The Node

At block 74,638, mined on 15 August 2010, Bitcoin printed 184 billion coins. A transaction exploited an integer overflow — two outputs so large their sum wrapped around to a tiny negative number — and sailed through validation on a chain whose entire premise is a cap of 21 million. Within about five hours, Satoshi and Gavin Andresen shipped a patched client that rejected the overflow; within 53 blocks the honest chain had overtaken the poisoned one, and the network had collectively agreed those 184 billion coins never existed. No court ordered it. No company reversed it. Thousands of independent validators adopted corrected rules and kept building. This module is about where Bitcoin's rules actually live: in the machines that check them.

A full node takes nobody's word for anything. It downloads every block since genesis and verifies the lot: every signature against its public key, every script against its spending conditions, every input against the UTXO set so nothing is spent twice, and every coinbase against the schedule so no block mints more than subsidy plus fees. The 21-million audit is not a metaphor — one RPC call, `gettxoutsetinfo`, walks the entire UTXO set and returns the total money supply, and any node operator can run it in minutes without anyone's permission. No auditor of any fiat currency can do the equivalent. Run a node and nobody — not a miner, not an exchange, not a government — can lie to you about your money.

Validity in Bitcoin is checked, not voted on. A block that breaks the rules is rejected whether one miner produced it or all of them did; hashpower decides the order of valid transactions, never what counts as valid. This is the quiet inversion at the system's heart: miners look powerful — the warehouses of ASICs, the gigawatts — but they are payment processors, auditioned continuously by a jury of cheap computers. The community's aphorism for it, don't trust, verify, has no citable author because it needed none; it is simply a description of what the software does. Every node that accepts payment against its own validation is enforcing the monetary rules personally, and a rule change it rejects does not exist for its owner, whatever the rest of the world runs.

Not every device can carry the full ledger, and Satoshi said so in the whitepaper's section 8: simplified payment verification. An SPV client keeps only the block headers, checks the proof-of-work chain, and confirms a transaction's inclusion in a block via a Merkle proof. What it cannot do is check the transactions themselves. Satoshi was candid about the trade: the verification is reliable as long as honest nodes control the network, and weaker if an attacker overpowers it. SPV proves inclusion and work, not validity — a light wallet would happily follow a most-work chain containing invented coins, showing green ticks all the way, where a full node would reject the first bad block. (A pruned node, by contrast, validates everything and merely discards old raw blocks afterwards: full security in about ten gigabytes.)

The value-overflow incident is why none of this is theoretical, and 2018 proved it was not a one-off. A performance optimisation from 2016 had quietly removed the duplicate-input check, leaving a path — live for two years, in software guarding hundreds of billions of dollars — by which a miner could have spent the same coin twice in one block and inflated the supply. CVE-2018-17144 was reported on 17 September 2018 and patched within hours of triage; it was first described publicly as a mere crash bug so the network could upgrade before the inflation vector was revealed, and it was never exploited. The sober lesson cuts both ways. The code guarding the 21 million cap is code, and code has bugs. The defence is not perfection but redundancy: thousands of independently run validators, and a culture of rapid, careful disclosure.

Then there is the political proof. In May 2017, around 58 companies commanding roughly 83% of hashrate signed the New York Agreement: SegWit would activate, followed by a doubling of the block size. Users were never party to it, so they answered in software — BIP148, a user-activated soft fork set for 1 August 2017, whose nodes would reject blocks that failed to signal SegWit. Under that pressure miners activated early, and SegWit locked in at block 481,824. The 2MB half of the bargain fared worse: on 8 November 2017 its own organisers cancelled the hard fork by email, conceding they had not built sufficient consensus. A supermajority of hashrate and the industry's largest firms proposed new rules; the people running validating nodes declined to run them; the change died.

Running a node today is deliberately unremarkable. The chain stood at roughly 756 GB in mid-2026 and grows by 80–90 GB a year, so a 1 TB SSD is the floor — about £300 of hardware all-in, with Raspberry Pi-class machines a well-trodden path, or around ten gigabytes if you prune. The initial block download is the painful part: hours to days of your machine re-checking seventeen years of history, which is not a bug but the entire point. What you get is independence and privacy. Use a block explorer or a custodial wallet and you are trusting someone else's node and handing them your addresses; your own node answers your wallet's queries without telling anyone what it looked up. Your node is your vote on what Bitcoin is.