Glossary

    Glossary

    Byzantine Fault Tolerance

    Byzantine fault tolerance is a network's ability to keep agreeing on one history even when some participants are broken or malicious. It is the consensus problem in human terms - generals who cannot trust their own messengers - and every blockchain is one answer to it.

    FoundationsLast reviewed 2026-10-01

    What it is

    Byzantine fault tolerance is the property of a distributed system that keeps working when some of its participants fail or act maliciously. The name comes from a 1982 thought experiment: several generals must agree on a joint attack, but they communicate only through messengers, and some of the generals - or messengers - may be traitors. How do loyal generals agree when they cannot know who to trust?

    Every blockchain is one answer to that question. Proof of work solves it probabilistically: rewriting history requires out-computing the honest majority, which gets more expensive as the network grows. Proof-of-stake networks solve it with rules: a supermajority of stake must agree, and misbehaving stake is penalised. The practical threshold for deterministic systems is that the network tolerates fewer than one-third of participants being faulty - beyond that, agreement is not guaranteed.

    Key takeaways

    • Byzantine fault tolerance means a network keeps agreeing despite broken or malicious participants.
    • The name comes from the 1982 Byzantine Generals thought experiment.
    • Deterministic systems tolerate fewer than one-third faulty participants.
    • Bitcoin solves the problem probabilistically via proof of work.
    • Proof-of-stake networks solve it with supermajority rules and penalties.

    How it works

    Walk the generals. Commanders on separate hills must agree to attack at dawn or retreat. Messengers carry the messages, and a traitor can send conflicting orders to different commanders or lie about what was received. The mathematical result, proven in 1982, is that no solution exists unless the loyal generals are a strong majority - for deterministic agreement, more than two-thirds. Below that threshold, the traitors can keep the honest participants from ever agreeing on one plan.

    Bitcoin's answer reframes the problem. Instead of requiring instant agreement, it makes agreement expensive to fake: each block requires work that costs real energy - the proof-of-work entry covers the mechanics - and the longest chain wins. Rewriting history means out-computing the honest network, so the traitor's cheapest option is honesty. The 51-percent-attack entry covers what happens when that assumption fails.

    Proof-of-stake networks answer deterministically. A supermajority of stake - two-thirds on Ethereum-style systems - must agree before history is finalised, and conflicting signatures are penalised - the slashing entry covers the penalties. The consensus-mechanism entry covers how the two rulebooks compare.

    Why it matters

    For an Australian reader, Byzantine fault tolerance is why you can trust a ledger nobody owns. Every exchange, wallet and price tracker on this site assumes the underlying chain settles correctly, and the chain settles correctly because its answer to the generals problem holds. When a network's tolerance is broken - the 2019 and 2020 attacks on Ethereum Classic are the worked cases - double-spends become possible and everything built on the chain inherits the risk.

    A colleague named Brent (yes, that Brent) once asked why crypto 'needs all this maths when a database works fine'. The answer is ownership: a database has an owner you trust, and a blockchain has none. The highest-leverage thing most Australians can do is prefer networks whose consensus answers are battle-tested - read the consensus-mechanism entry, check the 51-percent-attack entry for the failure cases, and treat novel consensus as a risk, not a feature.

    Real-world examples

    The generals problem in one message chain

    Five generals must agree to attack at dawn. One is a traitor sending conflicting orders: attack to two generals, retreat to two others. With fewer than the required majority loyal, the honest generals cannot distinguish honest messages from traitorous ones, and agreement fails. The 1982 proof showed no messaging scheme fixes this below the threshold - the problem is trust itself, not the post office.

    Bitcoin's probabilistic answer

    Bitcoin does not require instant agreement; it makes agreement expensive to fake. Each block costs real energy to produce, and the longest chain wins. A traitor wanting to rewrite history must out-compute the honest network - at roughly 700 exahashes per second of global hash rate, that is industrial money against industrial money. The traitor's cheapest option is to be honest, which is the design working. The hash-rate entry covers the security budget.

    The Ethereum Classic attacks as the failure case

    Ethereum Classic, a smaller proof-of-work chain, suffered 51 percent attacks in 2019 and 2020: an attacker rented enough hash rate to rewrite recent history and double-spend. Roughly $1 million USD and later smaller sums were double-spent, and exchanges paused deposits. The tolerance threshold failed because the honest majority was small enough to out-rent. Smaller networks inherit the risk, which is why depth of consensus belongs next to any yield you are eyeing.

