysis; and no proper subroutines. EOF ([[gloss:evm|EVM]] Object Format, EIP-3540 and related) aims to solve these problems: structured code sections, static jumps, and format versioning. EOF is included in the plan for future [[gloss:ethereum|Ethereum]] upgrades. :::

:::p{html} A separate direction is Parallel [[gloss:evm|EVM]]. The current [[gloss:evm|EVM]] executes [[gloss:transaction|transactions]] sequentially. Projects like Monad, Sei, and MegaETH are developing parallel execution: the system automatically identifies non-conflicting [[gloss:transaction|transactions]] and runs them simultaneously. Potential throughput gain: 10-100x. The main challenge is accurately determining state dependencies between [[gloss:transaction|transactions]]. ::: :::

:::further{label="Further Reading"}

• Ethereum Yellow Paper (formal EVM spec) | ethereum.github.io
• evm.codes — interactive opcode reference | evm.codes :::

:::quiz{id=q-2-2-s2 correct=2 explanation="Correct. Determinism means full predictability and reproducibility: the same input data always gives the same output, on any machine in any country, at any time. This is critical for blockchain — all [[gloss:node|nodes]] must come to identical results. [[gloss:evm|EVM]] has no randomness, no time dependency (except [[gloss:block|block]] number), no system calls."} Why is the [[gloss:evm|EVM]] described as "deterministic"?

• It runs very fast on modern hardware
• Developers can choose which opcodes to execute
• The same code produces the same result on every [[gloss:node|node]], regardless of the machine or location
• It automatically optimizes [[gloss:gas|gas]] costs :::

:::section{mins=2} Gas Economics :::

:::lead [[gloss:gas|gas]] measures the computational work a [[gloss:transaction|transaction]] does on [[gloss:ethereum|Ethereum]]. Every [[gloss:transaction|transaction]] has a fixed cost in [[gloss:gas|gas]]: a simple [[gloss:ethereum|ETH]] transfer is 21,000, a Uniswap swap around 200,000. The final fee = gas_used × (base_fee + priority_fee), paid in [[gloss:ethereum|ETH]]. :::

:::p Try [[gloss:gas|gas]] economics hands-on — pick an operation, adjust network conditions, see how much [[gloss:ethereum|ETH]] gets burned vs. tipped to the [[gloss:validator|validator]]. :::

:::p [[gloss:gas|gas]] is the mechanism that stops spam and infinite computations on [[gloss:ethereum|Ethereum]]. Without it, anyone could launch an infinite loop and freeze every [[gloss:node|node]]. Each [[gloss:evm|EVM]] operation has a fixed cost in [[gloss:gas|gas]]: a simple [[gloss:ethereum|ETH]] transfer costs exactly 21,000 [[gloss:gas|gas]], an ERC-20 token transfer around 65,000, a Uniswap swap 150,000-300,000, an [[gloss:nft|NFT]] mint 100,000-200,000, and a complex [[gloss:defi|DeFi]] operation can take 500,000 or more. :::

:::p{html} But [[gloss:gas|gas]] isn't a currency — it's a unit of computational work; you pay in [[gloss:ethereum|ETH]]. The [[gloss:transaction|transaction]] cost formula after EIP-1559 (the upgrade shipped in August 2021) looks like this: cost = gas_used × (base_fee + priority_fee). Here gas_used is how much [[gloss:gas|gas]] the [[gloss:transaction|transaction]] actually consumed, base_fee is the minimum [[gloss:gas|gas]] price the network sets algorithmically, and priority_fee (often just called the "tip") is a small extra you pay the [[gloss:validator|validator]] to get your [[gloss:transaction|transaction]] into a [[gloss:block|block]] sooner. The tip is usually just 1-2 gwei. There's also gas_limit — the maximum [[gloss:gas|gas]] you're willing to spend on a given [[gloss:transaction|transaction]]. With typical April 2026 numbers (base fee around 10-20 gwei, [[gloss:ethereum|ETH]] at about $2,300), a simple [[gloss:ethereum|ETH]] transfer costs $0.50-1.00, an ERC-20 transfer $1.50-3.00, a Uniswap swap $3.50-10, an [[gloss:nft|NFT]] mint $2-5, and a complex [[gloss:defi|DeFi]] operation can run $10-20. :::

:::p{html} EIP-1559 was a major rework of the fee model. Before it, the network ran a first-price auction: whoever bid more got in faster. This led to overpaying and unpredictable fees. After EIP-1559, base_fee is set algorithmically (it can change at most ±12.5% per [[gloss:block|block]]), and the user adds a tip on top as a small surcharge for priority. The crucial detail: base_fee is burned — it's destroyed entirely and reaches no one. The [[gloss:validator|validator]] only earns the priority_fee, so the tip is the [[gloss:validator|validator]]'s actual income from [[gloss:transaction|transactions]] on top of the base [[gloss:block|block]] reward. Why burn base_fee? It's a deflationary mechanism: the busier the network, the more [[gloss:ethereum|ETH]] gets destroyed. If burned [[gloss:ethereum|ETH]] exceeds new [[gloss:ethereum|ETH]] issuance over the same period, total [[gloss:ethereum|ETH]] supply shrinks. That's the foundation of the "ultrasound money" idea — money that can turn deflationary during periods of high network activity. :::

:::p An [[gloss:ethereum|Ethereum]] [[gloss:block|block]] has a [[gloss:gas|gas]] limit of 30 million units. A single [[gloss:block|block]] fits roughly 150-500 [[gloss:transaction|transactions]] depending on their complexity. When the network congests, base_fee climbs exponentially — it can double every 25 seconds with consistently full [[gloss:block|blocks]]. A notable case: during the Bored Ape Yacht Club mint in 2022, [[gloss:gas|gas]] price topped 5,000 gwei, and a single [[gloss:transaction|transaction]] cost more than $100. :::

:::callout{icon=💡} [[gloss:gas|gas]] is fuel for a car. A simple [[gloss:ethereum|ETH]] transfer is like a trip to the corner store — you need just a little fuel. A complex [[gloss:defi|DeFi]] operation is a cross-country drive, and it takes a lot more. [[gloss:gas|gas]] price is the fuel price at the pump, moving with demand: when everyone drives home on Friday night, it goes up. During rush hour or the launch of a hyped [[gloss:nft|NFT]] project, [[gloss:ethereum|Ethereum]] [[gloss:gas|gas]] behaves exactly the same way. :::

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

• [[gloss:transaction|Transaction]] cost = gas_used × (base_fee + priority_fee); base fee is set by the network, the tip by the user
• Base fee is burned rather than given to [[gloss:validator|validators]], creating deflationary pressure on [[gloss:ethereum|ETH]]
• A [[gloss:block|block]] fits 30 million [[gloss:gas|gas]] (around 150-500 [[gloss:transaction|transactions]]); during congestion, base fee rises exponentially
• A simple [[gloss:ethereum|ETH]] transfer costs 21,000 [[gloss:gas|gas]] (under a dollar); a complex [[gloss:defi|DeFi]] operation is 500,000+ [[gloss:gas|gas]] ($10-20) :::

