\section{Replication and tests}
\label{app:replication}

The replication package (UQO repository \texttt{wp24\_uqo}) reproduces every number, table and figure of the
paper with \texttt{make all} (Python 3.14, pinned requirements, TeX Live 2026). The steps and their outputs are:

\begin{enumerate}
\item \texttt{make test}: the test suite. Static block --- Propositions~1--6 of the source hold in the benchmark
      to $10^{-6}$; the Karush--Kuhn--Tucker conditions of the task allocation are satisfied to $10^{-6}$; the
      income identity holds to $10^{-6}$ relative to GDP and a deliberately broken accounting (compute services
      counted as final output) is rejected; utilization falls below one when energy binds. Dynamic block --- the
      frozen-technology balanced growth path solves the dynamic system to $10^{-6}$ without iteration; the banded
      Jacobian agrees with a dense finite-difference Jacobian to $10^{-6}$; a small technology shock converges and
      satisfies the no-arbitrage identity $(1+r_{t+1})W_t=\text{payouts}_{t+1}+W_{t+1}$ to $10^{-6}$; every saved
      scenario satisfies the residual, identity and no-arbitrage checks.
\item \texttt{make benchmark}: Figure~\ref{fig:benchmark}, Table~\ref{tab:benchmark}, Figure~\ref{fig:regimes}.
\item \texttt{make scenarios}: the \NumScenarios{} scenarios of Table~\ref{tab:diagnostics}, each a 275-year
      perfect-foresight path with its parameter file, meta-data (convergence, residual, horizon check, solution
      time) and path arrays.
\item \texttt{make ha}: the heterogeneous-agent general equilibrium for the baseline, the frozen economy and the
      citizen fund.
\item \texttt{make figures tables paper}: every figure and table of the paper and the file
      \texttt{numbers.tex} of macros, one per number quoted in the text.
\end{enumerate}

Data are downloaded by \texttt{make data} from the key-less endpoints listed in Appendix~\ref{app:data}; the
processed calibration targets are versioned with the package so that the paper compiles without network
access. No random seeds are used: all solvers are deterministic and the distribution of households is
simulated non-stochastically. Solver tolerances are $10^{-14}$ (static objective), $10^{-9}$ (maximum residual of the
transition) and $10^{-4}$ (log excess demand for claims in the heterogeneous economy). The research log
(\texttt{docs/research\_log.md}) records the numerical lessons learned during the construction, including three
specification changes made for numerical reasons: the removal of accessory tasks from the dynamic model, the
treatment of capitalists' expenditure as an unknown, and the re-calibration of the base point for each scenario.

\begin{figure}[htbp]
\centering
\includegraphics[width=\textwidth]{fig02_regimes.pdf}
\caption{The limit economy of Section~\ref{sec:theory} computed with the static block: output, price of compute,
wage of the higher skill and factor shares as compute grows with the complement $S$ fixed, for
$\sigma\in\{0.5,1,1.5\}$ (Proposition~\ref{prop:regimes}). Source: \texttt{scripts/11\_regime\_map.py}.}
\label{fig:regimes}
\end{figure}