What a Blockchain Actually Is, in Plain Terms

A blockchain is a record book that many computers keep identical copies of. New entries are bundled into blocks, and each block carries a digital fingerprint of the block before it, so altering an old entry breaks every fingerprint after it and the other copies expose the change. NIST, the US standards agency, describes it as a distributed digital ledger that is "tamper evident and tamper resistant", a more careful claim than the common word "unchangeable".

Key takeaways

A blockchain has three parts: blocks of grouped transactions, hashes that chain the blocks together, and many copies of the ledger that agree through a set of rules called consensus.
Changing one character in a block's data produces a completely different hash, so tampering with an old block is visible to anyone who compares copies.
Rewriting history is possible in theory but expensive, and the cost rises with every block added on top.
Blockchains are not fast or able to confirm facts about the physical world, and public chains are not private; a chain records what was submitted, not whether it is true.
Outside cryptocurrency, results are mixed: the TradeLens shipping platform was discontinued, with shutdown planned for early 2023, and ASX chose a product-based design for its stock-settlement replacement after a review of its blockchain-based one.

Blocks, hashes and a shared ledger

A ledger is a list of transactions. In a blockchain, transactions are grouped into blocks, and each block is linked to the previous one. NIST's Blockchain Technology Overview (NIST IR 8202, October 2018) puts it this way: transactions are cryptographically signed, grouped into blocks, and each block is linked to the previous one after validation and a consensus decision. New blocks are replicated across the copies of the ledger held in the network.

The ledger is not stored in one place. Each participating computer, called a node, keeps a copy and checks new blocks against the rules itself. The idea was laid out in the 2008 Bitcoin whitepaper, which proposed a peer-to-peer timestamp system so that payments would not need a trusted third party to prevent the same coin being spent twice. The technology became widely known with the launch of the Bitcoin network in 2009, according to NIST, but NIST also notes it applies to a wider range of uses than cryptocurrency.

What a hash does

A hash function turns any input into a fixed-length string of characters, a fingerprint. SHA-256, the function Bitcoin uses for block hashes (it applies SHA-256 twice, per the Bitcoin developer reference), always returns 256 bits, usually written as 64 hexadecimal characters. The same input always gives the same output, and a tiny change in the input gives an unrelated-looking output.

The table shows this, computed with Python's standard hashlib library. The two inputs differ by one character (5 versus 6).

Input text	SHA-256 hash (first 24 of 64 characters)
Alice pays Bob 5 coins	40a417b6c609285e3d3d1b9f
Alice pays Bob 6 coins	87b27ad83c9c5e9bae680e78

Comparing the full outputs, 61 of the 64 characters differ and 131 of the 256 bits differ. About half the bits flipping is what a good hash function should produce, and nothing in the new hash reveals how small the edit was.

How hashes make a chain

Each block stores the hash of the block before it. In Bitcoin, the block header includes the hash of the previous block's header, which is what the Bitcoin developer guide means by chaining the blocks together. The next example is hypothetical, with invented transactions, and uses the same method.

Illustrative example. Block 1 holds only the text "Alice pays Bob 5 coins". Block 2 holds the full hash of Block 1, then " | Bob pays Carol 2 coins". Real blocks carry much more data, but the linking works the same way.

Version	Block 1 hash (first 16 characters)	Block 2 hash (first 16 characters)
Original (5 coins)	40a417b6c609285e	f89b9b1e08f10c72
Edited (6 coins)	87b27ad83c9c5e9b	5fb36d07453bf757

Editing the amount in Block 1 changes its hash. Block 2 still holds the old Block 1 hash, so it no longer matches, and fixing it changes Block 2's own hash, which breaks Block 3, and so on to the end of the chain. The result: one edit forces a recalculation of every later block, and anyone holding an unedited copy sees the mismatch immediately.