:::details{badge=Practical title="gas Optimization for Developers"} :::p{html} [[gloss:solidity|Solidity]] developers use a full arsenal of techniques to minimize [[gloss:gas|gas]] costs. First, packing storage variables into 256-bit slots: two uint128 variables fit into one slot instead of two, which is a major saving. Second, using calldata instead of memory for read-only function parameters. Third, avoiding SSTORE wherever possible — writing to storage remains the single most expensive operation in the [[gloss:evm|EVM]]. Fourth, using events for non-critical data: they're about five times cheaper than storage. And fifth, batch operations instead of many separate [[gloss:transaction|transactions]]: one large [[gloss:transaction|transaction]] doing ten things is significantly cheaper than ten separate ones. :::

:::p{html} Understanding [[gloss:gas|gas]] economics is critical for any active [[gloss:defi|DeFi]] user. Picking the right time for a [[gloss:transaction|transaction]] (night hours in the US and Europe are usually cheaper), setting a correct [[gloss:gas|gas]] limit, and monitoring base fee via etherscan.io/gastracker — all of these can save tens or even hundreds of dollars over time. ::: :::

:::details{badge=Advanced title="Why Burn base_fee: Mechanics and Practical Sense"} :::p{html} At first glance, "burning part of the fee" looks strange — why destroy money? What's the point? This design has three practical purposes, and all three are intertwined.

1. Protection against miner manipulation

2. Linking [[gloss:ethereum|ETH]]'s value to network usage

3. The "ultrasound money" concept

Reality check 2022-2026

:::details{badge=Advanced title="EIP-4844 and Blob-transactions: A New Era of Cheap L2"} :::p{html} To understand why [[gloss:eip-4844|EIP-4844]] is so important, you need to understand how [[gloss:layer 2|L2]] networks work. They process thousands of [[gloss:transaction|transactions]] on their own side, off the main [[gloss:ethereum|Ethereum]] chain, and that's what makes them fast and cheap. But for those [[gloss:transaction|transactions]] to be protected by the same security as L1, L2s periodically have to publish data about their operations directly to [[gloss:ethereum|Ethereum]]. That receipt used to be the main cost line for [[gloss:layer 2|L2]] [[gloss:transaction|transactions]]: before 2024, L2s published their data as ordinary calldata, paying full price for every byte. :::

:::p{html} [[gloss:eip-4844|EIP-4844]] (the Dencun upgrade, March 2024) introduced a fundamentally new way to publish that data — [[gloss:blob|blobs]], or "binary large objects." It's a special storage type designed specifically for L2s: [[gloss:blob|blobs]] are visible to [[gloss:node|nodes]] but aren't executed by the [[gloss:evm|EVM]], they're stored for only about 18 days, and they have their own, much cheaper price than regular [[gloss:gas|gas]]. The result: the cost of publishing [[gloss:layer 2|L2]] data on [[gloss:ethereum|Ethereum]] dropped 90-99%. A typical Arbitrum [[gloss:transaction|transaction]] went from $0.50 down to $0.01.

:::further{label="Further Reading"}

• EIP-1559 Specification | eips.ethereum.org
• ultrasound.money — live ETH supply | ultrasound.money
• etherscan.io/gastracker | etherscan.io :::

:::quiz{id=q-2-2-s3 correct=1 explanation="Correct. base_fee is burned — destroyed forever. The [[gloss:validator|validator]] gets only the priority_fee (tip), 5-10% of the total fee. This kills the incentive for [[gloss:validator|validators]] to artificially congest the network and creates deflationary pressure on [[gloss:ethereum|ETH]] supply. The basis of "ultrasound money.""} What happens to base_fee after EIP-1559?

• It goes to the [[gloss:validator|validator]] as a reward
• It is burned — permanently removed from circulation
• It goes to the [[gloss:ethereum|Ethereum]] Foundation
• It is returned to the sender :::

:::section{mins=2} The Merge :::

:::lead On September 15, 2022, [[gloss:ethereum|Ethereum]] pulled off something no major blockchain had ever done — switching consensus mechanisms while live. It moved from [[gloss:proof of work|Proof of Work]] to [[gloss:proof of stake|Proof of Stake]] without halting the network for a single second. Energy use dropped by 99.95%, and miners were replaced by [[gloss:validator|validators]]. :::

:::p [[gloss:the merge|The Merge]] wasn't just a change of consensus algorithm — it reshaped [[gloss:ethereum|Ethereum]]'s economic model. Before [[gloss:the merge|The Merge]], [[gloss:block|blocks]] were produced by miners running GPUs and ASICs, burning roughly 78 TWh of electricity a year (comparable to Chile's national consumption). After [[gloss:the merge|The Merge]], [[gloss:block|blocks]] are produced by [[gloss:validator|validators]] who've put down 32 [[gloss:ethereum|ETH]] (around $96,000 at the time) as a deposit. Energy use dropped to about 0.01 TWh per year — a 99.95% reduction. :::

:::p In [[gloss:proof of work|PoW]], security came from the cost of hardware: to attack [[gloss:bitcoin|Bitcoin]], you'd need to buy more mining equipment than all the honest miners put together. In [[gloss:proof of stake|PoS]], security comes from the value of staked [[gloss:ethereum|ETH]]: an attacker risks losing their deposit through the [[gloss:slashing|slashing]] mechanism (a penalty for protocol violations). New [[gloss:ethereum|ETH]] issuance dropped from around 4.3% per year ([[gloss:proof of work|PoW]]) to 0.5-1% ([[gloss:proof of stake|PoS]]). The barrier to entry shifted too: instead of thousands of dollars in GPU spend, you now need 32 [[gloss:ethereum|ETH]]. :::

:::p Technically [[gloss:the merge|The Merge]] worked like this. The Beacon Chain (the [[gloss:proof of stake|PoS]] chain) launched separately in December 2020. For two years it ran in parallel with the main [[gloss:proof of work|PoW]] chain, giving the new consensus mechanism plenty of time to be tested. On September 15, 2022 at 06:42 UTC, the Terminal Total Difficulty was hit — a predetermined threshold after which the [[gloss:proof of work|PoW]] chain "merged" into the Beacon Chain. The switch happened with zero downtime: all infrastructure — dApps, [[gloss:defi|DeFi]] protocols, wallets — kept running without interruption. :::

:::p What did NOT change after [[gloss:the merge|The Merge]]: TPS stayed the same (around 15-30), and [[gloss:gas|gas]] fees didn't drop (a common misconception). What did change: [[gloss:block|blocks]] started arriving steadily every 12 seconds (instead of the random [[gloss:proof of work|PoW]] intervals), issuance dropped by roughly 90%, and [[gloss:ethereum|ETH]] holders gained the ability to earn staking income. :::

:::callout{icon=💡} Picture a notary stamping documents. In [[gloss:proof of work|Proof of Work]], that notary runs a huge office: rooms of staff, walk-in vaults, expensive equipment, a security desk. Forging a stamp would mean corrupting all of that — so expensive nobody bothers trying. In [[gloss:proof of stake|Proof of Stake]] the design flips: the notary works alone with no equipment, but before they can start they have to put down a large deposit at a bank. Each stamp is their personal signature, backed by that deposit. If they ever certify a fraudulent document, the bank automatically takes the deposit and their reputation is finished. Honest stamping becomes the only rational strategy — cheating costs more than the most generous bribe. :::

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

• [[gloss:the merge|The Merge]] (September 15, 2022) — the transition from [[gloss:proof of work|PoW]] to [[gloss:proof of stake|PoS]] with zero downtime
• Energy consumption dropped by 99.95% (from 78 TWh to 0.01 TWh per year)
• Issuance went from around 4.3% down to 0.5-1% annually
• TPS and [[gloss:gas|gas]] fees did not change — [[gloss:the merge|The Merge]] affected only consensus :::

:::details{badge=Practical title="Finality and PoS Security"} :::p{html} When we talk about [[gloss:transaction|transaction]] "finality," we mean the point after which a [[gloss:transaction|transaction]] can no longer be reversed. [[gloss:bitcoin|Bitcoin]] and [[gloss:ethereum|Ethereum]] reach that point along very different paths, and understanding the difference matters — especially for large amounts.

