Glossary · Blockchain basics

Fork

Published 28 Apr 2025 · Updated 21 Oct 2026 · 15 min read

A fork in blockchain is a split of one chain's history, rules or code into two versions that share everything before the split. Most forks involve two competing blocks, with the network choosing one within a block or two. Others change rules or create a separate chain and coin. Developers also fork code.

  • Most forks are temporary and settle within one or two blocks.
  • Soft and hard describe compatibility, while producer adoption shapes whether a fork splits the chain.
  • Coins from five chain splits since 2016 are each worth under 7% of their parent's value in October 2026.
Six meanings of a forkTemporary forks between competing blocksSoft forks and hard forksHow networks activate new rulesChain splits that keep separate historiesAccidental splits caused by software bugsCode forks of blockchains and appsStaying safe around a forkPlanning fork engineering workFork FAQ

Six meanings of a fork

A fork can mean six different things, from competing blocks to copied software. This overview separates them before examining activation, security and examples. Dedicated entries cover hard forks, soft forks and ommer blocks. First identify whether someone means a branch, a rule change or a codebase.

MeaningWhat divergesHow long it lastsExample
Temporary forkWhat divergesTwo valid blocks compete at the same heightHow long it lastsThe fork usually resolves within one or two blocksExampleBitcoin had competing blocks at height 966,500 on 11 September 2026
Soft forkWhat divergesNew nodes reject blocks that older nodes acceptHow long it lastsNo split occurs if most block producers enforce itExampleSegWit activated at block 481,824 on 24 August 2017
Hard forkWhat divergesOld nodes reject blocks accepted under changed rulesHow long it lastsNodes that do not upgrade are left behind for goodExampleThe Byzantium fork activated on 16 October 2017
Chain splitWhat divergesTwo groups follow different rules and separate coinsHow long it lastsBoth chains can continue indefinitely with supportExampleBitcoin Cash began at block 478,559 on 1 August 2017
Bug splitWhat divergesClient versions disagree about which blocks are validHow long it lastsRecovery usually takes hours rather than blocksExampleOlder Geth versions split off on 11 November 2020
Code forkWhat divergesCopied source code becomes a separate projectHow long it lastsThe forked project can continue independently of its sourceExampleLitecoin began with its own genesis block in 2011

Temporary forks between competing blocks

Two miners can produce valid blocks at the same block height almost simultaneously. Different nodes see different blocks first and initially accept them. The temporary fork usually ends when the next block strengthens one branch. The losing block leaves the accepted history. A reorg replaces previously accepted blocks, so a payment's first appearance does not guarantee permanence.

Figure 1 · Bitcoin stale blocks, 2011 to 2026Bitcoin's temporary fork historyMove through 2011 to 2026 to compare yearly rates, block counts and the ten latest recorded stale blocks.
2026
Stale blocks per 1,000 blocks
Stale blocks in the data53Includes 8 BIP-110 blocks
Blocks on the main chain42,171Mined from 1 January to 21 October
Stale per 1,000 blocks1.26Highest: 15.64 in 2012
Latest stale blocks in the data
#971,3452026-10-14, 03:31 UTC#969,8882026-10-04, 19:06 UTC#969,2982026-09-30, 11:14 UTC#967,1432026-09-15, 15:35 UTC#966,5002026-09-11, 11:35 UTC#964,1812026-08-26, 18:37 UTC#963,8532026-08-24, 11:25 UTC#963,8282026-08-24, 07:05 UTC#963,7872026-08-24, 00:41 UTC#963,1052026-08-18, 22:53 UTC

Through 21 October 2026, eight of 53 recorded blocks belong to BIP-110. Excluding them lowers the rate from 1.26 to 1.07.

The public bitcoin-data stale-blocks dataset was downloaded on 21 October 2026. Its crowd-sourced counts are lower bounds.

Bitcoin's fork choice rule follows the most cumulative work, not simply the most blocks. A losing block is a stale block. An orphan block can instead mean one whose parent is unknown. Faster relay helped lower rates after 2015. Since 2018, observed forks reach two blocks at most, excluding Bitcoin Cash and BIP-110, in data downloaded 21 October 2026. Coinbase outputs wait 100 blocks before spending because their blocks can become stale.

Ethereum's proof-of-work ommer blocks could receive partial rewards. Under proof of stake, LMD-GHOST weights branches by attestations, while Casper FFG adds finality. A 4 September 2025 study measured reorged slots at 0.1224% before Pectra and 0.1121% afterward. Etherscan's 17 to 21 October 2026 records show about six one-block reorgs daily. A seven-block beacon reorg on 25 May 2022 preserved finality. Solana also discards branches that do not finalize.

