AI Dividend Accounting Framework for the United States: Mathematical Specification, Conditional Applications and Reproducibility Package

Additional production does not translate automatically into household income. This working paper presents a reproducible accounting framework for tracing a stated AI-related production scenario through capital consumption, cross-border income, corporate saving, taxes and transfers to resident income and household groups. Australia was the first implementation. This United States application examines the framework under different institutional arrangements, including substantial income from abroad, retained corporate profits and pension-income attribution. It uses US Bureau of Economic Analysis sources and selected scenario settings from the Yale Budget Lab’s AI-Fiscal project. It does not replicate the Budget Lab’s tax model. The contribution is an explicit, inspectable measurement protocol: definitions and equations specify the required inputs; accounting checks reconcile the flows; and evidence labels distinguish reported observations, derived quantities, assumptions and unidentified results. Missing evidence is not silently replaced with zero. In four selected Moderate cases using a 2025 base, with government retaining additional revenue, household income ranges from USD 25.20 to USD 49.32 per USD 100 of additional production. Net resident income ranges from USD 55.96 to USD 82.62 per USD 100 of additional production. These differences illustrate the consequences of source factor-share assumptions and alternative foreign-income responses. They are selected conditional cases, not confidence intervals or estimates of realised US AI gains. The specification preserves the annual illustration using 2023 household component proxies. A separately dated application uses provisional 2024 group proxies and runs Q2 2026 conditional cases with explicitly carried-forward annual recipient ratios. These applications are not a time series of measured AI effects. The empirical real net dividend per person remains unidentified. Editorial release 0.3.1 preserves the computational core and dated application unchanged while clarifying definitions, fiscal labels, aggregation, rounding and presentation. The deposit includes an expert summary, mathematical specification, latest-data report, executable replication package and verification receipt. ChatGPT assisted development and internal verification; Claude was used for separate AI-assisted audits of earlier work. Independent human scientific validation remains outstanding. Research and release responsibility rests with Muhammad Imran Khan, acting in a personal capacity; no employer or institutional endorsement is implied. Author-owned documents and results are licensed under CC BY 4.0; code is MIT-licensed. Third-party materials retain their own terms. Readers should begin with the one-page expert summary. Feedback is particularly welcome on the scenario-to-national-accounts mapping, institutional boundaries and evidence needed to establish marginal household incidence. SHA-256 — AI_Dividend_US_Public_Release_0_3_1.zip2c8223064c7448317b7d0c661a77235eee8f9c95a1c2845697e0fb5c5e2f0262 SHA-256 — packaged MANIFEST.json6cfe2a98a4ea73044e5170f07ae2e93f4bf91654383f22cf7df213408076403e