Probabilistic finality ([[gloss:bitcoin|Bitcoin]])

Deterministic finality ([[gloss:ethereum|Ethereum]] [[gloss:proof of stake|PoS]])

Why "reversing" a finalized [[gloss:block|block]] costs 1/3 of the stake

The numbers

The key intuition

What this means for you

:::details{badge=Practical title="Client Diversity and Network Resilience"} :::p{html} To understand what a "client" is, picture the following. [[gloss:ethereum|Ethereum]] itself is a set of rules described in a specification. A client is a specific program that implements those rules, runs on a server, and does all the work: accepts [[gloss:transaction|transactions]], validates [[gloss:block|blocks]], talks to other [[gloss:node|nodes]], votes on [[gloss:block|blocks]] if it's running as a [[gloss:validator|validator]]. There's a useful parallel with web browsers: HTML and HTTP are specifications, while Chrome, Firefox, and Safari are different programs implementing the same rules. Each can work with any website because they all follow the same standards, but inside they are completely different codebases written by different teams.

Why multiple clients?

Two types of clients

Current distribution (April 2026)

Why 1/3 is the critical threshold

The conclusion

:::further{label="Further Reading"}

• The Merge Documentation | ethereum.org
• Vitalik: "Why Proof of Stake" | vitalik.eth.limo :::

:::quiz{id=q-2-2-s4 correct=2 explanation="Correct. [[gloss:the merge|The Merge]] changed the very foundation of [[gloss:ethereum|Ethereum]] security: from physical (the cost of mining hardware) to economic (the value of staked [[gloss:ethereum|ETH]]). It also eliminated 99.95% of energy consumption. But TPS and [[gloss:gas|gas]] fees did NOT change — those depend on [[gloss:block|block]] size and execution architecture, not on the consensus algorithm."} What is the most important consequence of [[gloss:the merge|The Merge]]?

• [[gloss:transaction|Transactions]] became significantly faster
• [[gloss:gas|gas]] fees dropped sharply
• Energy consumption dropped by 99.95% and network security is now backed by [[gloss:validator|validator]] deposits
• The number of [[gloss:transaction|transactions]] per second increased 10x :::

:::section{mins=2} Staking :::

:::lead Staking is how [[gloss:ethereum|ETH]] holders secure the network and earn ~3-4% per year. Solo: lock 32 [[gloss:ethereum|ETH]] and run your own [[gloss:validator|validator]]. Liquid (Lido, Rocket Pool): deposit any amount, get a tradable token, keep using it across [[gloss:defi|DeFi]]. :::

:::p After [[gloss:the merge|The Merge]], staking became the main way to help secure [[gloss:ethereum|Ethereum]]. There are two very different approaches. Solo staking requires 32 [[gloss:ethereum|ETH]] (around $96,000), full control over your own [[gloss:validator|validator]], and the work of setting up and maintaining a [[gloss:node|node]] — the complexity is high, but so is the yield (3-4% APY, Annual Percentage Yield — the annualized return on investment), and you directly support network decentralization. Liquid staking lets you stake any amount through a protocol like Lido or Rocket Pool: you deposit [[gloss:ethereum|ETH]] and get back a liquid token (stETH, rETH) that you can use across [[gloss:defi|DeFi]] (as collateral on Aave, or in a liquidity pool on Curve). Yield is slightly lower (3-3.5% after the protocol fee), but the entry is dead simple. :::

:::p Liquid staking dominates by volume: Lido (stETH) controls about 30% of all staked [[gloss:ethereum|ETH]], roughly $23 billion as of April 2026. Rocket Pool (rETH) holds about $2 billion, and Coinbase (cbETH) about $3 billion. stETH has effectively become one of the most important assets in [[gloss:defi|DeFi]]: it's used as collateral on lending protocols, traded on exchanges, and put to work in yield farming. :::

:::p The economics of a solo [[gloss:validator|validator]] look like this. A 32 [[gloss:ethereum|ETH]] stake yields about 1-1.3 [[gloss:ethereum|ETH]] per year. That income comes from three sources: consensus rewards (newly issued [[gloss:ethereum|ETH]] for producing [[gloss:block|blocks]], roughly 60% of income), execution rewards (priority fees from [[gloss:transaction|transactions]], around 30%), and [[gloss:mev|MEV]] tips through [[gloss:mev|MEV]]-Boost (around 10%). Hardware costs run $500-1,000 one-time, plus $50-100 per month for a VPS or a home server. Final yield: 3-4% APY, plus any potential upside from [[gloss:ethereum|ETH]]'s price. :::

:::callout{icon=💡} Picture [[gloss:ethereum|ETH]] as money, and you're deciding what to do with it. Solo staking is like opening your own bakery. You put up serious capital, you're at the oven at 5 AM, you handle suppliers and tax filings yourself — but every cent of profit is yours. It's real work and it takes know-how, and the yield is the highest. Liquid staking is like putting your money in a bank deposit. You don't have to do anything: the bank deploys it for you, hands you a receipt (stETH), and your balance grows on its own each month. The yield is a little lower — the bank takes a cut for managing it — but you don't have to bake bread or even know how it's done. And the bank receipt has a key advantage: you can use it as collateral, say for a mortgage. In the same way, stETH can be used as collateral across [[gloss:defi|DeFi]]. :::

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

• Solo staking: 32 [[gloss:ethereum|ETH]], 3-4% APY, full control, high complexity
• Liquid staking: any amount, stETH/rETH are liquid and used across [[gloss:defi|DeFi]]
• Lido controls about 30% of all staked [[gloss:ethereum|ETH]] (~$23B) — the largest liquid staking protocol
• [[gloss:validator|Validator]] income = consensus rewards + priority fees + [[gloss:mev|MEV]] tips :::

:::details{badge=Practical title="Slashing: Punishment for Malicious Behavior"} :::p{html} [[gloss:slashing|slashing]] is the mechanism that punishes [[gloss:validator|validators]] for actions threatening network security. There are two main violations: double voting (signing two conflicting [[gloss:block|blocks]]) and surround voting (an attempt to rewrite history). When a violation is detected, the [[gloss:validator|validator]] loses at least 1/32 of their stake (minimum 1 [[gloss:ethereum|ETH]]). :::

:::p{html} There's also a correlation penalty: if many [[gloss:validator|validators]] are penalized at the same time (which indicates a coordinated attack), the penalty grows up to 100% of the stake. This is a brilliant mechanism — a single [[gloss:validator|validator]]'s accidental error costs little, but a coordinated attack becomes devastatingly expensive for the attackers. Historically, about 400 [[gloss:slashing|slashing]] incidents have occurred, and the overwhelming majority came from misconfiguration rather than actual attacks. ::: :::

:::details{badge=Practical title="Pectra and MaxEB: The Era of Big Validators"} :::p{html} In May 2025, the Pectra upgrade introduced one of the most practically significant changes to staking economics since [[gloss:the merge|The Merge]] itself. The change is called EIP-7251: MaxEB (Maximum Effective Balance), and the gist of it is that the upper limit of stake per [[gloss:validator|validator]] was raised from 32 [[gloss:ethereum|ETH]] to 2048 [[gloss:ethereum|ETH]].

