Two words get used interchangeably in exchange interfaces and block explorers, and they describe different guarantees. Confirmation is a measure of probability; finality is a statement about what the protocol will refuse to undo.
Proof of work never promises finality
On a proof of work chain, a transaction included in a block can still be displaced if a competing branch grows longer. Nothing in the rules forbids it.
What protects the transaction is cost. Reversing it means redoing the work of every block built on top, faster than the rest of the network is extending the current chain.
Each additional block makes that more expensive, so confidence rises with depth without ever reaching certainty. This is why services quote a number of confirmations rather than a yes or no.
Proof of stake can make reversal punishable
Stake based systems can do something different, because validators sign explicit votes and those signatures are attributable. A validator who signs two conflicting histories has produced evidence against itself.
The protocol can then destroy part of that validator's deposit. Reversal stops being merely expensive and becomes self-incriminating.
Once enough stake has voted for a block, undoing it would require a large share of validators to accept losing their deposits. That threshold is what the word finality points at.
Finality arrives in stages
Even where finality exists it is not immediate. Votes accumulate over a defined window, and a block passes through justified and then finalised states as the window completes.
During that window the block behaves much like a confirmed proof of work block: very likely to stick, not yet guaranteed. The distinction only bites during network trouble.
If too much of the validator set is offline, finalisation can stall while the chain keeps producing blocks. The chain continues, but the guarantee is temporarily suspended.
Bridges and exchanges pick their own thresholds
Any service that credits an incoming deposit is making a bet about reversal. Crediting too early risks paying out against a transaction that later vanishes.
So each service chooses a depth it considers acceptable, weighing the value moved against the cost of a reorganisation. Larger deposits often wait longer than small ones.
Cross-chain bridges face the sharpest version of this, because a reversal on the source chain leaves assets released on the destination with nothing backing them.
Reorganisation depth is the practical measure
What matters in operation is not the theory but the observed depth of reorganisations on a given chain. Shallow single block reorganisations happen routinely on some networks and almost never on others.
Operators tune their waiting periods to that history rather than to the whitepaper. The protocol sets the ceiling; observed behaviour sets the working rule.
A chain with fast nominal finality but frequent stalls can end up slower in practice than one with no finality guarantee at all.