How a blockchain works, from the hash to the confirmation

A blockchain is a list that only grows at the back and whose entries can no longer be changed without it showing. All it needs for that are three building blocks: a computing function that boils data down to a fixed length, a rule that makes new entries expensive, and many computers that check the same thing independently of each other.

What a hash is and why it cannot be reversed

A hash is a fingerprint for data. Bitcoin uses the SHA-256 procedure: whether you feed in one letter or a file of 10 gigabytes, what comes out is always a value of 256 bits, written as 64 characters in hexadecimal. Two properties make it useful. First, moving a single comma changes the entire output value. Second, there is no way back: the text cannot be restored from the fingerprint, you can only guess and compare.

The size of the number space makes guessing hopeless. 256 bits mean 2 to the power of 256 possible values, a number with 78 digits. That hopelessness is exactly the basis for the next step. How a hash behaves when you change a single letter is something you can try out in the hash demo.

How blocks become a chain

A block consists of a header and the transfers. The header is small: the 2008 whitepaper works with about 80 bytes per block header and so arrives at 80 bytes times 6 times 24 times 365, that is 4.2 MB a year for all headers together. Among other things, the header holds the timestamp, a root value over all the transfers it contains and, decisively, the hash of the previous block.

That backward reference is the chain. Anyone changing block 500,000 after the fact changes its hash; the backward reference in block 500,001 then no longer fits, and all following blocks become invalid. A forger would have to redo the entire computing work from the changed point on and be faster than the rest of the network while doing it.

How much fits into a block is limited. Originally a cap of 1,000,000 bytes per block applied. Proposal BIP 141 replaced it in 2017 with a weight limit of 4,000,000 weight units, calculated as base size times 3 plus total size. How many transfers that allows depends on their size and how they are built. The often quoted figure of 7 transfers per second comes from an article in the Bitcoin Wiki that derives throughput expressly from the old cap of 1,000,000 bytes per block and does not take the move to weight units into account. It is still quoted as a rough order of magnitude, but it is no use as a current measurement.

All of this is distributed by nodes. A full node downloads every block, checks every signature and every rule itself, and passes on only what gets through. It asks nobody for permission and trusts no other computer. The whitepaper describes the sequence in 6 steps: transfers are distributed, every node collects them in a block, looks for the computing solution, distributes the find, the others check it, and their agreement consists in carrying on working on that block. Anyone who wants a different rule to take hold has to convince the operators of those nodes.

Why the difficulty is reset every 2,016 blocks

For a block to be valid, its hash has to sit below a target value. To get there, miners change a counting number in the header and keep trying until it fits. How hard that is, the network regulates itself: every 2,016 blocks the target value is recalculated. 2,016 blocks at intervals of 10 minutes would be exactly two weeks; if the last round was faster, the difficulty rises, if it was slower, it falls.

Difficulty 1 corresponds to the highest possible target value and is set in the code as 0x1d00ffff; the current difficulty is that starting value divided by the target value in force. How much computing power stands behind it is shown by a figure from the University of Cambridge survey: at the end of 2024 the derived network power stood at 796 EH/s, that is 796 quintillion hash attempts per second. The link between reward and computing power is set out in the piece on the halving.

How many confirmations are enough

A transfer counts as confirmed as soon as it sits in a block. Every further block on top is another confirmation. The widespread rule of thumb of 6 confirmations is no constant of nature: the Bitcoin Wiki explains that it rests on the assumption that an attacker brings hardly more than 10 per cent of the computing power and that a residual risk below 0.1 per cent is acceptable.

The whitepaper works this case through in section 11, under two clear assumptions: the attacker only starts working secretly on his own chain once the transfer is sent, and the honest blocks take the expected time on average. For an attacker with 10 per cent of the computing power, his probability of success then falls from 20.46 per cent after 1 block, through 1.32 per cent after 3 blocks, to 0.0243 per cent after 6 blocks. At 30 per cent of the computing power it still stands at 17.74 per cent after 5 blocks and only reaches 4.17 per cent after 10 blocks. For very large amounts the wiki therefore suggests 144 blocks, about one day.

The waiting time is random, not on a timer. Within 10 minutes a block is found with a probability of around 63 per cent, within 30 minutes with 95 per cent, within 60 minutes with 99.7 per cent.

Proof of work and proof of stake side by side

Both procedures solve the same task: they determine who may write the next block, and they make cheating expensive. The difference lies in what is put at stake. With proof of work it is electricity and hardware, with proof of stake it is deposited capital that is partly destroyed in case of misbehaviour. Ethereum switched over on 15 September 2022.

The switch cut annual electricity use by more than 99.988 per cent and CO2 output by around 99.992 per cent, to about 870 tonnes of CO2 equivalent a year. These figures appear on the pages of the ethereum project itself, where they come from a study by the Crypto Carbon Ratings Institute and are expressly described as an estimate based on publicly available data, not as an official figure. Anyone reading them should therefore know that the publisher is speaking on its own behalf. What further differences exist between the two networks is set out in the comparison of ethereum and bitcoin.

What you can check on a chain yourself

Four points can be settled without specialist knowledge before you trust a chain. First the computing power or the deposited stake: with small networks a fraction of the power of large networks is sometimes enough to get past 50 per cent, and that is exactly where the danger of a majority attack lies, in which the attacker recreates the changed block and every block after it and puts a longer chain alongside.

Second, the number of independent nodes: the more computers hold a full copy and check every rule themselves, the harder a change is to push through. Third, the confirmations: 1 confirmation is often enough for a small amount, 6 are the usual compromise, 144 count as cautious for very large sums. Fourth, the question of who maintains the software and how changes are decided.

What a blockchain does not do belongs in the picture as well: it checks no content, only rules. A false entry that meets every rule stays there for good. And it does not replace custody: the key sits with you or with a provider. The basics are set out in the piece on bitcoin, and specialist terms in the glossary.

Frequently asked questions

Sources