What changed

Why was this done

Does this hurt security?

The rest of Pectra

:::details{badge=Advanced title="Restaking and DVT: The Next Evolution of Staking"} :::p{html} Restaking (EigenLayer, rebranded to EigenCloud in 2025) is the concept of using already-staked [[gloss:ethereum|ETH]] to secure other protocols (AVS — Actively Validated Services). The same [[gloss:ethereum|ETH]] simultaneously protects [[gloss:ethereum|Ethereum]] itself, an oracle network, a bridge, and a DA layer (data availability layer). This improves capital efficiency, but it introduces a new risk — compound [[gloss:slashing|slashing]], where penalties arrive from several protocols at once. The additional yield for a [[gloss:validator|validator]] is about 1-5% on top of the base rate. Beyond EigenCloud, there are a few other notable protocols in this space: Symbiotic (around $500-900M [[gloss:tvl|TVL]], positioning itself as a fully permissionless alternative that accepts any ERC-20 token as collateral) and Karak (around $100M [[gloss:tvl|TVL]], with a multi-asset and multi-chain approach). Liquid restaking is a separate layer built on top of these protocols, represented by tokens like eETH (EtherFi), rsETH (Kelp), weETH, and pufETH. :::

:::p{html} This is controversial technology, and April 2026 showed exactly why. On April 18-20, 2026, the largest [[gloss:defi|DeFi]] exploit of the year occurred: attackers stole $292 million through a vulnerability in the LayerZero bridge used by KelpDAO. By exploiting a flaw in cross-chain message verification, they minted 116,500 rsETH with no real backing, then used those tokens as collateral on Aave to borrow against them. Protocols across [[gloss:defi|DeFi]] froze rsETH-related activity in panic, and a massive liquidity outflow began: total [[gloss:defi|DeFi]] [[gloss:tvl|TVL]] collapsed from $99.5 billion to $86 billion in two days, Aave lost about $6.6 billion and was left with bad debt of roughly $196 million. Attribution points to North Korea's Lazarus Group.

:::further{label="Further Reading"}

• Ethereum Staking | ethereum.org
• Lido | lido.fi
• EigenLayer | eigenlayer.xyz :::

:::quiz{id=q-2-2-s5 correct=2 explanation="Correct. Solo staking — full control, max yield, but high entry barrier (32 [[gloss:ethereum|ETH]] ≈ $96K) and complexity (running your own [[gloss:node|node]]). Liquid staking (Lido, Rocket Pool) lets you stake any amount through a protocol while keeping liquidity via stETH/rETH tokens — those can be used in [[gloss:defi|DeFi]] as collateral or in liquidity pools."} What is the key difference between solo staking and liquid staking?

• Liquid staking yields higher APY
• Solo staking is cheaper to run
• Solo staking requires 32 [[gloss:ethereum|ETH]] and a personal [[gloss:node|node]], while liquid staking lets you stake any amount and keeps your liquidity through stETH/rETH tokens
• Liquid staking doesn't use the [[gloss:proof of stake|PoS]] mechanism :::

:::section{mins=2} Account Model :::

:::lead [[gloss:ethereum|Ethereum]] uses an account model: every address has a balance, much like a bank account. This is fundamentally different from [[gloss:bitcoin|Bitcoin]]'s [[gloss:utxo|UTXO]] model, where a balance is the sum of "unspent outputs," like the set of bills in a wallet. :::

:::p{html} If you've ever opened MetaMask and seen a balance like "1.42 [[gloss:ethereum|ETH]]," you've already met [[gloss:ethereum|Ethereum]]'s account model in person. The address you see is your account, the number next to it is its current state, and that's it — there's no need to track which "incoming" [[gloss:transaction|transactions]] formed that balance. This is fundamentally different from how [[gloss:bitcoin|Bitcoin]] works (we covered that in 2.1 §3): [[gloss:bitcoin|Bitcoin]] doesn't actually have a "balance" — it has a list of "unspent outputs" that you have to add up to know how much you own. [[gloss:ethereum|Ethereum]] took the simpler path: every address is just an account with one number. That's why developers are happier writing [[gloss:smart contract|smart contracts]] on [[gloss:ethereum|Ethereum]], and that's why a wallet on a phone shows your balance in milliseconds. :::

:::p{html} But there's an important nuance: not all accounts are alike. There are two types in [[gloss:ethereum|Ethereum]], and the difference between them defines how the whole ecosystem works. The first type is an [[gloss:eoa|EOA]] (Externally Owned Account, an externally controlled account) — this is your MetaMask. It's controlled by a private key, can send [[gloss:transaction|transactions]] on its own, but has no logic of its own. It just sits there until the human owner decides to do something. The second type is a Contract Account — a [[gloss:smart contract|smart contract]]. It has its own code, but it can't act on its own initiative: it only reacts to incoming calls. Uniswap, Aave, every [[gloss:defi|DeFi]] protocol — they're all Contract Accounts that "wake up" only when an [[gloss:eoa|EOA]] pings them. That means every action on the network always begins with a person — someone, somewhere, signed a [[gloss:transaction|transaction]] with their key. :::

:::p{html} A practical detail you'll bump into: nonce. Every [[gloss:transaction|transaction]] from an account carries a sequential number — 0, 1, 2, 3 and so on — and the network executes them strictly in order. The [[gloss:transaction|transaction]] with nonce 5 won't go through until nonce 4 has been processed. This is what guarantees that you can't "double-spend" the same [[gloss:ethereum|ETH]] twice. But it also creates a familiar pain: if you set the [[gloss:gas|gas]] fee too low and a [[gloss:transaction|transaction]] "stalls," every subsequent [[gloss:transaction|transaction]] from that wallet gets stuck behind it. The fix is to "overwrite" the stalled [[gloss:transaction|transaction]] — send 0 [[gloss:ethereum|ETH]] to yourself with the same nonce and a higher tip. The second [[gloss:transaction|transaction]] muscles its way past the first one and frees up the queue. :::

:::p{html} One more thing the account model trades off: privacy. With [[gloss:bitcoin|Bitcoin]]'s [[gloss:utxo|UTXO]], you can use a fresh address for every [[gloss:transaction|transaction]], blurring the trail. With an account model, all your activity stacks up under a single identity, fully visible to any [[gloss:block|block]] explorer. Different design, different priorities — and the two leading platforms went different ways on it. :::

:::callout{icon=💡} An [[gloss:eoa|EOA]] is like a person with a banking app on their phone. For anything to happen, you have to open the app yourself, tap the button, sign the payment. Without your action, the account just sits there. A Contract is like a vending machine: it gets installed once, the rules are written into it (drop in a coin — get a can), and from then on it just stands there on its own. It can't go to the store or start selling on its own — it only responds to whoever walks up. Every [[gloss:defi|DeFi]] protocol — Uniswap, Aave, any exchange — works exactly like one of these vending machines: huge, complex, with billions of dollars inside, but unable to take the first step on its own. Activity on [[gloss:ethereum|Ethereum]] always begins with an [[gloss:eoa|EOA]] action — that is, with a person. :::

