Finality in blockchain has two distinct flavors, and the difference between them matters more than most explanations suggest.
The question "is this transaction final?" sounds simple. But the honest answer depends on which system you're asking about — and what you actually mean by "final."
Probabilistic finality means a transaction becomes increasingly difficult to reverse over time, but never reaches a point of absolute impossibility. Each new block added on top makes undoing it more computationally expensive. The guarantee is statistical, not absolute.
Deterministic finality means a transaction, once confirmed by a qualified majority of validators, cannot be reversed. Not unlikely to be reversed — cannot. The protocol rules it out by design.
How Probabilistic Finality Works
Bitcoin is the canonical example. When a transaction is included in a block, miners immediately begin building the next block on top of it. Anyone wanting to reverse the transaction would need to produce an alternative chain — one where that block never existed — and outpace the honest network long enough for the rest of the network to adopt the rewritten version.
The math is cumulative. After one confirmation, an attacker would need to redo the work for that block and race to produce new blocks faster than the honest majority. After six confirmations, they'd need to redo six blocks of proof-of-work and still win an ongoing race. Each additional block raises the required work, and the probability of a successful reversal falls toward zero without ever actually reaching it.
Bitcoin's whitepaper treats this explicitly. Satoshi modeled the probability of a successful double-spend attack as a function of confirmation depth and attacker hash rate. With 10% of network hash rate, an attacker has roughly a 0.1% chance of undoing a six-confirmation transaction. With 50% — what's called a 51% attack — even that becomes a calculation rather than a certainty.
Six confirmations became the de facto standard for exchanges and payment processors, but there's nothing magic about that number. It's a community convention reflecting a risk tolerance, not a protocol rule. High-value settlements sometimes wait for more.
How Deterministic Finality Works
BFT-based (Byzantine Fault Tolerant) consensus systems take a different approach. Instead of statistical guarantees from accumulated work, they achieve finality through explicit agreement among a known set of validators.
The mechanism: validators propose and vote on blocks. When a supermajority — typically two-thirds of validators by stake — sign off on a block, it's finalized. The protocol rules out reversal by construction: any chain that contradicts a finalized block would require at least one-third of validators to behave dishonestly, which the protocol detects and punishes by slashing their stake.
Cosmos uses Tendermint consensus, which works this way. Once a block is finalized in Tendermint, it's actually final — not probably final, not final for practical purposes. Light clients can trust it without downloading the full chain history.
The trade-off is structural. Deterministic finality requires a known validator set that can coordinate explicit votes. This works well for delegated systems with hundreds or a few thousand validators. It doesn't scale to Bitcoin's permissionless model, where anyone can mine with any hardware and no validator registry exists.
Ethereum's Hybrid
Ethereum after the Merge sits between these two models. The word "finality" gets used loosely here, so the details are worth being precise about.
Ethereum's consensus combines two mechanisms: LMD-GHOST (a fork choice rule tracking which chain has the most validator attestations) and Casper FFG (a finality gadget that periodically locks in checkpoints). The combination is called Gasper.
On a short time scale — slot by slot, every 12 seconds — Ethereum's confirmations are probabilistic. Validators attest to the head of the chain, and those attestations make reversal progressively harder. This looks similar to Bitcoin, though the mechanism is different.
Every two epochs — roughly 12.8 minutes — Casper FFG runs a finality checkpoint. If two consecutive epochs both achieve supermajority attestation (two-thirds of staked ETH), the earlier epoch is finalized. From that point, the chain cannot be reorganized without slashing at least one-third of all staked ETH. At current stake levels, that's tens of billions of dollars worth of collateral.
So Ethereum has probabilistic confirmation in the short term and deterministic finality every ~13 minutes. The distinction matters for applications that can wait vs those that need to act on soft confirmations.
Single Slot Finality (SSF) is the ongoing research direction attempting to compress this to a single 12-second slot — achieving deterministic finality as fast as block production. That's technically difficult at Ethereum's validator scale and doesn't have a committed deployment timeline as of mid-2026.
Where Constraints Live
Probabilistic finality's constraint is the honest majority assumption. Bitcoin assumes no entity controls more than 50% of hash rate. If that breaks, deep reorgs become possible. The assumption has held for Bitcoin since 2009, but smaller proof-of-work chains have been successfully attacked at depth.
Deterministic finality's constraint is the validator set. Systems assume fewer than one-third of validators by stake are acting maliciously. If that threshold is crossed through coordinated behavior — not honest mistakes, but active collusion — the finality guarantee fails. Slashing makes this economically irrational, not technically impossible.
What Would Confirm and Invalidate This
Confirmation signals: Ethereum SSF research producing a viable spec with a committed upgrade timeline; BFT chains sustaining years of operation without finality failures; exchange deposit windows shortening as deterministic finality speeds improve.
Invalidation signals: A successful deep reorg of a finalized Ethereum epoch (would require catastrophic validator collusion); a 51% attack on Bitcoin reversing transactions at depth greater than six blocks; SSF proving technically intractable, locking Ethereum's deterministic finality at its current ~13-minute cadence.
Timing
Now: Ethereum's hybrid model is live — probabilistic short-term, deterministic finality every ~13 minutes. Bitcoin has operated on probabilistic finality since 2009. BFT chains like Cosmos offer fast deterministic finality with its associated validator-set constraints.
Next: Ethereum SSF research is active but without a committed deployment timeline.
Later: Protocol-level innovations in finality speed and validator scaling remain open research problems without firm dates.
Boundary
This covers the distinction between probabilistic and deterministic finality at the protocol level. It doesn't address how applications should set confirmation thresholds for specific risk tolerances, the economics of 51% attacks in detail, or Ethereum's fork choice rule beyond what's relevant here.
The two models represent different design trade-offs, not a hierarchy where one is strictly better. Probabilistic finality works for permissionless environments where validator sets can't be enumerated. Deterministic finality works where validator coordination is acceptable. Most serious systems end up somewhere in between.
