What Is a Fork Choice Rule? Ethereum's LMD-GHOST Explained

A fork choice rule is the algorithm a blockchain's nodes use to decide which chain, among possibly several competing valid branches, counts as the canonical one right now. On Ethereum, that rule is called LMD-GHOST (Latest Message Driven Greediest Heaviest Observed Subtree), and it works by having every node tally up validator votes and follow whichever branch has accumulated the most support, rather than simply following whichever chain happens to be longest.

Why a fork choice rule is needed at all

In a distributed network, it's normal — even routine — for two validators to independently propose valid blocks for the same slot, or for network delays to cause different nodes to temporarily see different "current" chain tips. Without an agreed rule for resolving this, nodes could permanently disagree about which transactions actually happened. A fork choice rule gives every node in the network the same deterministic procedure to apply to the same set of observed data, so that — network delays aside — everyone eventually converges on the same view of the canonical chain.

This is a different problem from finality, which is about making a block's position in the chain cryptoeconomically irreversible. The fork choice rule operates continuously and provisionally, picking the current best chain tip moment to moment; finality is the separate, slower process that eventually locks in specific blocks permanently.

How LMD-GHOST works

Breaking down the acronym explains the mechanism:

• Latest Message Driven (LMD): Only each validator's most recent vote (attestation) counts. If a validator changes its vote, its earlier vote is discarded and only the newest one is tallied — this prevents validators from having outsized influence by voting multiple times.
• Greediest Heaviest Observed Subtree (GHOST): Rather than just picking the single chain with the most direct votes on its tip, the algorithm looks at entire subtrees of the block tree and follows whichever branch has the greatest total accumulated weight of votes anywhere beneath it, recursively, down to the current tip.

In practice, this means the algorithm tolerates a certain amount of messy, temporarily conflicting block proposals (which happen constantly in a global, latency-bound network) and still reliably converges on the branch that reflects the most validator support, block by block.

An example

Imagine a block at height 100 has two competing children proposed almost simultaneously: block A and block B. Some validators saw and attested to A first; others saw and attested to B first. Under LMD-GHOST, the network doesn't need every validator to agree instantly — it tallies which of A or B (and their descendants) has more attestation weight behind it, and every honest node running the same algorithm on the same observed votes arrives at the same answer: whichever branch is heavier becomes the head that new blocks build on.

Concept	What it does	Timescale
Fork choice rule (LMD-GHOST)	Picks the current best chain tip based on validator vote weight	Continuous, re-evaluated as new blocks/votes arrive
Casper FFG finality	Locks in specific checkpoints as irreversible	Roughly every 2 epochs (~12–15 min)
Slashing	Punishes provable rule violations (e.g., double voting)	Applied whenever detected

Reorgs under LMD-GHOST

A "reorg" happens when the canonical chain tip changes — a block that was briefly the head gets replaced because a different branch accumulated more weight. Small, shallow reorgs (a block or two) are a normal, expected occurrence in any proof-of-stake network with global latency, and LMD-GHOST is specifically designed to resolve them cleanly using the accumulated vote weight rather than arbitrary tie-breaking. Deep reorgs — many blocks reverting — are far rarer and generally require unusual network conditions or validator misbehavior, and become effectively impossible once blocks are finalized, since finality is a separate, stronger guarantee layered on top of the fork choice rule.

Why this matters beyond protocol design

For everyday users, you don't need to think about LMD-GHOST directly — it operates below the level of any wallet or dapp interface. But it's the reason a handful of confirmations on Ethereum is a reasonably reliable signal even before formal finality: the fork choice rule actively resolves competing branches using real accumulated stake-weighted votes, not just "whichever block arrived first," making it economically costly for an attacker to engineer a favorable reorg without controlling a large share of validator stake.

How LMD-GHOST relates to slashing and attack costs

Because votes are weighted by each validator's staked ETH, an attacker attempting to manipulate which branch the fork-choice rule favors would need to either control a very large share of total stake directly or somehow coerce enough honest validators into voting for a disfavored branch — neither of which is remotely cheap or easy on a network with hundreds of thousands of active validators. Attempts at certain kinds of provable manipulation (like a validator voting inconsistently to trick the fork-choice rule) fall under Ethereum's slashing conditions, meaning a validator caught doing so has a portion of their stake destroyed and is forcibly removed from the active set — a direct, automatic economic penalty layered on top of the fork-choice rule's ordinary vote-tallying logic.

Why this design choice over simply "longest chain wins"

Bitcoin's proof-of-work fork choice rule is simpler in spirit — nodes follow whichever valid chain represents the greatest accumulated mining difficulty, commonly summarized as "longest chain" (more precisely, heaviest chain by cumulative work). Ethereum's proof-of-stake system needed a different approach because there's no equivalent "work" being spent per block the way mining consumes real energy; instead, LMD-GHOST substitutes validator vote weight as the quantity being accumulated and compared, adapted specifically for a system where block proposers are pre-assigned rather than competing through computational effort.

Bottom line

LMD-GHOST is the rule Ethereum's nodes use, moment to moment, to agree on which chain is currently canonical — by following whichever branch has the greatest accumulated weight of the latest validator votes. It's distinct from finality, which is a slower, separate process that eventually makes specific blocks unconditionally irreversible. Together, the two mechanisms let Ethereum tolerate the routine messiness of a global, latency-bound network while still converging reliably on a single, agreed-upon history.