\section{Additional figures and tables}
\label{app:extra}

\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{figA_savings_news.pdf}
\caption{Savings mapping (negative result). Response of log capital to a one-log-point TFP news shock in partial
equilibrium (prices fixed) and in the symmetric rational-expectations equilibrium of the log-linearised economy
($\gamma=2$, $\rho_z=0.9$, $\delta=0.06$, $\beta=0.96$). The partial-equilibrium response, $\kappa_0$, is $-0.0015$:
wealth and substitution effects nearly cancel, so the consumption--savings decision carries a negligible stake for
delegation. Source: \texttt{scripts/10\_model\_a.py}.}
\label{fig:A_savings}
\end{figure}

\begin{figure}[htbp]
\centering
\includegraphics[width=0.7\textwidth]{figA_heatmap.pdf}
\caption{Sensitivity: variance of the log wage attributable to model noise at the equilibrium delegation share, over
the model-noise standard deviation (rows) and the effective number of independent foundation models (columns).
Source: \texttt{scripts/10\_model\_a.py}.}
\label{fig:A_heatmap}
\end{figure}

\begin{figure}[htbp]
\centering
\includegraphics[width=0.95\textwidth]{figC_designs.pdf}
\caption{Model C, Bayesian control: order-flow correlation among delegators and return volatility by market design,
delegation share and number of models (20 seeds, bootstrap 95\% intervals). Source: \texttt{scripts/30\_model\_c.py}.}
\label{fig:C_designs}
\end{figure}

\input{tables/model_b_decomposition}