1\section{Empirical discipline and a falsifiable short-run design}2\label{sec:empirical}34The model is calibrated to 2026 and simulated conditionally; nothing in it is a forecast, and the historical5sample cannot identify the post-AGI parameters ($\alpha^\infty_i$, the floors $\underline p$ and6$\underline\varepsilon$, $\sigma_T$ at compute intensities never observed). What the data can do is (i) pin the7base point and the pre-AI elasticities, (ii) reject the model's description of the \emph{current} phase of the8transition, and (iii) discipline the sign of the mechanisms. This section states what would falsify the model in9the next decade and what we have checked against the historical record.1011\subsection{Validation against pre-AGI patterns}1213The frozen-technology balanced growth path reproduces by construction the 2026 shares of Table~\ref{tab:calibration}.14Three untargeted comparisons are informative. First, the model's aggregate wealth-to-GDP ratio in 202615($\WoverGDPbase$) compares with household net worth of $6.0$ times GDP in 2026Q2 \citep{FedZ1_2026}: the16model's balance sheet, which has no government debt and no foreign assets, is of the right size, with the value of17land ($\LandOverGDPbase$ GDP) within the range implied by \citet{DavisHeathcote2007} and \citet{KnollSchularickSteger2017}.18Second, the model's price--earnings ratio of structures, $1/(r+\delta_N)=\PEstructuresbase$, is of the order of19the historical earnings yield of equities \citep{Shiller2026}; the compute price--earnings ratio of20$\PEcomputebase$ is what rapid obsolescence and falling prices imply and is not observed separately in national21accounts. Third, the heterogeneous economy's wealth Gini ($\GiniBase$) and top-10\% share ($\TopTenBase$) compare22with $0.85$ and $0.69$ in the Distributional Financial Accounts \citep{FedDFA2026}; the top-1\% share ($\TopOneBase$23against $0.325$) is under-predicted, as expected without return heterogeneity.2425The historical decline of the labor share \citep{KarabarbounisNeiman2014,Autor2020} is not evidence for or26against the model: the model's labor share is flat on the frozen path by construction and falls only through the27compute-equivalent channel. What the 2015--2026 record does show, and the model matches at the base point, is28that compute income is still a small share of GDP (information-processing-equipment investment of 2.6\% of GDP,29data-center electricity of 4.4\%): the economy is at the beginning of the transition, where the source's30Proposition~7 predicts gradual wage effects.3132\subsection{Short-run predictions that can be checked}3334Along the baseline path the model predicts, for 2026--2040, a set of co-movements that distinguish it from35competing accounts:36\begin{enumerate}37\item \emph{Wages by task exposure.} Wages of the skill whose task is being automated fall relative to the other38 skill around the automation date and recover afterwards (Restrepo's Proposition~7; Figure~\ref{fig:baseline}).39 A design that regresses occupation-level wage growth on the interaction of task exposure40 \citep{Eloundou2024,FeltenRajSeamans2021,Webb2019} with the time path of compute costs, with occupation and41 year fixed effects and pre-2022 trends as a placebo, tests the sign and the \emph{timing}: the model predicts42 that wages fall \emph{after} the compute-equivalent cost of the task crosses the wage, not when the43 technology becomes feasible.44\item \emph{Investment and margins by industry exposure.} Industries with high task exposure should show rising45 compute investment and \emph{falling} margins on the labor-intensive part of their business (compute is46 competitively supplied), while margins rise only where a scarce complement (sites, grid connections,47 proprietary data) is owned. The model's distinction between $R^Q$ (competitive) and the markup wedge of48 the AI-monopoly scenario gives the two hypotheses.49\item \emph{Valuations.} The model predicts that the market value of compute capital per unit of compute income50 is low ($\PEcomputebase$) and that the value of claims to scarce complements rises relative to GDP. A firm-level51 design would sort firms by the share of their capital that is compute (servers, accelerators; from52 property, plant and equipment footnotes in annual reports) versus land and long-lived structures, and test53 whether the price--earnings ratios of the former are lower and whether their earnings yields54 track the user cost $(1+r)(1+\pi_Q)-(1-\delta_Q)$. Data: Compustat if licensed, or the SEC XBRL55 financial statements, which are public.56\item \emph{Energy prices.} The energy-constrained scenarios predict a rising relative price of electricity in57 compute-intensive regions and a rising share of electricity going to data centers (4.4\% in 2023,58 6.7--12\% projected for 2028 by \citealp{Shehabi2024}). A region-level design using interconnection-queue data59 and wholesale prices would test the compute--energy price link that drives utilization in the model.60\end{enumerate}6162\subsection{Identification, confounding and measurement}6364Each design has the same threats. \emph{Pre-trends}: exposed occupations and industries were already on different65trends (offshoring, routine-biased change); the 2000--2021 window provides placebo tests. \emph{Confounding}: AI66exposure is correlated with education, tradability and firm size; the model implies that the relevant treatment67is the compute-equivalent \emph{cost} crossing the wage, which varies over time within exposure groups and allows68occupation-by-year designs. \emph{Measurement}: exposure indices measure feasibility ($\alpha$ finite), not69realized automation ($A\alpha<w$); the model predicts that feasibility without adoption has no wage effect, which is70itself a test. \emph{Alternative mechanisms}: a demand shock to AI-producing industries raises their margins71without any of the task mechanics; the distinction is that demand shocks raise margins of \emph{producers} of72compute while task displacement lowers wages of \emph{users}. Appendix~\ref{app:data} lists the sources, vintages73and licenses of every series used, and the document \texttt{docs/identification.md} of the replication package74states for each parameter whether it is externally identified, calibrated to a base-point share, or a scenario75input (Table~\ref{tab:calibration}).76