    Common mistakes

    Mistake: Assuming every blockchain tolerates one-third faulty participants.

    Why it bites you: That threshold applies to deterministic systems like proof-of-stake networks. Bitcoin solves the problem probabilistically, and small proof-of-work chains can fail the threshold entirely - the ETC attacks are the cases.

    Mistake: Treating novel consensus designs as a feature.

    Why it bites you: The consensus answer is the foundation everything else sits on. Novel designs have not survived the incentive attacks that Bitcoin and Ethereum's have, and the foundation is the wrong place to experiment with your money.

    Mistake: Assuming a database works fine instead.

    Why it bites you: A database has an owner you must trust; a blockchain does not. The maths is the price of removing the owner, and Brent's question skipped that trade.

    Mistake: Ignoring network size when evaluating a chain.

    Why it bites you: Tolerance thresholds mean nothing if the honest majority is cheap to out-rent. The [51-percent-attack entry](/glossary/51-percent-attack/) covers the worked cases, and network size is the check that matters.

    Ivy's take

    I spent an evening rereading the Byzantine Generals paper so you don't have to, and the 1982 result is refreshingly blunt: below the threshold, no clever messaging fixes trust. Every blockchain is one answer to that problem, and the answers split into two families - make agreement expensive to fake, or require a supermajority to agree. Bitcoin and Ethereum sit in different families and both hold, which is the closest thing consensus research has to a two-party system.

    I ran the failure cases through the same lens: Ethereum Classic's attacks failed the threshold because its honest majority was cheap to rent against, not because the maths was wrong. The advice I use is one line: prefer consensus that has survived its incentive attacks, and treat 'novel consensus' in a project's pitch as a cost, not a feature. The generals problem has been solved twice. The solutions are boring, and boring is the compliment.

    Related jargon, defined

    Consensus mechanism
    The rulebook a network follows to agree on one history.
    Proof of work
    Consensus by energy: rewriting history requires out-computing the honest network.
    Proof of stake
    Consensus by locked stake: a supermajority must agree, and misbehaviour is penalised.
    51% attack
    An attack where one party controls the majority of consensus power.
    Finality
    The point after which a network's history cannot be rewritten.
    Double-spend
    Spending the same coins twice, the failure consensus exists to prevent.

    Frequently asked questions

    What is Byzantine fault tolerance in simple terms?+

    It is a network's ability to keep agreeing on one version of history even when some participants are broken or malicious. The name comes from a 1982 thought experiment about generals who cannot trust their messengers.

    How do blockchains solve the Byzantine Generals problem?+

    Two ways. Proof of work makes agreement expensive to fake by requiring energy, and proof of stake requires a supermajority of stake to agree, with penalties for misbehaviour. Both hold on the major networks.

    What is the one-third threshold?+

    For deterministic agreement, a distributed system tolerates fewer than one-third of participants being faulty or malicious. Beyond that, honest participants cannot be guaranteed to agree, which is where consensus attacks become possible.

    Has a Byzantine fault tolerance failure happened in practice?+

    Yes. Ethereum Classic suffered 51 percent attacks in 2019 and 2020 where attackers rewrote recent history and double-spent. The chains that failed were small enough that the honest majority was cheap to out-rent.

    Why does this matter if I just track prices?+

    Every price you see assumes the underlying chain settles correctly, which is the consensus answer holding. When tolerance fails, double-spends and deposit pauses follow, and everything built on the chain inherits the risk.

    Disclaimer: This information is general in nature and does not constitute financial or legal advice. Always consult a qualified professional for your specific situation.

    Keep learning

    Ivy Sinclair-Wren

    Ivy Sinclair-Wren

    Financial Chaos Analyst

    Connect on LinkedIn

    Ivy Sinclair-Wren is a Financial Chaos Analyst covering crypto markets, investing behaviour, and the Australian habit of checking prices during family dinners. Based in Melbourne, she reviews AUD Crypto Tracker's coin guides for accuracy, ATO alignment, and ruthless utility — no hype, no predictions, just what the data says.