papers/hanzo-stack-comparison/sections/21-limitations-and-honest-assessment.tex

\section{Limitations and Honest Assessment}

\label{sec:limitations}
%% ============================================================================

We acknowledge the following areas where alternatives outperform the Hanzo + Lux stack:

\begin{enumerate}
    \item \textbf{Raw proxy throughput}: Nginx and Envoy outperform Hanzo Gateway by 15--35\% on pure pass-through proxying. For applications that need only proxying without aggregation or IAM integration, Nginx is faster.

    \item \textbf{Dynamic secrets}: HashiCorp Vault's dynamic secret generation (database credentials, PKI certificates) has no equivalent in Hanzo KMS. Organizations with strict credential rotation requirements (e.g., SOC~2 Type~II with 24-hour credential lifetimes) should use Vault.

    \item \textbf{Inference throughput}: Triton + TensorRT-LLM achieves 5\% higher throughput than Hanzo Engine on NVIDIA hardware due to proprietary kernel optimizations. For NVIDIA-only deployments, Triton remains the throughput leader.

    \item \textbf{Ecosystem maturity}: Auth0, Stripe, and Pinecone have larger user bases, more documentation, and more third-party integrations. Teams with limited engineering capacity may prefer the wider support ecosystem.

    \item \textbf{GraphQL}: Hasura provides a superior GraphQL experience. Hanzo Base does not expose a GraphQL API.

    \item \textbf{P2P networking}: libp2p provides NAT traversal and DHT routing that ZAP does not. For fully decentralized peer-to-peer applications, libp2p remains necessary (and can be used alongside ZAP).
\end{enumerate}

These limitations are engineering trade-offs, not fundamental architecture constraints. Each represents a conscious decision to optimize for the AI-native blockchain use case at the expense of generality.

%% ============================================================================