The problem in one sentence
How do many parties agree on a plan when some may lie, be offline, or try to sabotage the outcome? That’s the Byzantine Generals Problem. In blockchains, the “plan” is the next block; the liars are malicious or faulty nodes.
The crux: honest participants need a way to converge on one story, even if some mess with the messages. If they don’t, double-spend and censorship become easy.
| Role | Behavior | Consensus impact |
|---|---|---|
| Honest node | Follows rules, relays messages | Helps convergence |
| Faulty node | Offline or misconfigured | May slow convergence |
| Malicious node | Sends conflicting messages | Attempts to fork/confuse |
Why it matters for money
If honest nodes can’t agree on the same ledger, attackers can double-spend or censor. The network needs a way for honest participants to converge even if some peers are malicious or sending conflicting messages.
Money without a referee only works if conflicting versions of history get resolved the same way everywhere. Consensus rules are the script; the network plays it out every block.
Classic BFT vs. open networks
Traditional Byzantine Fault Tolerance (BFT) assumes a known set of validators and can tolerate up to ~1/3 bad actors with fast finality. Great for permissioned systems; less suited to open, anonymous networks with unknown participants.
Open networks need sybil resistance: proof that a participant has skin in the game (work or stake) so one actor can’t spin up thousands of fake identities for free.
| Aspect | Classic BFT | Open (PoW/PoS) |
|---|---|---|
| Validator set | Known, fixed | Open, permissionless |
| Fault tolerance | ~1/3 by count | Depends on hash/stake weight |
| Sybil resistance | Identity/PKI | Work or stake cost |
| Finality | Fast (round-based) | Probabilistic or economic |
Proof-of-Work’s approach
PoW doesn’t ask nodes to vote; it asks them to burn energy. The longest (most work) valid chain wins. An attacker needs majority hashpower to consistently override honest miners. There’s no fixed validator list; sybil resistance comes from cost.
Proof-of-Stake’s approach
PoS bonds capital instead of energy. Validators propose and attest; misbehavior can be slashed. Finality gadgets let nodes agree that certain blocks are locked in. Sybil resistance comes from stake weight and economic penalties.
Faults and attacks
Eclipse: Isolate a node so it only sees attacker-controlled peers, feeding it a fake view. Mitigation: diverse peers, multiple connections, checkpoints.
Equivocation: A validator/miner builds conflicting blocks. PoW wastes their hashpower; PoS can slash equivocations if detected.
51%/finality attacks: Majority hashpower (PoW) or supermajority stake (PoS) can reorder recent history. Visible, expensive, and reputationally costly, but possible if control concentrates.
| Attack | What it does | Mitigation |
|---|---|---|
| Eclipse | Isolate victim’s view | Peer diversity, checkpoints |
| Equivocation | Build conflicting blocks | Waste (PoW) or slash (PoS) |
| 51% / finality flip | Reorg or censor | Avoid concentration; make it costly |
Finality and confidence
In PoW, more confirmations lower the probability of a successful reorg. In PoS with finality gadgets, finalized blocks require slashing to revert, making deep changes economically painful.
“Byzantine fault tolerance in the wild means honest nodes agree despite liars, because cheating costs real resources or gets you slashed.”
Limits to remember
No consensus is magic. If the majority of work or stake is malicious, they can misbehave—though they pay in energy or capital and leave evidence. Users rely on open verification, diverse peers, and, in PoS, honest checkpoints.