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

• Two account types: [[gloss:eoa|EOA]] (a wallet with a private key) and Contract Account (controlled by code)
• An account stores nonce, balance, storageRoot, and codeHash
• Nonce guarantees the order of [[gloss:transaction|transactions]], but a stuck [[gloss:transaction|transaction]] [[gloss:block|blocks]] all the following ones
• The account model is simpler for developers; [[gloss:utxo|UTXO]] is better for privacy and parallelization :::

:::details{badge=Practical title="Account Abstraction: Wallets of the Future"} :::p{html} [[gloss:account abstraction|Account Abstraction]] ([[gloss:account abstraction|ERC-4337]]) is a standard that blurs the line between an [[gloss:eoa|EOA]] and a Contract Account, effectively turning every wallet into a [[gloss:smart contract|smart contract]]. This opens possibilities that aren't available with ordinary [[gloss:eoa|EOAs]]: social recovery (you lose your key, and friends or trusted parties help you restore access), [[gloss:gas|gas]] sponsorship (the dApp pays the [[gloss:gas|gas]] fee for you — a UX closer to Web2), batch [[gloss:transaction|transactions]] (several actions in a single [[gloss:transaction|transaction]], like approve + swap in one click), and session keys (limited permissions for a dApp without having to sign each action). :::

:::p{html} Leading implementations include Safe (formerly Gnosis Safe), a standard for multisig wallets, along with Kernel and Biconomy. Looking ahead, the expectation is that all wallets become [[gloss:smart contract|smart contract]] wallets, which will substantially improve both UX and the security of blockchain interactions. ::: :::

:::details{badge=Advanced title="State Management and Data Scaling"} :::p{html} [[gloss:ethereum|Ethereum]] state is the totality of all accounts, balances, and storage data across all contracts. Technically it's stored in a Modified Merkle Patricia Trie. State size is about 100-150 GB as of April 2026, and it grows by roughly 50 GB per year. This creates a mounting load on full [[gloss:node|nodes]] — disk and RAM requirements keep climbing, and if nothing changes, in a few years running a full [[gloss:ethereum|Ethereum]] [[gloss:node|node]] will only be possible on very powerful servers. That's a serious threat to decentralization. :::

:::p{html} Inside that trie, every account has just four fields: nonce (the [[gloss:transaction|transaction]] counter), balance (the balance in wei), storageRoot (the root of the contract's data tree), and codeHash (a hash of the contract's code). For an [[gloss:eoa|EOA]], the last two are empty — it really is just a number with a counter. For a Contract, all four matter, because that's where contract state actually lives. Every [[gloss:ethereum|Ethereum]] [[gloss:block|block]] is, in effect, an updated snapshot of these four fields across hundreds of millions of accounts. That's why state grows the way it does: every new contract deployed and every storage write adds another entry to that trie.

:::further{label="Further Reading"}

• ERC-4337 Specification | eips.ethereum.org
• Accounts documentation | ethereum.org :::

:::quiz{id=q-2-2-s6 correct=2 explanation="Correct. An [[gloss:eoa|EOA]] (Externally Owned Account) is a regular wallet controlled by a private key — like your MetaMask. A Contract Account is a [[gloss:smart contract|smart contract]] controlled by code (Uniswap, Aave). The key difference: an [[gloss:eoa|EOA]] can initiate [[gloss:transaction|transactions]], a Contract Account cannot — it only reacts to incoming calls. [[gloss:account abstraction|Account Abstraction]] ([[gloss:account abstraction|ERC-4337]]) blurs this line."} What is the key difference between an [[gloss:eoa|EOA]] and a Contract Account?

• [[gloss:eoa|EOA]] is faster than a Contract
• Contracts cost more to deploy
• An [[gloss:eoa|EOA]] is controlled by a private key and can initiate [[gloss:transaction|transactions]]; a Contract is controlled by code and only reacts to calls
• [[gloss:eoa|EOAs]] cannot hold [[gloss:ethereum|ETH]] :::

:::section{mins=2} Roadmap :::

:::lead Vitalik Buterin split [[gloss:ethereum|Ethereum]]'s future development into five stages with memorable names: Surge (scaling), Scourge (the fight against [[gloss:mev|MEV]]), Verge (verification), Purge (simplification), and Splurge (final improvements). :::

:::p{html}

The [[gloss:ethereum|Ethereum]] roadmap isn't a rigid plan but a set of five parallel directions of work. Each addresses its own problem, and all of them are actively in progress — there's already shipped work on every track, and more upgrades coming on each.

Surge — scaling

Goal: 100,000+ [[gloss:transaction|transactions]] per second through [[gloss:layer 2|L2]].

• Done: [[gloss:eip-4844|EIP-4844]] with [[gloss:blob|blobs]] — Dencun upgrade, March 2024 (the first major step)
• Pectra — May 2025 (staking improvements, [[gloss:validator|validator]] consolidation)
• Fusaka — December 2025 (raised [[gloss:blob|blob]] limits, PeerDAS)
• Coming: Glamsterdam — first half of 2026 (will introduce ePBS, parallel execution, [[gloss:gas|gas]] limit toward 100M)
• Final goal: full Danksharding with 64-128 [[gloss:blob|blobs]] per [[gloss:block|block]] and data availability sampling

Scourge — fighting [[gloss:mev|MEV]] and censorship

Goal: prevent [[gloss:validator|validators]] and [[gloss:block|block]] builders from manipulating [[gloss:transaction|transaction]] order for their own profit. [[gloss:mev|MEV]] ([[gloss:mev|Maximal Extractable Value]]) is the profit extracted by reordering [[gloss:transaction|transactions]] within a [[gloss:block|block]].

• Done: [[gloss:mev|MEV]]-Boost — running since 2022 as an off-protocol solution that 90%+ of [[gloss:validator|validators]] use voluntarily
• Coming: ePBS (enshrined Proposer-Builder Separation) in Glamsterdam — moves the same role split into the protocol itself
• In discussion: FOCIL ([[gloss:fork|Fork]]-choice Enforced Inclusion Lists) — likely in Hegota, would make it impossible for proposers to censor specific [[gloss:transaction|transactions]]

Verge — verification

Goal: a full [[gloss:ethereum|Ethereum]] [[gloss:node|node]] should run even on a smartphone. Verkle trees and SNARK proofs will dramatically shrink proof sizes and hardware requirements.

• In progress: several years of active R&D on Verkle trees, working devnets
• Coming: Hegota upgrade — second half of 2026, expected to introduce Verkle trees on mainnet
• Final goal: stateless clients — [[gloss:node|nodes]] that don't store state at all and verify each [[gloss:block|block]] via compact proofs

Purge — removing technical debt

Goal: lighten the load on [[gloss:node|nodes]] by getting rid of old data and obsolete protocol features.

• Done: history expiry (EIP-4444) — already supported in clients, Geth's prune-history command frees hundreds of GB
• In progress: active discussion of state expiry — proposals from the [[gloss:ethereum|Ethereum]] Foundation in late 2025
• Coming: banning legacy opcodes (SELFDESTRUCT and others), simplifying [[gloss:gas|gas]] mechanics

Splurge — final improvements

Goal: a basket of smaller upgrades that don't fit neatly into the other tracks.