To reproduce both tables, run this in Python 3:

import hashlib
def h(s): return hashlib.sha256(s.encode()).hexdigest()
for amount in ["5", "6"]:
    block1 = h("Alice pays Bob " + amount + " coins")
    block2 = h(block1 + " | Bob pays Carol 2 coins")
    print(block1, block2)

How nodes agree on a block

With thousands of independent copies, the network needs a rule for deciding which block is added next and which copy is correct. That rule is the consensus mechanism. NIST notes that networks open to anyone usually make publishing a block cost something, so that a malicious publisher cannot easily take over the system.

Two designs dominate. In proof of work, publishers compete to find a number that makes the block's hash meet a difficulty target, which the Bitcoin whitepaper describes as incrementing a nonce until the hash begins with a required number of zero bits. In proof of stake, validators lock up cryptocurrency as collateral and are chosen to propose blocks. Ethereum's documentation says that in each 12-second slot one validator is randomly selected to propose a block, other validators re-execute its transactions to check them, and validators that act dishonestly can have their stake destroyed, per the proof-of-stake page.

When two valid blocks appear at nearly the same time, the Bitcoin whitepaper's rule is that nodes work on the first one they received but keep the other branch, and the tie breaks when one branch becomes longer. The network then follows the longest chain, meaning the chain with the most proof-of-work behind it.

Why rewriting history is costly

Rewriting an old block means redoing the work for that block and all later ones, then overtaking the honest chain. The whitepaper states that the attacker would have to redo the proof-of-work of the block and all blocks after it, and that the chance of a slower attacker catching up falls exponentially with each later block. Even a successful rewrite has limits: the paper notes that honest nodes reject invalid transactions, so an attacker cannot create coins from nothing or take money that was never theirs, only try to reverse a payment of their own.

The paper gives a formula for the probability that an attacker with a share q of the block-producing power ever catches up once the honest chain is z blocks ahead. The formula assumes the attacker has less than half the power; with half or more, catching up is only a matter of time.

Blocks behind (z)	Attacker with 10% of power (q = 0.1)	Attacker with 30% of power (q = 0.3)
0	100%	100%
5	0.09%	17.7%
24	far below 0.01%	0.08%

These values were recomputed from the paper's formula. The z = 5 results match the paper's printed table (0.0009137 for q = 0.1 and 0.1773523 for q = 0.3). The paper does not print z = 24, but it lists 24 as the number of blocks needed to push a 30% attacker below 0.1%; the recomputed value is 0.0008106. A larger share of power makes an attack far more plausible; NIST describes a "51% attack" as one where an attacker holds more than half the resources. And the deeper a transaction is buried, the safer it is, which is why NIST says most users wait several block creations before treating a transaction as settled.

NIST also stresses that "immutable" is not strictly true. The most recent blocks can be replaced by a longer competing chain, and a permissioned network's owner or consortium can replace any number of blocks through legitimate means. The accurate summary is tamper evident and costly to alter, not impossible to alter.

Public and permissioned chains

NIST sorts blockchains by who is allowed to publish blocks. The comparison below follows its categories: it likens a permissioned network to a controlled corporate intranet and a permissionless one to the public internet.

Feature	Permissionless (public)	Permissioned
Who can publish blocks	Anyone	Only authorized participants
Who can read	Anyone	Anyone or only authorized users, set by the operator
Consensus cost	Usually requires spending or locking up resources	Often does not; usually faster and less computationally expensive
Whom users must trust	The rules and the spread of participants	The owner or consortium that controls membership

Whoever controls membership on a permissioned chain also controls who can change it, so its main benefit is shared record-keeping among organizations that do not fully trust each other, not removal of a central party.

What a blockchain is not good at

Speed and capacity

Every node re-checks and stores what the network records, so capacity is deliberately limited. Ethereum caps each block's total computation, and its documentation explains the cap exists to keep blocks from growing so large that ordinary full nodes could no longer keep up, which would centralize the network. Blocks arrive on a schedule: Ethereum aims for one per 12-second slot, which is at most 7,200 per day if no slot is missed, while Bitcoin targets two weeks (1,209,600 seconds) for every 2,016 blocks, about 600 seconds or 10 minutes per block and 144 per day on average. Resource costs show up on the storage side too: NIST reported in 2018 that Bitcoin's blockchain data was over 175 gigabytes and growing. Accenture's 2022 review of ASX's ledger-based settlement design, covered below, named higher latency from distributed systems as one underlying driver of its problems.

