A blockchain has no board and no arbitration. When participants disagree about the rules and neither side yields, the only available resolution is for both to continue separately, and the chain becomes two chains with a shared history and different futures.
Every participant runs software that enforces a set of rules about which blocks are valid. As long as everybody runs compatible rules, everybody agrees on the same chain. A fork happens when some participants start enforcing different rules, at which point they stop accepting blocks that the others produce, and the shared history stops being shared.
That framing is worth holding because it removes the drama. There is no vote, no authority, and nothing to appeal to. There is a set of rules encoded in software, a set of people choosing which software to run, and an outcome determined by who runs what. Everything else is persuasion aimed at that choice.
It also explains why forks concentrate around a particular kind of question. Disputes about a parameter can usually be settled by discussion because the sides are not far apart. Disputes about what the system is for, who it serves, or whether a past event should be reversed, cannot be split down the middle, and those are the ones that end in a split.
There is nobody to appeal to. A fork is what a disagreement looks like in a system where the only decision procedure is which software people choose to run.
A soft fork tightens the rules: blocks valid under the new rules are also valid under the old ones. Participants who do not upgrade continue to accept everything the upgraded participants produce, so the network stays together even though not everybody has changed. It is a compatible change and it is how most upgrades happen.
A hard fork loosens or changes the rules: blocks valid under the new rules are rejected by the old software. Participants who do not upgrade stop accepting the new blocks, which means everybody must change or the network splits. There is no version where a hard fork is optional for some participants and invisible to others.
The practical distinction for a holder is simply whether they need to do anything. A soft fork requires nothing from most participants. A hard fork requires everybody who runs software to update it, and everybody who holds through it to understand that they may end up holding two things.
At the moment of a split, both chains contain the entire history up to that point, which means every balance exists on both. If you held an amount before the fork, you hold that amount on each chain afterwards, controlled by the same keys, because the keys are part of the shared history too.
This is the part that gets described as free money and it is more complicated than that. You hold two assets where you held one, and their combined value is not automatically greater than what you had before. Frequently the original falls by roughly what the new one is worth, because the market is dividing an existing valuation rather than creating one.
Access is the other complication. Holding the balance on both chains and being able to move it on both are different things, and the second depends on having software that supports the new chain, and on that chain having somewhere to trade. A balance you cannot move or sell is a balance in the abstract.
Immediately after a split, the two chains accept identical transaction formats, because they were the same software a moment earlier. That means a transaction you broadcast on one chain can frequently be rebroadcast by anybody on the other and will be valid there too, moving the corresponding balance without your intention.
The practical consequence is specific and it has cost people their holdings. Sending your balance on one chain to an exchange, in order to sell it, can result in the same movement occurring on the other chain, sending the asset you intended to keep to the same address. The transaction is valid, it is signed by you, and nothing about it is reversible.
Responsible forks include replay protection, a deliberate change making transactions on one chain invalid on the other, and it removes the problem entirely. Whether a given split has it is publicly stated in advance and it is the single most important technical fact for anybody intending to transact around one. Where it is absent, the safe action is to transact on neither chain until it is added or until you have split your holdings carefully.
Without replay protection, spending on one chain can spend on the other. It has happened, it is not reversible, and whether protection exists is stated in advance.
The great majority of forks produce a chain that trades briefly and then fades, and the reason is that a chain needs several things simultaneously and a split usually only takes one of them. It needs participants securing it, developers maintaining it, applications built on it, venues willing to list it, and holders who want it. A minority faction typically starts with the last two and not the first three.
The security problem is the sharpest. A chain that inherits a small fraction of the original's mining or staking is substantially cheaper to attack, and several minority chains have suffered exactly that. The economic argument is straightforward: the cost of attacking a chain is proportional to the resources securing it, and a split divides those resources without dividing the value at stake proportionally.
What survives, when anything does, is a chain that attracted a genuine constituency with a genuinely different view of what the system should be. That is rare because most disagreements are about parameters rather than purposes, and a parameter disagreement does not sustain a separate ecosystem.
There is no registry and no authority, so the name goes to whoever the ecosystem treats as the continuation. In practice that is settled by a combination of which chain retains the majority of participants, which one the major venues list under the original ticker, and which one the developer community continues to work on.
The process is messy and it has been contentious every time. Both sides claim continuity, both produce arguments about which is the true version, and the resolution arrives through coordination rather than through a decision anybody made. The chain that ends up with the name is the one people carried on calling by it.
