RIADeFiPublished by Ketju Research
The AtlasReviewed 2026-08-14

Avalanche: who orders transactions?

599 Primary Network validators, top validator 1.72% of stake; no single operator

Governed, no freeze

This page grades the chain layer only. Each asset and protocol on the chain carries risks of its own.

Who controls it

Control assessmentApproved with limits
Stagen/a
Reviewed2026-08-14
Transaction ordering
599 Primary Network validators, top validator 1.72% of stake; no single operator
Escape hatch
n/a: L1
Upgrade authority
No admin key controls user funds. Upgrades run through the open ACP process, but in practice Ava Labs writes, releases and schedules every fork.
Halts and outages
Avalanche has had two full outages, both from single-client bugs that an emergency Ava Labs release fixed: 2023-03-23 (AvalancheGo v1.9.12) and 2024-02-23 (a finality stall of about 5 hours). Neither caused a rollback or a loss of funds. Ketju found no halt in the roughly 2.5 years since, the same recency as Solana’s record. Unlike Solana, Avalanche has no second client on the way. Validator count has roughly halved since 2022, from about 1,200 to 599. The fall coincides with Etna (ACP-77), which let L1 operators stop validating the Primary Network. Ketju measured a Nakamoto coefficient of 23 on-chain at this review.

Ketju assessment

Ketju approves Avalanche with limits: no party can freeze or seize C-Chain funds, but one company, Ava Labs, writes the only production client and leads recovery from any halt. Every venue Ketju tracks on Avalanche runs on the C-Chain. The C-Chain sits on the Primary Network and inherits nothing from subnets. A subnet failure shrinks that L1’s security budget but does not affect C-Chain consensus. The ordering layer is decentralised: 599 validators, a Nakamoto coefficient of 23, no admin key over funds, and no freeze mechanism shown or designed. The operating layer is concentrated. AvalancheGo is the only production client, and both past outages were single-client bugs that waited on one vendor’s patch. Messari reports over 33% of stake hosted on AWS, and Ava Labs runs the default RPC, indexer and wallet pipeline. A compelled Ava Labs could not take funds, but it could ship a censoring client or filter the default RPC. This verdict lets Ketju judge benqi-staked-avax and the other Avalanche-only venues on their own merits.

Review triggers

  • Any full-network halt requiring a coordinated restart
  • Nakamoto coefficient falling below 15
  • Primary Network validator count falling below 450
  • A single hosting provider exceeding 40% of stake
  • The Avalanche Bridge wardens (6-of-8 as last published, 2022; Ava Labs-selected) moving collateral outside the published process

Each trigger names a fact anyone can observe. If one occurs, Ketju will review the assessment again. None is a forecast.

Questions before use

  1. Does the client’s thesis require censorship resistance, or primarily low-cost settlement?
  2. Can the client withdraw without help from the party that normally orders transactions?
  3. Who can upgrade bridges or contracts, and what delay applies?
  4. Does the asset introduce an issuer weaker than the chain grade?
  5. What evidence will show a halt, censorship event, or governance change?

← The full Atlas