Probabilistic finality is the property of some blockchains under which a confirmed transaction never becomes absolutely irreversible, but the chance of it being reversed falls quickly as more blocks are added on top of it. It is the finality model of the Bitcoin Blockchain and other networks secured by proof of work, where nodes follow the chain with the most accumulated work. Users therefore wait for several confirmations before treating a payment as settled; six blocks, roughly an hour on Bitcoin, is a common convention.
It contrasts with deterministic finality, used by many proof of stake and Byzantine fault tolerant systems, where a block is final once a defined voting step completes. Both differ from the legal payment finality of central bank systems, which is fixed by law and system rules rather than by computation.
Probabilistic finality in plain terms
Think of fresh concrete. Straight after it is poured, someone could still smooth it over and leave no trace of a footprint. An hour later it is much harder to change, and a day later it would take a jackhammer. The footprint was never guaranteed to be permanent at any single moment; it simply became harder and harder to erase.
A Bitcoin payment hardens in the same gradual way. When it first appears in a block, a rival version of recent history could still replace that block. Each new block stacked on top makes a rewrite more expensive, because anyone trying it would have to redo all that work faster than the rest of the network. After enough blocks, reversal becomes so costly and unlikely that people treat the payment as done.
How probabilistic finality works
Miners compete to add blocks, and two can occasionally find a valid block at almost the same moment. For a short time the network holds two competing tips. Under Nakamoto consensus, nodes build on whichever branch accumulates more work, and the losing block is dropped. Transactions in it return to the pool and are usually included again later. Replacing recent blocks this way is called a chain reorganisation, or reorg.
Most reorgs are one block deep and harmless. The danger is deliberate: an attacker who controls a large share of mining power can secretly build a longer branch that omits a payment, then release it after receiving goods, achieving a double spend. That is why recipients count transaction confirmations and why the double-spend problem is central to proof-of-work security.
The attacker catch-up probability
Satoshi Nakamoto's 2008 white paper models the race as a gambler's ruin problem. When honest miners hold the majority of mining power, the chance that an attacker ever catches up from a given number of blocks behind is:
- is the attacker's share of total mining power.
- is the honest miners' share, equal to .
- is the number of blocks the attacker is behind, roughly the confirmations the recipient waited for.
- is the probability the attacker eventually overtakes the honest chain.
Worked example: an attacker with 10% of mining power faces honest miners with 90%. After six confirmations, , about two in a million. The paper's fuller calculation also allows for blocks the attacker mines while the recipient waits, which gives a higher figure; on that basis, a 10% attacker needs five blocks for the risk to fall below 0.1%, and a 30% attacker needs 24. If the attacker holds half or more of the mining power, the formula no longer applies: the probability of catching up is 1 and no number of confirmations is safe, which is the logic of a 51% attack.
Deterministic finality in proof of stake
The Ethereum Blockchain has used proof of stake since the Merge in September 2022. Time is divided into 12-second slots grouped into 32-slot epochs, and validators vote on checkpoints. Under the Casper FFG gadget, a checkpoint becomes finalised after two epochs, about 12.8 minutes. Reverting a finalised block would require validators controlling at least one-third of staked ether to break protocol rules, and their stake can be destroyed by slashing, so finality is economic rather than merely probable.
Networks built on Byzantine fault tolerance go further, finalising each block as soon as a supermajority of validators signs it, at the cost of halting if too many validators go offline.
Reorganisations in practice
In March 2013 an incompatibility between two versions of the Bitcoin software briefly split the chain into two forks until miners coordinated to return to the older rules. Smaller proof-of-work chains with little hash power have proved more vulnerable: Bitcoin Gold suffered double-spend attacks in 2018, and Ethereum Classic suffered several deep reorganisations from 51% attacks in August 2020. Exchanges respond by setting their own confirmation thresholds per network, often far higher for small chains.
For payments and settlement, probabilistic finality means the moment of completion is a risk judgement rather than a fixed point. Protections in laws such as the Settlement Finality Directive apply to designated systems, not to public blockchains, so firms using chains for settlement must manage the residual settlement risk themselves.