Key Principle
Consensus is "a system of strict rules without rulers" (Chapter 14): with no central authority to decide what is true, every node must algorithmically converge on one world-state. Consensus secures the chain and keeps control decentralized — currency issuance is the incentive, not the goal.
Why This Matters
Consensus is the layer that makes a public blockchain's value proposition (censorship resistance, permissionlessness) possible. It sits below the smart-contract layer — contract authors needn't understand it — but it determines what "immutable" actually guarantees. The DAO fork proved immutability is a value the community chose to defend, not an unbreakable law of physics.
Proof of Work vs Proof of Stake
Both coerce honesty via reward (newly minted currency) + punishment. The core distinction is where the punishment lives:
- PoW (Ethereum's Ethash): punishment is extrinsic — loss of funds spent on electricity. Ethash evolved from Dagger–Hashimoto (Buterin's Dagger + Dryja's Hashimoto) and was made deliberately memory-hard for ASIC resistance: every hash must traverse a large, frequently-accessed dataset, the DAG, so memory bandwidth (not raw compute) is the bottleneck — and GPUs already have abundant bandwidth, so an ASIC can't be orders of magnitude faster. DAG: ~1 GB initial, grows linearly, regenerated each epoch (30,000 blocks ≈ 125 hours). ASIC resistance was expected to erode because Ethereum's stated plan to abandon PoW could be sidestepped by ASICs mining other Ethash coins (PIRL, Ubiq, Ethereum Classic).
- PoS (Casper): punishment is intrinsic — loss of staked ether. A validator locks ether as a deposit, validators take turns proposing/voting weighted by deposit size, and a validator "risks losing their deposit if the block they staked it on is rejected by the majority." A difficulty bomb was engineered into PoW to exponentially raise mining difficulty and force migration to PoS.
At publication (2018), Casper was "still under active research... not implemented," pursued in two variants: Casper FFG ("Friendly Finality Gadget," first proposed as a hybrid PoW/PoS step) and Casper CBC ("Correct-by-Construction"). In June 2018 Buterin scrapped the hybrid for pure PoS. Buterin's trade-off: "CBC seems to have nicer theoretical properties, but FFG seems to be easier to implement."
[DATED 2018]The Merge (Sept 2022) replaced PoW entirely with PoS. "Miners" → "validators" (staking 32 ETH); Ethash, the DAG, epochs, and GPU-mining dynamics were removed from mainnet; the difficulty bomb is now moot. What shipped is Gasper = Casper FFG (finality) + LMD-GHOST (fork choice). The hybrid FFG+PoW model was abandoned (the text itself notes this) and CBC Casper never reached mainnet. The intrinsic-punishment design the book describes — dishonesty punished by slashing — is exactly what was realized. The trade-off framework below remains useful.
Principles of Consensus (evaluation framework — still useful)
A durable way to interrogate any "this consensus is better" claim by forcing "Better at what?" Ask:
- Who can change the past, and how? (immutability)
- Who can change the future, and how? (finality)
- What is the cost to make such changes?
- How decentralized is the power to make them?
- Who will know if something changed, and how?
Optimizing all dimensions at once is likely impossible — "no algorithm can optimize across all dimensions of the problem of decentralized consensus" — so different applications need different answers, and the trade-offs play out empirically.
The DAO Hard Fork & the ETH/ETC Split
The case where immutability collided with intervention:
- The DAO (Slock.it, 2016) amassed
14% of all ether ($150M) — one contract bug = network-wide crisis. - Reentrancy bug: a withdrawal function did the external ether transfer before updating the caller's entitlement, so the attacker's fallback re-entered and drained repeatedly. Drained ~3.6M ether (~30%). A built-in 28-day withdrawal delay was the only window in which any response could be designed.
- Irregular state change — block 1,920,000 (July 20, 2016): a hard fork that rewrote balances (not just forward rules) to return the ether "as though The DAO had never even existed... technically easy, if rather politically controversial."
- Chain split: nodes holding "immutability should be a fundamental principle... without exception" rejected the fork and continued the original chain as Ethereum Classic (ETC) ("code is law"); the forked chain is ETH. A committed minority running old client software keeps the old chain alive — falsifying "majority hash power = consensus." A replay hazard existed until replay protection was added (ETC: Jan 13, 2017). Governance lesson: the CarbonVote had only 5.5% turnout with ~a quarter of the pro-fork vote from a single address — low turnout + whale concentration undermines any "community consensus" claim.
Key Quotes
"the punishment in PoS is intrinsic to the blockchain (e.g., loss of staked ether), whereas in PoW the punishment is extrinsic (e.g., loss of funds spent on electricity)." — Antonopoulos & Wood, Chapter 14
"The purpose of the DAG is to make the Ethash PoW algorithm dependent on maintaining a large, frequently accessed data structure." — Antonopoulos & Wood, Chapter 14
"no algorithm can optimize across all dimensions of the problem of decentralized consensus." — Antonopoulos & Wood, Chapter 14
"The entire blockchain industry is one giant experiment... In the end, history will answer the controversy." — Antonopoulos & Wood, Chapter 14
Rules of Thumb
- Treat the five-question framework (past/future/cost/decentralization/visibility) as durable; treat PoW/Ethash specifics as historical.
- Never fork without whole-community consensus; a contentious fork without it splits the chain (ETH/ETC).
- Distrust low-turnout, whale-weighted "community consensus."
- Remember: agreement among nodes secures one state; it does not make external data trustworthy.
Related References
- Implementation Playbook — Build, Test, Deploy - node modes, clients, and the spec-over-implementation security property
- Smart Contract Vulnerability Catalog - the reentrancy bug that triggered the DAO fork
- Rules of Thumb — Heuristics Across All Chapters - Architecture & Consensus heuristics