Consensus

Nobody votes and nobody is in charge. Each node checks every block against the same rulebook, then follows whichever valid chain carries the most work, which is how a laptop syncing from scratch reconstructs the identical history back to January 3, 2009. When you buy bitcoin, this is the machinery that decides the coins you were sent are actually yours.

How it works

Agreement is reached by asking two questions in a fixed order, never by taking a poll.

The first question is validity. A node receives a block and checks it alone: signatures, amounts, the merkle root, the subsidy claimed by the miner, the weight, the timestamp. Anything that fails is dropped on the spot, and the amount of electricity burned producing it is irrelevant. The second question is choice between valid options. When two competing histories both pass, a node keeps the one with the greatest total work summed across its headers, which is not the same as the one with the most blocks: a shorter branch of harder blocks beats a longer branch of easier ones.

Notice what is missing. Nodes never exchange opinions, never announce a preference, and never count each other. They relay blocks and transactions, apply the same test to what arrives, and end up in the same place because the test is deterministic. Agreement is a side effect of everyone doing arithmetic separately, which is why the network has no quorum to corrupt and no committee to subpoena.

Brief disagreement is routine. Two miners occasionally solve within a second of each other and half the world sees one block first, so for a few minutes there are two candidate tips. The next block breaks the tie, one branch is abandoned as a stale block, and its transactions return to the mempool. Deep disagreement is the interesting case, and it only happens when nodes are running different rules. Tighten the rules and old software still accepts everything new software produces, which is a soft fork: Taproot arrived that way in November 2021 and split nothing. Loosen them and the two groups permanently follow different chains, which is a hard fork, and that is exactly what happened on August 1, 2017 when Bitcoin Cash separated.

Why this matters when you buy bitcoin

Consensus is what makes a bitcoin balance yours rather than a promise from whoever is holding it.

The chain agrees about outputs and the keys that can spend them. It has no concept of an account holder. While your coins sit on an exchange, consensus sees the exchange's addresses and nothing else, and your claim on them is a row in that company's database, enforced by contract law rather than by nodes. Move them to a wallet whose keys you hold and the position reverses: the global rulebook now recognises your spending authority directly, and no support ticket, freeze, or insolvency changes that. This is the entire argument behind self custody, and it is a statement about consensus rather than about trust.

Forks make the distinction visible. Anyone holding their own keys on August 1, 2017 ended up controlling coins on both chains, because both chains inherited the same history of outputs. Anyone holding through a platform got whatever that platform chose to do, and several took weeks to credit the new asset or declined to support it at all. The same logic applies to any future split.

Consensus is also global in a way regulation is not. The 231 country guides on this site differ enormously on what is legal, what must be reported, and which platforms will accept you, but none of that reaches the ledger. A coin bought in Germany and a coin bought in Nigeria are validated by the same rules, live in the same UTXO set, and are indistinguishable to a node. What changes at a border is who will sell to you and what you owe your tax authority, never what the network considers valid.

The six hours in March 2013 when bitcoin ran two chains

Block 225,430, mined on March 11, 2013, was accepted by one version of the software and rejected by the version before it.

Bitcoin 0.8 had switched its database from Berkeley DB to LevelDB, and in doing so it quietly dropped an undocumented limit on the number of locks the older database could take while processing a large block. The block exceeded that limit. Nodes running 0.7 refused it, nodes running 0.8 built on it, and the network carried two chains for about six hours. One double spend of roughly 10,000 dollars got through against the payment processor OKPay.

The resolution says something no protocol document does. Developers asked mining pools to downgrade to 0.7 so that the more restrictive chain would win, and the pools did it, discarding blocks they had already been paid for. Version 0.8.1 shipped shortly after with a temporary cap of 500,000 bytes on the blocks it would produce, and BIP50 was written as the postmortem. The episode is the standard proof that bitcoin's rules are whatever the deployed software actually enforces, bugs included, and that when the software disagrees with itself, humans have to choose.

Consensus vs consensus rules

Consensus is the outcome; the consensus rules are the criteria used to reach it. The rules are a fixed list of validity checks written into code and identical for everyone running compatible software. Consensus is the moving result of applying that list to whatever blocks have arrived, which is why two honest nodes can be momentarily out of consensus, holding different tips after a near simultaneous block, while agreeing perfectly about the rules. Change the rules and you change what consensus is possible; change nothing and consensus still shifts every ten minutes as new work lands.