papers/zoo-quasar-benchmarks/sections/07-comparative-evaluation.tex

\section{Comparative Evaluation}

\label{sec:comparison}

\subsection{Single-Layer Alternatives}

We compare Zoo's dual-layer design against three single-layer alternatives:

\begin{enumerate}
    \item \textbf{Quasar-only}: No compute verification. Faster but no AI quality guarantees.
    \item \textbf{PoAI-only}: Every block requires compute verification. Slower but thorough.
    \item \textbf{Combined}: Compute verification integrated into consensus voting. Medium speed.
\end{enumerate}

\begin{table}[t]
\centering
\begin{tabular}{lcccc}
\toprule
\textbf{Design} & \textbf{Finality} & \textbf{TPS} & \textbf{Useful Compute} & \textbf{Safety} \\
\midrule
Quasar-only & 6.3ms & 4,760K & 0\% & \checkmark \\
PoAI-only & 45ms & 890K & 99.5\% & \checkmark \\
Combined & 28ms & 1,420K & 95\% & \checkmark \\
\textbf{Dual-layer} & \textbf{6.7ms} & \textbf{4,712K} & \textbf{98.2\%} & \checkmark \\
\bottomrule
\end{tabular}
\caption{Comparison of consensus designs. Dual-layer achieves near-Quasar performance with near-PoAI compute utilization.}
\label{tab:designs}
\end{table}

The dual-layer design achieves 99.3\% of Quasar-only performance while capturing 98.7\% of PoAI-only useful compute. This is strictly better than the combined approach in both dimensions.

\subsection{Comparison with Other AI Chains}

\begin{table}[t]
\centering
\begin{tabular}{lcccc}
\toprule
\textbf{Chain} & \textbf{Consensus} & \textbf{Finality} & \textbf{AI Verify} & \textbf{TPS} \\
\midrule
Zoo Network & Quasar + PoAI & 6.7ms & TEE + scoring & 4.7M \\
Bittensor & Yuma & 12s & Emission-based & 200 \\
Ritual & DPoS & 2s & Infernet oracles & 1,000 \\
Gensyn & Truebit-variant & 30s & Re-execution & 500 \\
\bottomrule
\end{tabular}
\caption{Zoo Network vs other AI-focused chains. Zoo achieves orders-of-magnitude higher throughput with faster finality.}
\label{tab:ai-chains}
\end{table}