What Is Snowball Consensus?

In the Snow family, to decide on a transaction or block, each node picks a small random sample of validators on the network and asks for their preference. If a majority of the responses above a certain threshold support the same option, this counts as a successful query for that round. The process repeats over many rounds, and the network quickly converges on a shared decision without requiring many messages.

Snowball adds to the Snowflake protocol. Snowflake only counts how many times in a row it has received the same outcome, and this counter resets when the preference changes. Snowball, on the other hand, keeps a separate confidence counter for each option; an option gains points with each successful query and the node prefers the option with the highest confidence. Thus, a temporary fluctuation in the majority cannot immediately reverse the preference the node has built up over a long time.

The decision is finalized when the preferred option receives enough successful queries in a row. Snowball is the basic building block of the Avalanche (DAG) and Snowman protocols used in practice. Its security is probabilistic: when parameters are chosen appropriately and the share of malicious validators stays below a certain limit, the probability of an incorrect decision is negligible.