papers/zoo-per-llm-chains/sections/08-token-economics.tex

\section{Token Economics}
\label{sec:economics}

\subsection{Two-token model}

Each per-LLM chain operates with two distinct economic instruments:

\begin{enumerate}[nosep]
\item \textbf{ZOO} (network-wide) --- the \Zoo{} Network governance and staking token, fixed supply, used for cross-chain validator stake and DAO governance. ZOO is shared across all per-LLM chains and is not minted by any individual chain.
\item \textbf{\$<MODEL>} (chain-specific) --- a per-chain reward unit (e.g., \$ZEN4-CODER, \$ZEN4-MAX). Minted by the chain to pay training, hosting, and research contributors.
\end{enumerate}

The split mirrors the established \Zoo{} pattern: \$ZOO is a fixed-supply ecosystem token; per-application reward tokens are emitted at the application layer to compensate work specific to that application. Per-LLM chains apply this pattern to model lifecycle work.

\subsection{Per-chain token supply}

A per-LLM chain's reward token has a fixed cap of $10^9$ units. The default allocation:

\begin{longtable}{p{4cm}rp{6cm}}
\toprule
\textbf{Bucket} & \textbf{Share} & \textbf{Purpose} \\
\midrule
\endhead
Training rewards & 50\% & Distributed to gradient submitters per attribution \\
Hosting rewards & 20\% & Distributed to inference operators \\
Research rewards & 10\% & Distributed to artifact authors \\
Data contributors & 10\% & Distributed to contributors of training data \\
DAO treasury & 5\% & On-chain governance treasury \\
DEX liquidity & 5\% & Initial liquidity on \Zoo{} DEX \\
\bottomrule
\end{longtable}

\subsection{Emission schedule}

Emissions follow a halving schedule. Default parameters:
\begin{itemize}[nosep]
\item Block reward $r_0$: chain genesis.
\item Halving period: $2^{22}$ blocks ($\approx$ 13 months at 1\,s blocks).
\item Total emissions converge to the fixed cap.
\end{itemize}

The emission rate per block is split across buckets in proportion to the allocation. For example, in any given block, 50\% of the emitted reward token enters the training-rewards distribution pool and is paid out per accepted gradient that block.

\subsection{Inference revenue split}

When the \Hanzo{} AI Chain forwards a fraction (default 30\%) of inference revenue to the per-LLM chain, the funds enter the chain in stable units (USDC-equivalent on \Lux{}) and are distributed without conversion to the chain's reward token:

\begin{equation}
\text{contributor share} = \text{revenue} \cdot \frac{w_{\text{contributor}}}{\sum_c w_c}
\end{equation}

Distribution is per-epoch: the attribution graph snapshot at billing time governs the payout vector for that epoch's revenue.

\subsection{Staking and governance}

Validators that produce blocks for a per-LLM chain stake \$ZOO; operators that commit \GPU{} capacity stake \$ZOO plus a per-chain capacity stake; contributors hold the chain's reward token to vote on chain governance (slashing parameter changes, attribution policy amendments, license amendments).

The chain's reward token is non-transferable for an initial vesting period (default 90 days) to prevent spot-flipping of training rewards.

\subsection{Slashing economics}

Slashed stake is split: 50\% burned, 30\% redistributed to the validator that signed the slashing block, 20\% returned to the chain's DAO treasury. The split discourages collusive slashing while still giving validators an incentive to detect violations.