Skip to content

Guide · Protocol

Ethereum finality and reorgs: what “settled” actually means

What Ethereum finality actually guarantees, why shallow reorgs are normal at the chain head, and how justified checkpoints differ from finalized ones. Not financial advice.

By Etherelysium Editorial TeamUpdated 18 November 20267 min read

Ethereum finality and reorgs: what “settled” actually means
Share X LinkedIn

On Ethereum, “included in a block” and “settled” are not the same claim. A transaction can sit in a block that later gets reorganized out of the chain entirely, even after several more blocks have been built on top of it. “Settled,” in the strict sense this site uses, means the block has been finalized under Ethereum’s own consensus rules — a state that comes with a concrete economic guarantee, not just a low probability of reversal.

Slots, epochs, and the finality clock

Ethereum’s proof-of-stake chain is built from slots and epochs. A slot is a 12-second window in which one randomly selected validator may propose a block; an epoch is 32 consecutive slots, so roughly 6.4 minutes. Validators are split into committees so that every active validator casts one attestation — a vote for what it believes is the correct head of the chain — once per epoch, not once per slot. The chain’s fork-choice rule, LMD-GHOST, follows whichever branch has accumulated the most attestation weight, which is how the network converges on a single view of “the chain” even when a few validators briefly disagree.

Finality itself comes from a separate layer called Casper FFG (Friendly Finality Gadget), which operates on epoch boundaries rather than individual blocks. The first slot of each epoch is a checkpoint. Validators vote on pairs of checkpoints, and when a checkpoint pair receives votes backed by at least two-thirds of all staked ETH, the more recent checkpoint becomes justified. The earlier checkpoint in that pair — which was itself justified the epoch before — then becomes finalized.

In practical terms, a transaction included in a block typically needs two full epochs to clear this process, which puts ordinary finality at a little over twelve minutes after inclusion, not immediately and not in seconds. That delay is a deliberate design choice, not a performance limitation that will simply disappear with a future upgrade — though research into single-slot finality aims to shrink it, that design is not live on mainnet.

What “finalized” actually guarantees

Finality is not a statement about likelihood. It is backed by a slashing condition: for a finalized checkpoint to be reverted, validators representing at least one-third of all staked ETH would have to cast contradictory, provably slashable votes. The protocol can detect that kind of double-voting automatically and destroy the offending stake. So reverting a finalized block is not merely “very unlikely” — it is a scenario where an attacker would have to destroy a huge, auditable share of the network’s own staked capital to pull it off, and the protocol is built specifically to catch and punish that behavior.

This is a meaningfully different kind of guarantee than “many confirmations,” the heuristic used on proof-of-work chains, where there is no hard rule for when a block becomes irreversible — only a shrinking probability that grows more comfortable the longer you wait, without ever hitting a defined finish line. Ethereum’s design trades some speed for a rules-based endpoint: a point past which reversal isn’t just unlikely, it’s provably expensive to the attacker in a way the protocol can measure and act on.

Justified vs. finalized

These two words get used loosely, but they describe different points in the same process. A checkpoint can be justified — meaning it has the two-thirds vote needed to move forward — while the chain is still waiting for the next epoch’s vote to actually finalize it. Blocks built on a justified-but-not-yet-finalized checkpoint carry more weight than a fresh, unjustified block, but they are not yet backed by the full slashing guarantee that comes with finality. If you see a service describe something as “justified” rather than “finalized,” that is a meaningfully earlier and weaker state.

Reorgs: where they still happen

A reorg (reorganization) happens when the canonical chain switches from one sequence of blocks to a different one, dropping blocks that had briefly looked canonical. On Ethereum today, reorgs are a head-of-chain phenomenon: they happen among the most recent, unfinalized blocks, where the fork-choice rule is still settling on which branch has more attestation weight. They do not happen to finalized blocks under normal operation — that is the entire point of finality, and it is what separates a finalized block from one that merely has a lot of blocks stacked on top of it.

Shallow, one-block reorgs at the tip of the chain are a known, studied occurrence, usually tied to network latency or a validator proposing a block based on slightly stale information about the current head. They resolve within the unfinalized window and never touch anything that has already finalized. They are a normal, expected feature of a live, globally distributed consensus system, not evidence that something has gone wrong.

