cerkon1/ai-robotics-2032-window

This repository is the data appendix for a three-part Substack article series called The 2032 Window. The series, written by an author who goes by Cerkon, argues that AI capability, robotics costs, execution-layer constraints, and macroeconomic forces will all converge on a window between 2028 and 2035 during which large parts of specific US labor groups become economically replaceable. The repo's job is simple: every number quoted in the articles maps to a row in a CSV stored here, and every chart in the articles is reproducible from a Python pipeline included in the repo. The stated purpose is so that readers who disagree with the analysis can disagree from the same starting point. The folder layout is organised around four legs of the thesis. There is a leg for AI capability and compute cost, drawing on the METR autonomy benchmark and Epoch AI's notable models database. There is a leg for robotics deployment cost, using International Federation of Robotics data and humanoid funding and pricing figures. A third leg covers execution-layer constraints like the power grid, manufacturing, autonomous vehicle reliability, demographics, and regulation. The fourth leg covers macro forces, including capital expenditure from the Mag7 companies, the K-shape split in labor share versus equity returns, and scarce asset performance. Each leg folder has its own README describing the data sources and a research notes file with analytical observations. Alongside the data folders there are top-level documents that set the rules. THESIS.md gives the canonical one-paragraph claim and a section outline. methodology.md spells out projection rules, confidence tiers, data freeze cutoff dates, and a retraction policy. sources.md lists more than thirty sources with status tags such as proposed, pulled, verified, or cited. A separate file in the checkpoints folder contains thirty-five specific falsifiable predictions across 2026, 2028, 2030, 2032, and 2035, plus five explicit failure criteria that would invalidate the thesis. The author has committed publicly to score the 2026 predictions in May 2027 and to issue a sixty-day retraction if any failure criterion fires. Reproducing the charts is short. You clone the repo, change into data/charts, install pandas and matplotlib with pip, and run build_charts.py. This regenerates thirteen PNG files at 200 DPI. The pipeline is described as idempotent, meaning runs against the same input data produce byte-identical outputs. If a user wants to test against newer source data they update the relevant CSVs and re-run, with no code changes needed. The analytical work in this repo, meaning the CSV cleaning, chart code, methodology, and notes, is released under Creative Commons Zero, so it can be reused without attribution. The underlying third-party data sources keep their own licenses, and the README gives a table of which attribution string each upstream source requires.