Beyond PoW: PoS, BFT

Proof of Work is one point in a design space, optimized for one extreme: nobody knows who the participants are, and anyone may join or leave at will. Relax that assumption and much cheaper consensus becomes available. An engineer's job is matching the algorithm to the trust model — which is literally this lesson's exercise.

Proof of Stake

Replace "one hash, one lottery ticket" with "one staked coin, one lottery ticket." Validators lock coins as collateral; the protocol pseudo-randomly selects a proposer, weighted by stake; provable misbehavior (like signing two conflicting blocks) gets the stake slashed. Security still costs something an attacker must acquire — but it's capital at risk instead of burned electricity. Ethereum switched in the 2022 Merge, cutting its energy use by ~99.95%; Cardano, Solana, Polkadot, and Cosmos chains are PoS-native. The honest trade-offs: the classic nothing-at-stake objection (voting on every fork is free unless slashing punishes it — that's precisely what slashing is for), long-range rewrites of old history using once-valid keys (countered by checkpointing / "weak subjectivity"), and a rich-get-richer tilt, since stake earns yield.

The BFT family

Byzantine Fault Tolerance comes from a different lineage — distributed-systems theory, not cryptocurrency. Lamport, Shostak, and Pease formalized the Byzantine generals problem in 1982; Castro and Liskov's PBFT (1999) made it practical: among n = 3f + 1 known validators, up to f can be arbitrarily malicious and the rest still agree, with immediate finality — a committed block is final, no confirmations, no reorgs, ever. Cost: validators are a fixed, known set, and communication is O(n²) messages per decision, so counts stay in the dozens-to-hundreds. Descendants refine it: Tendermint (started by Jae Kwon in 2014, later developed with Ethan Buchman; the engine of Cosmos) integrates BFT with PoS validator selection; HotStuff (2019) got messages to O(n) and powered Facebook's Diem. Modern "PoS" chains are mostly hybrids: PoS chooses who may vote, a BFT-style protocol decides how votes commit.

And plain old Raft

If nodes can crash but never lie — one company's internal cluster — Byzantine defenses are pure overhead. Raft (and Paxos) tolerate crash faults only: f failures among 2f + 1 nodes, a simple elected leader, no signatures, no adversary model. This is etcd, Consul, and every "strongly consistent" database you've deployed. Using PoW for a microservice cluster is malpractice; so is using Raft between mutually distrusting banks.

The decision rubric

Two axes decide everything: is membership open or closed? and can participants be malicious or merely crash? Open + malicious → PoW/PoS. Closed + malicious → BFT. Closed + honest-but-crashy → Raft.

Your exercise

Each input line describes a scenario; print exactly one token: POW, POS, BFT, or RAFT (uppercase). The graded confusion is BFT vs RAFT — both run among known nodes, so scan for trust: "Permissioned consortium of 7 banks" and "12 known financial institutions" are mutually distrusting parties → BFT; "Internal microservice cluster needing strong consistency" and "3 trusted nodes coordinating leader election" are one administrative domain → RAFT. Second catch: "Public, super-fast finality (Solana-like)" is POS — fast public finality is a stake-plus-BFT hybrid, not raw BFT, because the validator set is open.