Privacy

Public chains make transactions visible to everyone. The Bitcoin whitepaper's privacy section explains that the public can see someone sending an amount to someone else but, if public keys stay anonymous, without information linking it to a person, and it warns that linking is still possible and can reveal other transactions if an owner is identified. NIST calls such networks pseudonymous. Pseudonymous is not private.

Facts that live off the chain

A blockchain can check that a transaction follows its rules. It cannot check that a sensor was working or a person told the truth. NIST calls this the "oracle problem" and gives an example: if a cryptocurrency payment buys a physical item, nothing inside the network can show whether the shipment took place without outside input. Garbage in remains garbage on the chain, permanently recorded.

Real uses, with dates and outcomes

Cryptocurrencies are the original use: a ledger of who owns what, kept without a bank. Outside them, uses are mostly record-keeping between organizations, and documented results are mixed. Two dated examples:

TradeLens (shipping). IBM and GTD Solution, a division of A.P. Moller - Maersk, jointly developed a blockchain-enabled shipping platform, announced in 2018. On 29 November 2022 they announced they would discontinue it, with the platform expected to go offline by the end of the first quarter of 2023. Their stated reason: they had developed a viable platform but did not achieve the full global industry collaboration needed for commercial viability. The shortfall was adoption, not the technology.
ASX CHESS replacement (stock settlement). The Australian Securities Exchange began replacing its clearing and settlement system with a distributed-ledger design. After an independent Accenture review, published in November 2022, found significant challenges with the solution design, ASX reassessed it, and on 20 November 2023 it said it would proceed with a product-based solution from Tata Consultancy Services, while remaining able to connect to distributed-ledger technology later.

Neither shows that a blockchain cannot work in finance. A shared ledger helps only when participants adopt it and the design meets the system's requirements. Treat any claim that a blockchain produced a specific percentage improvement as unproven unless it cites a published report.

Frequently asked questions

Is a blockchain the same thing as Bitcoin?

No. Bitcoin is one application built on a blockchain. NIST notes that the technology is often viewed as bound to Bitcoin or cryptocurrency in general, but that it is available for a broader variety of applications.

Does every blockchain need a cryptocurrency?

No. NIST says networks open to everyone usually reward block publishers with a native cryptocurrency, but permissioned networks identify their members and often do not need resource-costly consensus. Such networks can still use shared, replicated storage.

What is a node, and can anyone run one?

A node is a computer that keeps a copy of the ledger and checks new blocks. On a permissionless network anyone can take part, though a new full node has to download most or all of the chain's data, which NIST says uses a lot of network bandwidth. On a permissioned network, membership is controlled by the operator.

Does a blockchain have to use a lot of electricity?

It depends on the consensus method. NIST describes proof of work as resource intensive and says many permissioned systems use other mechanisms. Ethereum moved to proof of stake in September 2022, and the Crypto Carbon Ratings Institute estimates, per ethereum.org, that this cut its annualized electricity use by more than 99.988%, a bottom-up estimate rather than a measurement.

How is this different from an ordinary database?

An ordinary database has an administrator who can edit or delete records. A blockchain spreads that control across participants and links entries by hash, so edits are detectable. NIST's caveat applies: on a permissioned chain, the owner or consortium can still replace blocks if it chooses.

A test for any blockchain claim

Ask who is allowed to publish blocks. If it is a single company or a small group, the trust sits with them.
Ask what the chain actually verifies. If the claim depends on sensors, paperwork or people, the chain only preserves what was entered.
Ask whether the claim cites a dated, published source. Outcomes such as "cut costs by 40%" need a report behind them.
Look for the cost of a change: how many blocks deep an entry must be, and what it would take to outpace the network.
Check whether the use needs shared record-keeping among parties who do not trust each other. If one party can keep the records, an ordinary database may serve better.