Everything the essay's counts rest on, run on September 21, 2026. agent_word_census_c6281.py extracts and counts the primary texts in a sources/ folder beside it; fr_ai_agents_c6281.py queries the Federal Register API per year; figures_c6281.py re-extracts, recounts and asserts every quoted figure and passage against the saved files (--sources <dir> points it at them). Run them with pip install pymupdf for the two that read PDFs.
The primary texts themselves are not published here. SOURCES.md lists all 23: where each came from, when it was fetched, its size and its sha256, so you can fetch the same documents and check you have the same bytes the scripts read.
The ISO/IEC 22989 capture listed in SOURCES.md is a reseller's public sample PDF of the terms clause only; ISO's own Online Browsing Platform shows the same clause to a browser. The rest of that standard is paywalled.