• Done: EIP-7702 in Pectra — simplified [[gloss:account abstraction|Account Abstraction]] that lets a regular [[gloss:eoa|EOA]] temporarily "become" a [[gloss:smart contract|smart contract]]
• In progress: full native [[gloss:account abstraction|Account Abstraction]] ([[gloss:smart contract|smart contract]] wallets as the default)
• Long-term: VDFs (Verifiable Delay Functions) for trustworthy on-chain randomness, quantum-resistant cryptography in the long run

The point of running all five tracks in parallel is precisely that progress in one (say, scaling via Surge) doesn't block another (say, verification via Verge). This is what allows [[gloss:ethereum|Ethereum]] to evolve much faster than if upgrades came one after another.

:::callout{icon=💡} Picture [[gloss:ethereum|Ethereum]] as a busy city that's being constantly rebuilt — and you can't shut it down for even a day, because hospitals, schools, and banks are running. Each of the five tracks tackles a different problem of city life. Surge widens the streets — more lanes for the same cars ([[gloss:layer 2|L2]] scaling). Scourge cracks down on corrupt traffic cops who let bribers through ([[gloss:mev|MEV]] from [[gloss:validator|validators]]). Verge hands every commuter a smartphone navigator instead of huge paper maps (lightweight verification). Purge tears down abandoned warehouses no one uses anymore (old data and obsolete features). Splurge covers the small finishing touches — better signs, smoother sidewalks, modern traffic lights ([[gloss:account abstraction|account abstraction]], VDFs, and more). All five crews work simultaneously, and the city never stops. :::

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

• Five parallel stages: Surge, Scourge, Verge, Purge, Splurge
• Surge: [[gloss:eip-4844|EIP-4844]] (2024) → Pectra (2025) → Fusaka (2025) → Glamsterdam (2026) → full Danksharding
• Verge: Verkle trees + SNARKs will let a full [[gloss:node|node]] run on a phone or a Raspberry Pi
• All stages develop in parallel, not sequentially — this accelerates the network's evolution :::

:::details{badge=Advanced title="Verkle Trees and Enshrined PBS"} :::p{html} Verkle trees are a new data structure meant to replace [[gloss:ethereum|Ethereum]]'s current state storage (Modified Merkle Patricia Trie). We touched on them in §6 as one of the solutions to the problem of growing state size. Their main advantage is how compact the "proofs" they produce are. When some [[gloss:node|node]] checks that a specific address has a specific balance, it doesn't need the whole state — just a small proof, a cryptographic demonstration that the record in question really exists in the network's state. Today, such a proof weighs 1-2 kilobytes; in Verkle trees it shrinks to about 150 bytes — 10 times smaller. This difference makes the next architectural shift possible — stateless clients, [[gloss:node|nodes]] that don't store the full state at all. Even an ordinary smartphone or a Raspberry Pi will be able to run as a full [[gloss:ethereum|Ethereum]] [[gloss:validator|validator]], receiving compact proofs alongside each [[gloss:block|block]]. :::

:::p{html} Enshrined PBS (Proposer-Builder Separation) is the idea of baking a role separation directly into the [[gloss:ethereum|Ethereum]] protocol. Today, two roles are already performed separately off-chain. The Proposer is the [[gloss:validator|validator]] whose turn it is to propose the next [[gloss:block|block]]. The Builder is a specialized company or service that assembles the most profitable set of [[gloss:transaction|transactions]] from the [[gloss:mempool|mempool]]. Right now this scheme runs via [[gloss:mev|MEV]]-Boost — an add-on program that [[gloss:validator|validators]] run voluntarily, outside the protocol. The [[gloss:validator|validator]] picks the [[gloss:block|block]] from whichever builder offers the highest bid, and this helps [[gloss:validator|validators]] earn more from [[gloss:mev|MEV]] (Maximum Extractable Value — the profit from reordering [[gloss:transaction|transactions]] within a [[gloss:block|block]]). The problem is that [[gloss:mev|MEV]]-Boost is an external solution, living outside the protocol. In the enshrined version, planned for the Glamsterdam upgrade (May 2026), the same role separation will be built into the protocol itself. This provides two important things: censorship resistance — a proposer can't unilaterally refuse to include a [[gloss:transaction|transaction]], and a fairer builder market, where everyone competes on equal terms. ::: :::

:::further{label="Further Reading"}

• Ethereum Roadmap Documentation | ethereum.org
• Vitalik's blog — roadmap essays | vitalik.eth.limo :::

:::quiz{id=q-2-2-s7 correct=1 explanation="Correct. Surge is the scaling stage. The path: [[gloss:eip-4844|EIP-4844]] ([[gloss:blob|blobs]], March 2024) → Pectra (May 2025) → Fusaka (December 2025) → Glamsterdam (May 2026) → full Danksharding (64-128 [[gloss:blob|blobs]] per [[gloss:block|block]], 2-4 years). Goal: 100,000+ TPS through [[gloss:layer 2|L2]] inheriting L1 security. Other stages tackle [[gloss:mev|MEV]] (Scourge), verification (Verge), simplification (Purge), final improvements (Splurge)."} What is the main goal of the Surge stage?

• Fighting [[gloss:mev|MEV]] and censorship
• Scaling [[gloss:ethereum|Ethereum]] to 100,000+ TPS through L2s
• Reducing state size
• Full [[gloss:account abstraction|account abstraction]] :::

:::section{mins=2} Ecosystem :::

:::lead [[gloss:ethereum|Ethereum]] is the world's largest blockchain ecosystem, combining [[gloss:defi|DeFi]] (with a total [[gloss:tvl|TVL]] around $86-100 billion), NFTs, over a hundred [[gloss:layer 2|L2]] solutions (Arbitrum, Base, Optimism, zkSync, and others), stablecoins (USDC, USDT, DAI), and thousands of decentralized applications. :::

:::p{html} The scale of [[gloss:ethereum|Ethereum]]'s ecosystem is best shown by sector-by-sector numbers for April 2026. [[gloss:ethereum|Ethereum]] itself holds about $57 billion in [[gloss:tvl|TVL]] — roughly half of all [[gloss:defi|DeFi]]. The largest individual protocols are Aave (lending) with about $20 billion after the recent outflow caused by the KelpDAO exploit (it had been $26 billion before), and Lido (liquid staking) with about $23 billion. Stablecoins on [[gloss:ethereum|Ethereum]] (USDC, USDT, DAI, and others) total more than $130 billion combined. [[gloss:layer 2|L2]] networks hold around $30 billion in [[gloss:tvl|TVL]]. We'll cover the [[gloss:layer 2|L2]] ecosystem in detail in 2.5 [[gloss:layer 2|L2]]. :::

:::p For a long time, [[gloss:ethereum|Ethereum]]'s strategy was described as the [[gloss:layer 2|L2]]-centric, or [[gloss:rollup|rollup]]-centric roadmap. The idea was that L1 would remain the expensive, slow security layer, while all activity moved onto [[gloss:layer 2|L2]] networks — "branded shards" of [[gloss:ethereum|Ethereum]] that would inherit its security through cryptographic proofs. Arbitrum, Optimism, Base, zkSync — all of them emerged as part of this strategy. But on February 3, 2026, Vitalik published a post on X that marked a major turning point: he directly acknowledged that the [[gloss:rollup|rollup]]-centric roadmap as originally conceived "no longer makes sense." There are two reasons. :::

