Comparing consensus mechanisms
No consensus mechanism wins everywhere. This lesson lines up the trade-offs so you can read any "X kills Bitcoin" headline critically.
In this lesson
- Compare PoW, PoS and BFT-style consensus on the same axes
- Explain probabilistic vs deterministic finality
- Identify which mechanism fits which use case
The scorecard
- "Safe" grows with depth
- Simple, battle-tested at scale
- Bitcoin, pre-merge mining view
- Final once a supermajority attests
- Fast confirmation, larger message overhead
- Ethereum checkpoints, Tendermint chains
| Mechanism | Sybil resistance | Energy | Typical finality | Main risk |
|---|---|---|---|---|
| PoW | Hashrate | High | ~60 min (6 blocks) | Pool/miner concentration |
| PoS | Staked capital | Very low | ~15 min (2 epochs) | Stake concentration / slashing bugs |
| PBFT/Tendermint | Delegated stakes | Low | Seconds | Validator collusion at small N |
| PoA | Reputation of approved nodes | Negligible | Seconds | Permissioned — trust the operators |
Sybil resistance is the core question: how do you stop one actor from creating 10,000 fake nodes to outvote everyone? PoW answers "make identities cost hardware and electricity"; PoS answers "make them cost locked capital"; permissioned chains answer "don't allow strangers in".
Example · Which chain for which job?
Settlement where censorship-resistance is paramount → PoW/PoS public chains. High-throughput application chain with known validators → Tendermint-style BFT. Internal audit trail among partner banks → permissioned ledger, maybe no token at all.
Is one consensus "more decentralised"?
Measure it: Nakamoto coefficient for PoW, Nakini/Gini for stake distribution, node counts and geographic spread for both. Marketing claims are not metrics.
What about layer 2s?
Rollups settle onto L1 consensus and add their own fraud/validity proofs — they inherit L1 security while moving execution off it.