papers/hanzo-ai-chain/sections/10-deployment.tex

\section{Deployment}
\label{sec:deploy}

\subsection{Activation Boundary}

The AI Chain activates on \textbf{25 December 2025}, the same boundary at
which Lux Quasar 3.0 (LP-105), QuasarSTM (LP-010), QuasarGPU (LP-132),
and the unified A-Chain attestation surface (LP-134) become canonical on
the Lux primary network. The spec described in this paper is frozen as
of \textbf{15 December 2025}.

There is no separate activation for the AI Chain. Activating Quasar 3.0
activates the AIVM, the HMM registry, the agent registry, the three
markets, and the AI block reward.

\subsection{Pre-Activation Genesis State}

At activation block 0, the chain state is seeded with:

\begin{itemize}[leftmargin=1.2em]
  \item The 1{,}000{,}000{,}000 AI genesis allocation (\S\ref{sec:aicoin})
        distributed to the addresses recorded in the genesis manifest;
  \item The five core Hanzo model registrations
        (\texttt{hmm:zen-text}, \texttt{zen-vision}, \texttt{zen-audio},
        \texttt{zen-code}, \texttt{zen-embed}) at version 1.0
        (\S\ref{sec:hmm});
  \item The vendor root CA set required to validate TEE quotes (Intel
        TDX, AMD SEV-SNP, NVIDIA H100/H200/B100 Confidential Computing,
        ARM CCA);
  \item The genesis validator set inherited from the Lux primary
        network at the activation block;
  \item The initial epoch parameters: 500\,ms block time, 4 AI block
        reward, 70/30 base-fee burn split.
\end{itemize}

\subsection{Operator Onboarding}

Operators may begin onboarding from the spec freeze date (15 December
2025). Onboarding steps:

\begin{enumerate}[leftmargin=1.4em]
  \item Acquire AI from genesis distribution (public sale or
        ecosystem grant);
  \item Stake $\geq$10{,}000 AI to the operator registry;
  \item Provision TEE-capable hardware (any of the supported vendor
        attestation roots);
  \item Load one or more registered HMMs;
  \item Post inference, training, or pool listings.
\end{enumerate}

Operator listings posted before activation are queued; matching begins
at block 0.

\subsection{Caller Onboarding}

Callers (humans, applications, agents) need only an AI balance and an
EVM-compatible wallet to interact with the AIVM precompiles. The Hanzo
LLM Gateway (\texttt{\textasciitilde/work/hanzo/llm}) and the Hanzo Agent
SDK (\texttt{\textasciitilde/work/hanzo/agents}) integrate the chain
calls behind their existing OpenAI-compatible APIs, so existing client
code requires no changes beyond endpoint and credential swaps.

\subsection{Activation Roadmap}

\begin{table}[h]
\centering
\small
\begin{tabular}{lp{8.5cm}}
\toprule
\textbf{Date (2025)} & \textbf{Milestone} \\
\midrule
15 December & Spec freeze. This document final.  AIVM client + node binaries
              tagged. Operator staking opens. \\
17 December & Genesis manifest published. HMM root commitments anchored.
              Vendor TEE root CA set pinned. \\
20 December & Final review window closes. Validator set enumerated. \\
22 December & Operator dry run on the AI Chain shadow network. \\
24 December & Validator readiness gate. Quorum confirms LP-010 / LP-105 /
              LP-132 / LP-134 readiness. \\
25 December & \textbf{Activation block.} AIVM live, AI block reward live,
              compute markets open, agent registry open. \\
\bottomrule
\end{tabular}
\caption{Activation timeline. All dates 2025; all dates inside or before
the LP-105 activation boundary.}
\label{tab:roadmap}
\end{table}

\subsection{Post-Activation: First 90 Days}

\begin{itemize}[leftmargin=1.2em]
  \item \emph{Operator ramp}: target 200 staked operators by activation
        + 30 days, 1{,}000 by + 90 days.
  \item \emph{Inference TPS}: target sustained 10{,}000 inference receipts
        per minute by + 30 days, scaling with operator count.
  \item \emph{Training bounties}: first community-funded fine-tune
        bounty by + 14 days; first cleared bounty by + 45 days.
  \item \emph{Cross-chain receipts}: first F-Chain confidential
        inference receipt by + 7 days; first M-Chain custody-gated
        model load by + 21 days.
\end{itemize}

\subsection{Where Things Run}

The AI Chain code lives under \texttt{\textasciitilde/work/hanzo/aci}
(node binary, AIVM precompiles, registry contracts). Reference clients
live under \texttt{\textasciitilde/work/hanzo/agents},
\texttt{\textasciitilde/work/hanzo/llm}, and the Hanzo Cloud control
plane (\texttt{\textasciitilde/work/hanzo/cloud}). The chain is operated
on the Lux mainnet validator set; AI-specific operators run anywhere
they can produce a valid vendor TEE quote.