Soft forks and hard forks

A soft fork upgrade tightens validation rules, so blocks following them still satisfy older software. A hard fork permits blocks that old software rejects. These labels describe compatibility, not the outcome. A lasting split needs block producers to keep both branches alive. Adoption determines whether the network converges or continues along separate paths.

Figure 2 · A simplified rule changeCompatibility and producer adoptionChoose a fork type and adjust producer adoption from 10% to 100%. Compare the branches and the chains each node follows.
Change
60%
One chainOne chain. Upgraded and old nodes follow chain A.
Chain A, valid under both rule setsFollowed by everyone1234567891011Old nodes accept these blocks, because the new rules only forbid things the old rules allowed.
Chain B, a block that breaks the new rulesOrphaned1234Upgraded producers never build on it, and chain A grows faster, so old nodes drop it too.
Chains after the fork1Short forks can still happen
Upgraded nodes followChain ANew rules enforced
Old nodes followChain AThey see nothing invalid

With most producers enforcing this model's soft fork, one chain remains. Any rule-breaking block is orphaned.

This simplified model uses eight new blocks, one rule change and no network delay.

Bitcoin favors soft forks because older nodes can follow the resulting chain without enforcing every new restriction. To them, those restrictions resemble policy rather than consensus. Ethereum instead uses a hard fork for every network upgrade. When clients upgrade together, they can still leave only one chain. The distinction concerns what each node will accept, not a promise of a new asset.

Enforcement can fail to match signaling. P2SH left 91 invalid blocks and 22 descendants between April and July 2012. BIP66 produced six-block and three-block forks in July 2015 because miners extended blocks without full validation. The BIP-110 fork required bit 4 signaling from block 961,632 on 8 August 2026, despite limited support. Its branch found eight blocks through 28 August 2026. The proposal closed on 9 August 2026 after mining stalled.

How networks activate new rules

A proposed fork needs a shared activation point. Bitcoin has activated upgrades using dates, block heights and miner signaling. Ethereum coordinates client releases around scheduled triggers. Signaling does not prove miners validate new rules. A user-activated soft fork (UASF) instead depends on users enforcing them.

MethodWhere and whenWhat triggers itExample
Flag dayWhere and whenBitcoin used date and height triggers for forks from 2009What triggers itNodes enforce the upgrade at a preset height or dateExampleBitcoin enforced nLockTime at block 31,000 on 22 December 2009
Miner supermajorityWhere and whenBitcoin used block versions for BIP-34's 2013 upgradeWhat triggers itNew-version rules bind at 750 of 1,000, and old versions fail at 950ExampleBIP-34 activated at block 227,931 on 25 March 2013
Version bitsWhere and whenBitcoin used version bits (BIP-9) for CSV's July 2016 upgradeWhat triggers itAfter 1,916 of 2,016 blocks signal, activation follows one period laterExampleCSV activated at block 419,328 on 4 July 2016
User-activated soft forkWhere and whenBitcoin's BIP-148 fork proposal targeted 1 August 2017What triggers itUsers enforce mandatory signaling from a preset dateExampleBIP-91 preceded it at block 477,120 on 23 July 2017
Speedy TrialWhere and whenBitcoin activated the Taproot upgrade in November 2021What triggers it1,815 of 2,016 must signal, with activation waiting for a minimum heightExampleTaproot activated at block 709,632 on 14 November 2021
Scheduled hard forkWhere and whenEthereum has used this method for every network upgrade since Homestead in March 2016What triggers itClients apply changed rules at an agreed activation pointExampleFusaka activated at block 23,935,694 on 3 December 2025

Ethereum moved from block-height triggers to total difficulty for The Merge on 15 September 2022, then timestamps for upgrades with Shanghai in April 2023. On 21 October 2026, its 19 execution-layer activations include only two that produced separate coins. A fork ID (EIP-2124) helps incompatible peers disconnect. Glamsterdam reached the Sepolia testnet on 6 October 2026. Hoodi is scheduled for 26 October 2026. The Ethereum Foundation's 16 October 2026 update leaves mainnet's date undecided.

Not every protocol change needs a chain fork. Canton's logical synchronizer upgrade changes a synchronizer's protocol version while participants remain attached to the same logical synchronizer. Super validators prepare the upgrade deployment and schedule ahead of time. The process allows very limited network downtime. After 30 days, they remove the old physical synchronizer deployment.

Chain splits that keep separate histories

A lasting chain split leaves two communities, two rule sets and two coins. Both inherit history and balances through their last shared block. The DAO response made an irregular state change, not an EVM rule change, at block 1,920,000 on 20 July 2016. The minority coins compared here retain much smaller relative values on 21 October 2026 than in their early snapshots.

