A blockchain is a shared record maintained by participants who follow identical rules. A fork occurs when that agreement breaks and two versions of the record continue in parallel.
Consensus depends on shared validity rules
Each node independently checks whether a block satisfies the protocol's rules before accepting it. Nodes running the same software reach the same conclusion about the same block.
The chain everyone follows is selected by a rule, commonly the version with the most accumulated work or stake weight. Nodes converge because they apply identical criteria.
Change the criteria on some nodes but not others, and those nodes can accept different blocks. The single record becomes two.
Temporary forks are ordinary operation
Two miners can produce valid blocks at nearly the same moment, and different parts of the network see them in different order. Both chains extend briefly.
The next block extends one branch, which then carries more accumulated work, and nodes abandon the shorter branch. Transactions in the discarded block return to the pending pool.
This resolves without intervention and is the reason participants wait for additional confirmations before treating a transaction as settled.
A soft fork tightens rules that old nodes still accept
A soft fork narrows what counts as valid. Blocks produced under the new stricter rules remain valid to nodes running older software, because they satisfy the older, looser criteria.
Old nodes continue to follow the chain without upgrading, though they cannot fully verify the new constraints. Adoption by a majority of block producers is enough to make the change effective.
Backward compatibility is why soft forks are the usual route for incremental protocol changes.
A hard fork creates blocks old software rejects
A hard fork loosens or alters rules so that new blocks fail validation on old software. Nodes that do not upgrade cannot follow the new chain at all.
If every participant upgrades, the old chain simply stops being extended and the network continues as one. Coordination, not technology, decides that outcome.
If a meaningful group continues running the old rules, both chains persist. Each has its own history before the split and separate histories afterward.
A permanent split duplicates balances and creates confusion
Because the shared history is identical up to the fork point, every balance exists on both chains. Holders end up with units on each, though the two are separate assets with separate markets.
Transactions can be vulnerable to replay across chains unless the protocols add distinguishing markers, and exchanges typically suspend deposits and withdrawals while the situation stabilizes.
The mechanics are governance made visible: a fork is what disagreement looks like in a system with no central authority to settle it.