:::p The first is that L2s have decentralized far more slowly than expected. L2Beat introduced a stage system for [[gloss:rollup|rollups]]: Stage 0 (fully centralized control), Stage 1 (partial decentralization with safeguards), Stage 2 (full decentralization). Even by 2025, only three major L2s had reached Stage 1: Arbitrum, Optimism, and Base. The rest were stuck at Stage 0 — which means their operators can technically freeze, upgrade, or change any state of the network. Some projects have stated outright that they never intend to go further — their fintech clients demand retained "ultimate control" for regulatory reasons. Vitalik responded bluntly: "if you build a 10,000 TPS [[gloss:evm|EVM]] that connects to L1 via a multisig bridge, you're not scaling [[gloss:ethereum|Ethereum]] — you're running an L1 with a bridge." The second reason is that L1 itself started scaling much faster than the original strategy assumed. The [[gloss:gas|gas]] limit rose from 30 million at the start of 2025 to 36 million by mid-year, Pectra and Fusaka expanded capacity, and L1 fees dropped to levels where most users can already operate directly on the main chain. :::

:::p{html} Vitalik's new vision is this: L2s should exist not as generic scaling clones of [[gloss:ethereum|Ethereum]], but as specialized chains that solve problems L1 can't. The examples he gives: application-specific chains for concrete use cases (gaming, trading, social networks), chains with extremely low latency for high-frequency trading, privacy-oriented solutions, chains for non-financial data and AI coordination. At the Hong Kong Web3 Festival on April 20, 2026, Vitalik said it plainly: "an [[gloss:layer 2|L2]] that copies [[gloss:ethereum|Ethereum]] and becomes more centralized is meaningless." For the ecosystem, this means a phase shift: from "many generic L2s competing for [[gloss:tvl|TVL]]" to "specialized chains with a clear use case and real decentralization." As of April 2026, the picture looks like this: Base (Coinbase's [[gloss:layer 2|L2]]) is #1 by [[gloss:tvl|TVL]] ($4.63B at the end of 2025), the most active by users; Arbitrum is historically the largest, with a developed [[gloss:defi|DeFi]] ecosystem; Optimism pursues the Superchain concept through OP Stack (an open framework for building L2s, on which Base, Zora, and others are built); ZK-[[gloss:rollup|rollups]] (zkSync, Scroll, Polygon zkEVM, Starknet) offer stronger mathematical guarantees but have been slower to gain adoption. A detailed comparison of all L2s is available at L2Beat.com. :::

:::p ENS ([[gloss:ethereum|Ethereum]] Name Service) is a domain-name system for the blockchain. Instead of a 42-character address like 0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045, you can use vitalik.eth. More than 2 million names have been registered. Some names trade on the secondary market for tens or even hundreds of thousands of dollars. :::

:::callout{icon=💡} [[gloss:ethereum|Ethereum]] initially tried to work on a "metro in a big city" model: the network itself was the central transit hub, and L2s were the buses and commuter trains that carried passengers out to the districts. But it turned out that the bus companies operated opaquely and could cancel routes at any time, while the metro itself had expanded in the meantime and was handling more passengers directly. So the strategy changed: now the metro handles the main load, and other forms of transport have their place only where something specific is needed — express trains for distant suburbs, airport express, cargo transport. :::

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

• [[gloss:ethereum|Ethereum]]'s ecosystem includes [[gloss:defi|DeFi]], NFTs, [[gloss:layer 2|L2]], and stablecoins with total [[gloss:tvl|TVL]] around $86-100 billion
• In February 2026 Vitalik announced a strategic shift: the [[gloss:rollup|rollup]]-centric roadmap "no longer makes sense"
• New vision: L2s should be specialized (app-specific, privacy, ultra-low latency), not [[gloss:ethereum|Ethereum]] clones
• Only 3 major L2s have reached Stage 1 decentralization — Arbitrum, Optimism, Base. Base is #1 by [[gloss:tvl|TVL]] :::

:::details{badge=Advanced title="Ethereum as a Future Settlement Layer"} :::p{html} Despite the rethinking of the [[gloss:rollup|rollup]]-centric roadmap, [[gloss:ethereum|Ethereum]] L1 remains the foundation of the whole system. Its role gets refined: L1 is the layer where the most important settlements finally land and where assets that need maximum reliability live. For many use cases — institutional [[gloss:transaction|transactions]], large transfers, critically important [[gloss:smart contract|smart contracts]] — L1 is still the best choice, because the security guarantees are maximal and the network cannot be stopped by anyone. :::

:::p{html} Value accrual for [[gloss:ethereum|ETH]] comes from several sources: [[gloss:gas|gas]] for L1 [[gloss:transaction|transactions]], payment from L2s for posting data to L1 (via [[gloss:blob|blobs]]), staking yield for [[gloss:validator|validators]], and the fee burn through EIP-1559. This multi-layered model creates sustainable demand for [[gloss:ethereum|ETH]] regardless of where activity actually happens — L1 or [[gloss:layer 2|L2]]. ::: :::

:::details{badge=Advanced title="Composability and L2 Fragmentation"} :::p{html} When you use Aave, Curve, and Yearn together inside L1 [[gloss:ethereum|Ethereum]], they work like Lego bricks: a single [[gloss:transaction|transaction]] can flow through multiple protocols, picking up actions in each. This property is called composability, and it's what made [[gloss:defi|DeFi]] possible in the first place. The catch is that on [[gloss:layer 2|L2]] this magic stops working. :::

:::p{html} When liquidity is spread across dozens of L2s, the user has to "bridge" assets — that is, manually move them between networks. A bridge transfer takes minutes (sometimes days), costs [[gloss:gas|gas]] in two networks, and the bridge itself is the most attacked component in [[gloss:defi|DeFi]]: in 2022, hacks of the Wormhole, Nomad, and Ronin bridges produced $2.5 billion in losses. Liquidity for the same token (say, USDC) gets fractured across Arbitrum, Base, Optimism, zkSync, Linea, and the rest, instead of pooling in one place. The user has to remember where they have what, while protocol developers have to redeploy their code on every new [[gloss:layer 2|L2]] separately.

:::further{label="Further Reading"}

• L2Beat — comparison of L2 solutions | l2beat.com
• DeFiLlama — TVL tracker | defillama.com :::

:::quiz{id=q-2-2-s8 correct=2 explanation="Correct. On February 3, 2026, Vitalik publicly acknowledged that the original strategy needs revision. Reasons: L2s decentralized too slowly (only Arbitrum, Optimism, Base reached Stage 1), and L1 itself accelerated faster than expected. New vision: L2s should solve specific tasks (gaming, privacy, ultra-low latency for HFT), not be generic [[gloss:ethereum|Ethereum]] clones. This is a phase shift in the entire ecosystem."} What strategic shift did Vitalik announce on February 3, 2026?

