papers/zoo-quasar-benchmarks/sections/02-background.tex

\section{Background}


\subsection{Quasar Consensus}

\Quasar{}~\cite{lux-quasar-consensus} is the Lux Network's consensus protocol, descended from the HotStuff family of BFT protocols. Key properties:

\begin{definition}[\Quasar{} Consensus]
\Quasar{} is a leader-based BFT protocol with:
\begin{enumerate}
    \item \textbf{Deterministic finality}: Once a block is committed, it cannot be reverted
    \item \textbf{$O(n)$ message complexity}: Linear in the number of validators per round
    \item \textbf{Pipelining}: Multiple blocks in flight simultaneously
    \item \textbf{Responsive}: Advances at network speed, not at a fixed timer
\end{enumerate}
\end{definition}

\Quasar{} achieves finality in two rounds of voting:

\begin{equation}
\text{Finality} = 2 \cdot \text{RTT} + \text{verification} \approx 2 \cdot 3\text{ms} + 4\text{ms} = 10\text{ms}
\end{equation}

where RTT is the round-trip time between validators in a co-located deployment.

\subsection{Proof of AI}

\PoAI{}~\cite{zoo-poai-consensus} is Zoo Network's compute verification mechanism:

\begin{definition}[Proof of AI]
\PoAI{} verifies that claimed AI computation was performed correctly and assesses its quality:
\begin{enumerate}
    \item \textbf{Compute Verification}: \TEE{} attestation proves execution
    \item \textbf{Quality Scoring}: Multi-party evaluation assesses output value
    \item \textbf{Reward Distribution}: Proportional to verified compute quality
\end{enumerate}
\end{definition}

\subsection{HotStuff Lineage}

\Quasar{} extends HotStuff with three innovations:

\begin{itemize}
    \item \textbf{Optimistic fast path}: When all validators are honest and responsive, finality is achieved in a single round (5ms)
    \item \textbf{Stake-weighted voting}: Votes are weighted by validator stake, not uniform
    \item \textbf{Compute-aware leader selection}: Leaders are selected based on both stake and compute capacity, ensuring AI-capable validators produce blocks
\end{itemize}