Figure 3 · Six chain splits, 2016 to 2026What became of minority chainsSelect a fork to compare its cause, replay safeguards and relative value. Follow the timeline or compare all six cases.
Split
Ethereum ClassicContested hard fork20 July 2016 · Ethereum block 1,920,000An irregular state change moved the ether in The DAO and its child contracts to a refund contract. Miners who rejected it kept the old chain.Split fromEthereumReplay protection at launchNoneValue 11 days later15.1% of ETH market capValue on 21 October 20260.42% of ETH market cap
What happened next20 Jul 2016About 85% of miners follow the fork. Block 1,920,000 moves about 12 million ETH into a refund contract.Aug 2016Coinbase says replay attacks took about $40,000 of ETC from it.Jan 2017Ethereum Classic adds replay protection with its own chain ID, 61.Aug 2020Three 51% attacks in one month rewrite thousands of blocks, the largest over 7,000.
Minority chain value as a share of its parentChainSoon after21 Oct 2026Ethereum Classic15.1%0.42%Bitcoin Cash6.86%0.338%Bitcoin Gold3.92%0.00024%Bitcoin SV58.3%6.55%EthereumPoWAbout 0.4%0.0098%BIP-110No coinNo coin

Ethereum Classic continued without the DAO state change. Attacks reorganized thousands of its blocks in August 2020.

Values compare CoinMarketCap snapshot market caps with CoinGecko data on 21 October 2026. ETHW uses prices instead.

Replay protection prevents a signature intended for one chain from authorizing the same transaction on another. EIP-155 added a chain ID to signed data in November 2016, while old unprotected signatures remained valid. Bitcoin Cash requires SIGHASH_FORKID. Coinbase reported about $40,000 in ETC replay losses in August 2016. Ethereum Classic added chain ID 61 in January 2017. An ETHW bridge replay attack yielded 200 ETHW in September 2022 because a contract read a stored chain ID. BIP-110's August 2026 branch had no such protection.

Copied balances do not make the two assets interchangeable. Exchanges and apps need to distinguish networks before crediting payments, and service interruptions can include paused withdrawals. Smaller mining communities can leave branches exposed to reorganizations. Ethereum Classic suffered three 51% attacks in August 2020, reorganizing 3,693 blocks, about 4,000 blocks and more than 7,000 blocks. Shared history does not guarantee shared security afterward.

Accidental splits caused by software bugs

A consensus bug can make clients calculate different valid histories, even when nobody intends a split. These accidental forks were repaired with patches, downgrades or coordinated software updates. They differ from ordinary block races because another block alone may not restore agreement.

  1. August 2010

    Bitcoin's value overflow

    Block 74,638 created about 184 billion bitcoin through integer overflow. A patched client arrived within five hours, and its chain prevailed at block 74,691.

  2. March 2013

    Database limits divide Bitcoin

    At block 225,430, Bitcoin 0.7 rejected a block accepted by 0.8 because of Berkeley DB limits. BTCGuild and Slush downgraded to restore agreement.

  3. July 2015

    Signaling without full validation

    BIP66 reached 95% at block 363,725. Miners practicing SPV mining extended invalid block 363,731 without full validation, forming a 6-block fork on 4 July. A separate 3-block fork followed on 5 July 2015.

  4. November 2016

    Geth versions disagree

    A consensus bug in Geth 1.4.19 and 1.5.2 split the network at block 2,686,351. Users needed Geth 1.5.3 to recover.

  5. November 2020

    Older Geth disrupts services

    Older Geth versions diverged at block 11,234,873 on 11 November 2020. Infura went offline, and exchanges including Binance paused ETH and ERC-20 withdrawals.

  6. August 2021

    Memory corruption changes state

    A crafted transaction exploited older Geth versions at block 13,107,518, producing a different state root. Geth 1.10.8 had shipped the fix three days earlier.

Code forks of blockchains and apps

A software fork need not divide an existing transaction history. A code fork can start a new network or adapt an application, including a decentralized exchange. The new project shares an implementation with its source, not necessarily its users, balances or consensus.

Bitcoin's code

Litecoin reused Bitcoin's code but began with its own genesis block in 2011. It uses 2.5-minute blocks and 840,000-block halvings. Dogecoin also adapts Bitcoin Core.

Geth's code

BNB Smart Chain and Polygon Bor forked Geth, sharing potential bugs. In October 2026, op-geth is end of life and will not implement hard forks from Glamsterdam onward.

Uniswap's code

SushiSwap forked Uniswap V2 in August 2020. Its September 2020 liquidity migration moved over $800 million from Uniswap, an event known as a vampire attack.

Licenses and reuse

Uniswap V2 uses GPL-3.0. V3's Business Source License became GPL-2.0 or later on 1 April 2023. V4 uses the Business Source License until 15 June 2027 at the latest.

DefiLlama lists 773 Uniswap V2 forks holding $2.276 billion on 21 October 2026. It also lists 52 Aave V3 forks holding $5.621 billion and 253 Uniswap V3 forks that day. A fork inherits useful features and potential bugs together. Independent development does not remove the need to track upstream fixes.