The scenario that actually threatens finalized state is different and far more serious: it is not a “longer reorg” so much as a consensus failure. If the chain cannot reach the two-thirds supermajority needed to finalize for more than four consecutive epochs, an inactivity leak mechanism gradually reduces the stake of validators who are not participating. Over time, that lets the remaining active majority regain the two-thirds threshold and resume finalizing on its own. This is a recovery mechanism for a stalled finality process, not a routine event, and it says nothing about reverting blocks that had already finalized before the stall began — those stay final.

How to read this as a user

If you are interacting with Ethereum directly — sending ETH, interacting with a DeFi protocol, watching a deposit arrive — the practical question is usually “how long until this cannot be undone.” A block with a handful of confirmations on top of it is very likely to stay canonical, but “very likely” and “finalized” are different claims, and conflating them is where the risk hides. Services that handle large value, such as exchanges and bridges crediting sizable deposits, often wait for finality specifically because it closes off reorg risk rather than just reducing it to a small number.

For everyday, lower-value transactions, most people do not wait a full twelve-plus minutes — they accept the much smaller residual risk of an unfinalized but deeply buried block. That is a reasonable tradeoff for a small payment and a poor one for a six-figure withdrawal. Knowing which regime you are in is the point of this guide, not a suggestion about what any specific amount should wait for; that judgment call depends on your own exposure and tolerance, not on anything this page can tell you.

This distinction matters even more once you leave L1. An L2 can give you a fast “included” confirmation on its own sequencer long before anything related to that transaction has touched Ethereum’s finality process at all — that is the subject of how to read an L2’s finality and its settlement to L1, where a fast confirmation and a settled one can be separated by a much longer and more consequential gap than on L1 alone.

Risk note

Waiting for finality is a timing choice, not a security purchase. Shallow reorgs at the head of the chain are normal and resolve on their own; they say nothing about the safety of anything that has already finalized. Treat any claim of “instant finality” on a different system with skepticism, and check what exactly it is being finalized against before comparing it to Ethereum’s own guarantee.

Not financial advice. Nothing on this page is a recommendation to buy, sell, stake, bridge, or avoid any asset or protocol.

FAQ

Is a transaction with 10 confirmations “settled”?

Not in the strict sense used here. Ten confirmations make reversal unlikely, but only a finalized checkpoint — reached after two epochs of two-thirds supermajority votes — carries the protocol’s slashing-backed guarantee against reversal.

Can a finalized block still be reorged?

Under normal operation, no. Reverting a finalized block would require provably slashing at least one-third of all staked ETH, which the protocol is designed to detect. The scenarios worth understanding are different: shallow reorgs at the unfinalized chain head, and a chain-wide failure to finalize, which triggers an inactivity leak rather than reverting anything already final.

Why does finality take over twelve minutes instead of being instant?

Finality works on epoch boundaries, not individual blocks. It takes two consecutive epochs — about 12.8 minutes — for a checkpoint to go from proposed to finalized under Casper FFG. Proposals for single-slot finality aim to shrink this, but it is not live on mainnet.

Does this guide apply to Layer 2 transactions too?

Not directly. An L2 can confirm a transaction on its own sequencer almost immediately, well before that transaction’s effects are reflected in any batch or proof that lands on L1. See the companion guide on reading an L2’s finality and settlement for how those two clocks relate.

Frequently asked questions

Is a transaction with 10 confirmations "settled"?
Not in the strict sense used here. Ten confirmations make reversal unlikely, but only a finalized checkpoint -- reached after two epochs of two-thirds supermajority votes -- carries the protocol's slashing-backed guarantee against reversal.
Can a finalized block still be reorged?
Under normal operation, no. Reverting a finalized block would require provably slashing at least one-third of all staked ETH, which the protocol is designed to detect. The scenarios worth understanding are different: shallow reorgs at the unfinalized chain head, and a chain-wide failure to finalize, which triggers an inactivity leak rather than reverting anything already final.
Why does finality take over twelve minutes instead of being instant?
Finality works on epoch boundaries, not individual blocks. It takes two consecutive epochs -- about 12.8 minutes -- for a checkpoint to go from proposed to finalized under Casper FFG. Proposals for single-slot finality aim to shrink this, but it is not live on mainnet.
Does this guide apply to Layer 2 transactions too?
Not directly. An L2 can confirm a transaction on its own sequencer almost immediately, well before that transaction's effects are reflected in any batch or proof that lands on L1. See the companion guide on reading an L2's finality and settlement for how those two clocks relate.

Educational content, human-edited and possibly AI-assisted. It describes how systems work and is not financial, legal or tax advice. Report an error via our corrections page.

More in Protocol

Related articles

See all