papers/zoo-tokenomics/sections/05-inference-credit-system-earn-by-contribu.tex

\section{Inference Credit System: Earn by Contribution}


Zoo's economic innovation: \textbf{replace pay-to-play with contribute-to-access}. Users earn Inference Credits (IC) through five contribution types:

\subsection{Data Contribution}

Upload quality datasets to earn IC:

\begin{equation}
\text{IC}_{\text{data}} = Q \times V \times R
\end{equation}

where:
\begin{itemize}
\item $Q$: Quality score (0-1, assessed by DAO validators)
\item $V$: Volume (number of examples)
\item $R$: Rarity bonus (underrepresented domains get 2-5× multiplier)
\end{itemize}

\textbf{Example:}
\begin{itemize}
\item Upload 1,000 medical Q\&A pairs
\item Quality score: 0.85 (assessed by domain experts)
\item Rarity multiplier: 3× (medical data scarce)
\item IC earned: $0.85 \times 1000 \times 3 = 2,550$ IC
\item Equivalent value: 25.5M $\KEEPER$ (at 10K $\KEEPER$/IC)
\end{itemize}

\subsection{Experience Contribution}

Submit semantic experiences (training-free GRPO insights):

\begin{equation}
\text{IC}_{\text{experience}} = \frac{U \times E}{1000}
\end{equation}

where:
\begin{itemize}
\item $U$: Usage count (how many inferences used the experience)
\item $E$: Effectiveness score (average improvement vs baseline, 0-100)
\end{itemize}

\textbf{Example:}
\begin{itemize}
\item Submit geometry problem-solving experience
\item Usage: 5,000 inferences over 3 months
\item Effectiveness: +15\% accuracy over baseline
\item IC earned: $\frac{5000 \times 15}{1000} = 75$ IC
\item Equivalent: 750K $\KEEPER$
\end{itemize}

This creates \textbf{perpetual income} for quality contributors - experiences continue generating IC as long as they're useful.

\subsection{Feedback Contribution}

Rate model outputs to improve quality:

\begin{equation}
\text{IC}_{\text{feedback}} = 0.1 \times N_{\text{ratings}} \times A_{\text{agreement}}
\end{equation}

where:
\begin{itemize}
\item $N_{\text{ratings}}$: Number of ratings submitted
\item $A_{\text{agreement}}$: Agreement with consensus (prevents spam)
\end{itemize}

\textbf{Example:}
\begin{itemize}
\item Rate 1,000 model outputs (thumbs up/down, identify errors)
\item Agreement rate: 0.92 (92\% match consensus)
\item IC earned: $0.1 \times 1000 \times 0.92 = 92$ IC
\end{itemize}

\subsection{Code Contribution}

Contribute to Zoo's open-source infrastructure:

\begin{itemize}
\item \textbf{Major Features}: 100-500 IC (e.g., new training algorithm)
\item \textbf{Bug Fixes}: 1-10 IC (depends on severity)
\item \textbf{Documentation}: 0.5-5 IC (per page/tutorial)
\item \textbf{Testing}: 0.1-1 IC (per test case)
\end{itemize}

Assessment by technical DAO committee (5 elected developers review PRs).

\subsection{Compute Contribution}

Donate GPU/TPU time for training/inference:

\begin{equation}
\text{IC}_{\text{compute}} = \frac{H \times P}{100}
\end{equation}

where:
\begin{itemize}
\item $H$: Hours of compute donated
\item $P$: Performance score (relative to baseline A100 GPU)
\end{itemize}

\textbf{Example:}
\begin{itemize}
\item Donate 100 hours of RTX 4090 (performance score: 0.8× A100)
\item IC earned: $\frac{100 \times 0.8}{100} = 0.8$ IC
\end{itemize}

This enables \textbf{resource-rich but cash-poor} participants (students with gaming PCs, researchers with institutional access) to earn inference access without monetary payment.

\subsection{Economic Impact of Contribution Model}

Traditional AI platforms (OpenAI, Anthropic) extract 95%+ of value from data contributors. Zoo inverts this:

\begin{table}[h]
\centering
\begin{tabular}{lccc}
\toprule
\textbf{Platform} & \textbf{Data Contributor Cut} & \textbf{Platform Cut} & \textbf{Model} \\
\midrule
OpenAI & 0\% & 100\% & Pay-to-play \\
Anthropic & 0\% & 100\% & Pay-to-play \\
Scale AI & 5-10\% (contractors) & 90-95\% & Pay labor \\
Bittensor & 30\% (miners) & 70\% (validators/TAO) & Mine-to-earn \\
\textbf{Zoo} & \textbf{90\%} & \textbf{10\%} & \textbf{Contribute-to-access} \\
\bottomrule
\end{tabular}
\caption{Value distribution across AI platforms}
\label{tab:value_dist}
\end{table}

Zoo's 90/10 split (contributors retain 90\%, DAO takes 10\% for infrastructure) represents \textbf{18-fold improvement} over traditional platforms.

\textbf{Empirical Validation:}

From September-October 2025 (first 6 weeks post-airdrop):
\begin{itemize}
\item \textbf{Data Contributions}: 12.4M examples uploaded (worth \$6.2M at \$0.50/example market rate)
\item \textbf{IC Distributed}: 142,000 IC (\$2.84M equivalent at \$20/IC)
\item \textbf{Contributors}: 8,472 unique contributors
\item \textbf{Average Earnings}: 16.8 IC per contributor (\$336)
\item \textbf{Top Contributor}: 2,847 IC earned (\$56,940)
\end{itemize}

This demonstrates that contribute-to-access creates \textbf{real economic value} for community participants, not just speculative token appreciation.