• [[gloss:ethereum|Ethereum]] transitions back to [[gloss:proof of work|Proof of Work]]
• [[gloss:ethereum|ETH]] supply will be capped at 100 million
• The [[gloss:rollup|rollup]]-centric roadmap "no longer makes sense" — L2s should become specialized rather than generic [[gloss:ethereum|Ethereum]] clones
• All L2s become mandatory for all [[gloss:transaction|transactions]] :::

:::section{mins=2} Programmability Risks :::

:::lead A programmable platform has a dark side: [[gloss:smart contract|smart contracts]] have no brakes. If there's a bug in the code, the attacker takes the money — and no one in the protocol can roll back the [[gloss:transaction|transaction]]. From 2024 through April 2026, more than $9 billion has been stolen from the Web3 ecosystem. :::

:::p{html} The core [[gloss:ethereum|Ethereum]] principle is "Code is Law." Once a [[gloss:smart contract|smart contract]] is deployed, it executes exactly as written. No one can step in to change its logic, cancel a [[gloss:transaction|transaction]], or freeze funds at their discretion. That's the guarantee the whole system was built on — it's exactly why you can safely hold billions of dollars on [[gloss:ethereum|Ethereum]]. But the same principle cuts both ways. If there's a bug in a contract's code, an attacker can exploit it, and no one in the protocol can "roll back" the [[gloss:transaction|transaction]]. The stolen money becomes the attacker's, just as legitimately as if it had been transferred willingly. With a regular bank, you can call and freeze a transfer. With a [[gloss:smart contract|smart contract]], you can't. That puts a steep demand on developers: every significant [[gloss:smart contract|smart contract]] effectively has to be flawless from version one, because the cost of a mistake is users' money. :::

:::p The scale of the problem is hard to overstate. In 2024, more than $2.9 billion was stolen across the Web3 ecosystem (Hacken data). In 2025 it was $3.4 billion (Chainalysis) — the third-worst year in crypto's history. All told, the top-100 [[gloss:defi|DeFi]] exploits between 2014 and 2024 lost $10.77 billion combined. That list includes seriously big names: Bybit — $1.5 billion in February 2025 (the largest crypto theft in history), Ronin Bridge — $625 million in 2022, Poly Network — $611 million in 2021, and KelpDAO — $292 million in April 2026, which we discussed above. :::

:::p{html} Why does this keep happening? Several reasons, all direct consequences of programmability. The complexity of modern protocol code has grown to the point where ordinary tools can't verify every possible execution path. [[gloss:defi|DeFi]]'s composability (the ability of one protocol to use another like Lego bricks) is usually considered a feature, but it also acts as a risk multiplier: one bug in one protocol can cascade through dozens of dependent protocols. The open code is transparent not only to developers — an attacker reads exactly the same thing the authors do, and often finds the flaw faster. And crucially — once a contract is deployed, it typically can't just be fixed: the immutability that's required for trust becomes a trap when a bug is discovered. These vulnerabilities and their classes ([[gloss:reentrancy|reentrancy]], flash loans, access control, oracle manipulation) are discussed in detail in 6.8 [[gloss:smart contract|Smart Contracts]] §8, and composability-specific cascade risks are covered in 6.9 Governance and Composability §5. :::

:::p{html} In 2024-2026, a new dimension got added to all of this — artificial intelligence — and it works in both directions. On the defensive side, AI agents have become a powerful tool for auditors: modern tools have learned to automatically find vulnerabilities that conventional tests miss, and have already prevented hundreds of millions in potential losses. On the other side, the same technology has become a tool for attackers. Research shows that autonomous AI agents can already find and exploit a large share of known vulnerability classes on their own, and the cost of such scanning for an attacker is just a few dollars per contract, while the capability is growing fast. In practice, this means the window between deployment of a vulnerable contract and its exploitation is shrinking rapidly: developers used to have weeks to fix a discovered bug, now they have hours. We'll cover AI auditing tools, bug bounties, and formal verification in detail in 4.2 Audits and 6.8 §8. :::

:::p How is the ecosystem responding? The industry is maturing and moving away from the "move fast and break things" culture of early [[gloss:defi|DeFi]]. Layered defense is becoming the norm: professional audits before deployment, continuous online monitoring, substantial bug bounty programs, automatic "circuit breakers," insurance funds, and emergency security councils like Arbitrum's Security Council, capable of freezing stolen funds (though such measures themselves raise valid questions about decentralization — we touched on this in §5). No single one of these instruments provides full protection on its own; only the combination works. :::

:::p{html} The philosophical takeaway is simple but important: this is the price [[gloss:ethereum|Ethereum]] pays for programmability. [[gloss:bitcoin|Bitcoin]] consciously rejected Turing-completeness (2.1 §3) specifically to avoid this class of risk — and not without reason. [[gloss:ethereum|Ethereum]] and all programmable blockchains chose the other side of the trade-off: maximum flexibility in exchange for permanent responsibility for code quality. For a developer in 2026, that means writing [[gloss:smart contract|smart contracts]] is closer to designing aircraft control systems than to web development — every line has to be scrutinized to a much stricter standard. For a user, it means: know what you're putting your assets into — check audits, security history, the protocol's degree of decentralization. And remember that even the most thoroughly vetted [[gloss:smart contract|smart contract]] can be exploited tomorrow if someone finds a zero-day in it. :::

:::callout{icon=💡} With a regular bank, you can always reach someone human — call support, file a complaint, escalate to a regulator, go to court. With a [[gloss:smart contract|smart contract]], there's no one to reach. It runs like an automatic vault, configured once and put into perpetual operation: no owner, no "undo" button, no customer service. If a flaw is found in its logic, the money is gone — and legally it's as valid as any ordinary [[gloss:transaction|transaction]]. That's simultaneously the guarantee of reliability (no one can arbitrarily take your funds) and the source of risk (no one can help if something goes wrong). :::

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

• "Code is Law" cuts both ways: a guarantee against arbitrary interference, and at the same time an inability to reverse an exploit
• Scale: more than $10 billion stolen through [[gloss:defi|DeFi]] attacks between 2014-2024, $3.4 billion lost in 2025 alone
• AI amplified both sides: helps auditors find vulnerabilities, but gives attackers cheap tools for autonomous flaw-hunting
• The ecosystem's response — layered defense: audits, monitoring, bug bounties, circuit breakers, emergency councils :::

:::further{label="Further Reading"}

• Halborn: "Top 100 DeFi Hacks Report" | halborn.com
• Anthropic: "AI agents find $4.6M in smart contract exploits" | red.anthropic.com
• Immunefi — bug bounty platform | immunefi.com :::

:::quiz{id=q-2-2-s9 correct=2 explanation="Correct. "Code is Law" is the foundational principle: a [[gloss:smart contract|smart contract]] executes exactly as written, and no one can intervene. This works both ways: a guarantee against arbitrary interference (no government can freeze your funds), but also an inability to reverse anything if a bug is found in the code. The price of programmability — code quality must be perfect from version one. More than $10B has been stolen through [[gloss:defi|DeFi]] exploits over 2014-2024."} What is "Code is Law" and why does it cut both ways?

• A law that mandates [[gloss:ethereum|Ethereum]] developers must open-source their code
• A rule stating that only lawyers can deploy [[gloss:smart contract|smart contracts]]
• A deployed [[gloss:smart contract|smart contract]] runs exactly as written — a guarantee against arbitrary interference, but also an inability to reverse an exploit if a bug is found
• A framework for regulating [[gloss:defi|DeFi]] protocols :::