For a holder, the practical point is that the naming outcome determines a great deal about liquidity and support, and it is not predictable in advance from the technical merits. A chain can be technically preferable and lose the name, at which point it is a separate asset with a separate and much smaller market.
Whether you receive the new asset at all depends on where your holdings sit. Assets held under your own keys exist on both chains automatically. Assets held on a platform depend on that platform deciding to credit you, and there is no obligation for it to do so.
Most established venues announce their intention in advance and most do credit holders, because not doing so is a customer relations problem. But the decision is theirs, the timing is theirs, and there have been cases where a platform supported a fork late, partially, or not at all. The announcement, when it exists, is the only commitment.
There is also a suspension period. Deposits and withdrawals of the affected asset typically stop for a window around the split, which means capital on a platform is immobile at exactly the moment the market is deciding what the two chains are worth. Anybody planning to act around a fork should know that the window exists and how long it is expected to last.
Underneath the technical vocabulary, a fork resolves a question about authority: when there is no procedure for settling a disagreement, what decides. The answer these systems have arrived at is that nothing decides, and both answers proceed, and the market allocates value between them afterwards.
That is a genuinely unusual arrangement and it has a real virtue. Nobody can be forced onto rules they reject, because leaving is always available and costs only the network effect. It is the ultimate form of the exit option, and it is why these systems can change at all without a governing body.
The cost is equally real. The threat of a split makes controversial changes extremely difficult to adopt, which produces conservatism, which is a feature for a settlement layer and a limitation for a platform trying to evolve. Chains have made different choices about how much of each they want, and those choices are visible in how often each one forks.
Four things, in order. Establish whether replay protection exists, because if it does not, the safe action is to transact on neither chain until you have understood how to separate your holdings. Find out whether the platforms holding any of your balance intend to credit the new asset and when, since that is a decision they announce rather than an entitlement you have.
Then decide in advance what you will do with the new asset if you receive it, because the decision is much harder afterwards when a price exists and is moving. And expect a suspension window on deposits and withdrawals around the date, which means capital you might want to move should be moved before rather than during.
None of this requires a view on which chain is correct. It is operational preparation for an event with a known date and a known set of mechanical consequences, and doing it in advance costs an hour while doing it during costs whatever the confusion costs.
Contentious splits were common in one period and are much less so now, for reasons worth understanding. The economic argument against them became clear: minority chains generally lose value, lose security, and lose developers, and the outcome has been consistent enough that fewer people believe a split will succeed.
The technical culture also changed. Upgrade mechanisms that avoid requiring everybody to change simultaneously, and a stronger preference for compatible changes, mean that more disagreements can be resolved without a split. And governance processes, whatever their limitations, at least provide a forum where a disagreement can be aired before it becomes a fork.
What has not changed is the underlying possibility, and that is the part that matters. The option to split is what makes the systems voluntary, and its existence constrains what anybody can impose even when it is never exercised. A fork that does not happen still did work.
A disagreement about the rules that neither side drops. Participants run software enforcing rules; when some start enforcing different ones, they stop accepting each other's blocks and the chain becomes two. There is no vote and no authority, only which software people choose to run.
A soft fork tightens the rules, so blocks valid under the new rules are also valid under the old ones and participants who do not upgrade stay compatible. A hard fork changes them in a way the old software rejects, so everybody must upgrade or the network splits.
Both chains contain the entire history up to the split, so your balance exists on both, controlled by the same keys. It is not automatically more value: the original frequently falls by roughly what the new asset is worth, because the market divides an existing valuation rather than creating one.
Immediately after a split both chains accept identical transaction formats, so a transaction you broadcast on one can be rebroadcast by anybody on the other and will be valid there too. Sending to an exchange to sell one asset can move the other to the same address, signed by you and not reversible.
A chain needs participants securing it, developers maintaining it, applications, venues and holders, and a minority faction typically starts with only the last two. The security problem is sharpest: a chain inheriting a small fraction of the original's mining or staking is substantially cheaper to attack.
Nobody, formally. It is settled by which chain retains the majority of participants, which one major venues list under the original ticker, and which one the developer community continues to work on. A chain can be technically preferable and lose the name, at which point it has a much smaller market.
Assets under your own keys exist on both chains automatically. Assets on a platform depend on that platform choosing to credit you, which most established ones do and none are obliged to. Expect a suspension of deposits and withdrawals around the date, so move anything you need to move beforehand.
The economic argument against them became clear as minority chains consistently lost value, security and developers. Upgrade mechanisms that avoid simultaneous changes also improved, and governance forums let disagreements be aired first. The possibility remains, and its existence constrains what anybody can impose even when unused.