Staying safe around a fork

Checks around a fork depend on what is changing. A short reorg affects payment confidence, a rule change affects client compatibility, and a persistent branch affects transaction validity. Users and application teams can separate these risks before deciding which checks apply.

Wait for finality

Treat payments as settled only after finality or enough block confirmations. Ethereum's JSON-RPC API offers a finalized block tag for this check.

Check the chain ID

Verify the chain ID used in EIP-155 signatures. Contracts should read CHAINID at runtime rather than trust a value stored during deployment.

Track client releases

Track upgrade schedules, hard fork dates and client releases. Bitcoin Core warns when an invalid chain exceeds its accepted chain by more than six blocks.

Test copied mainnet state

A mainnet fork copies state locally without publishing transactions. Pin its block number so runs repeat. The PixelPlex smart contract cost guide of 12 February 2026 lists fork tests that simulate real mainnet conditions.

Planning fork engineering work

A team planning a fork needs more than a copied repository or an activation date. Work also includes maintaining compatibility, separating transaction domains and testing application behavior. We build blockchain software and plan its ongoing maintenance. Our Canton development work includes participant node management and planning for future protocol upgrades.

PixelPlex case icon

In our 2020 to 2021 Rio DeFi project, we built a Substrate-based fork, adapted the EVM pallet and forked the explorer. We also built a Uniswap-based exchange prototype. For Qtum's blockchain implementation, we implemented proof-of-stake consensus on a Bitcoin fork and integrated Ethereum and WASM virtual machines.

  1. Scope and licensing

    Identify the code being forked and its license terms. Separate implementation reuse from any decision to create a new network.

  2. Upstream compatibility

    Map dependencies on upstream clients and scheduled upgrades. A forked client still needs compatible changes to keep following its target chain.

  3. Transaction separation

    Define chain identity and replay protection before supporting parallel histories. Review transaction signatures and how contracts obtain the current chain ID.

  4. Application settlement

    Specify how the application treats replaced blocks and finalized payments. Keep branch selection separate from the business decision to credit a payment.

  5. Pinned state testing

    Choose a pinned mainnet fork for reproducible tests. Check integration behavior before changing the software that connects to live networks.

Discuss your code and network compatibility needs

We can discuss your project context, existing dependencies and the engineering work involved in its next upgrade.

Discuss your project

Fork FAQ

What does a fork mean in crypto?

A fork is a point where a blockchain splits into two versions. Most often, two miners find blocks at the same height, and the network keeps one. It can also mean a soft or hard fork of the rules, a lasting chain split with its own coin, or copied code.

How do soft forks and hard forks differ?

A soft fork narrows what blocks are accepted, while a hard fork allows blocks that older software rejects. Compatibility is the distinction. Taproot activated on 14 November 2021 with older software still able to operate unchanged. Those older nodes do not enforce every new restriction.

Does every hard fork create a new coin?

A new coin is not an automatic outcome. It requires a continuing separate chain, not just incompatible rule changes. When participants adopt the same changes, one network continues. Ethereum's December 2025 and January 2026 Blob Parameter Only forks changed blob limits without creating another coin.

What happens to coins during a chain split?

Balances at the last shared block appear on both branches. Bitcoin Gold's 24 October 2017 snapshot copied balances at block 491,406 into its separate chain beginning at block 491,407. That duplication does not establish equal market value or prevent transaction replay.

What is a stale or orphan block?

A stale block is a valid block left outside the accepted chain after another branch wins. The word orphan is also used for a block with an unknown parent. Stale blocks are rare on Bitcoin, about 1 to 2 per 1,000 blocks in 2025 and 2026. On proof-of-work Ethereum, stale blocks included by later blocks were called ommer or uncle blocks.

What does replay protection do after a split?

Replay protection makes a transaction signed for one chain invalid on another. EIP-155 includes the chain ID in signed data, and Bitcoin Cash makes SIGHASH_FORKID mandatory. Without these safeguards, the same signed transaction can be valid on both chains, as on BIP-110's August 2026 branch.

Can anyone fork a blockchain or DeFi app?

Anyone can copy open-source code if its license allows it, creating a separate project without changing the original. Uniswap V2 uses GPL-3.0, while V3's Business Source License ended on 1 April 2023. V4 uses that license until 15 June 2027 at the latest. A running network changes only when its users and block producers adopt new rules.

Related terms

All definitions
Sources (35)

Get updates about blockchain, technologies and our company

We will process the personal data you provide in accordance with our Privacy policy. You can unsubscribe or change your preferences at any time by clicking the link in any email.

Follow us on social networks and don't miss the latest tech news

  • facebook
  • X
  • linkedin
  • instagram
Stay tuned and add value to your feed