# AI Demand Shock — public analysis package

The package contains the published assumptions, observations, source register and independent scalar calculations. It contains no original Office files or private research archives.

Unzip the archive and run `python3 analysis/reproduce.py` from its top-level folder. Python's standard library is sufficient. The script reproduces the scenario funding gap, price-volume identity, default utilization and the three OpenRouter four-week blocks. `analysis/calculations.js` exports the same four scenario identities as the interactive page and can be loaded with Node's `require`.

All scenario inputs remain assumptions. Historical company evidence uses a 27 September 2026 availability cutoff. The separately labeled current layer runs through 23 October 2026. Unknown incremental AI profit is retained as null or explicitly unidentified, never zero.

The supplied presentation's 62%/40%/12% headline probabilities are not reproduced by a complete original model. Alternate reconstruction frequencies and conditional intervals are not calibrated crash probabilities. A potential revenue calculation can exceed physical serving capacity; that result is not realizable sales.

`data/analysis.json` contains all public datasets. Individual JSON files and the OpenRouter CSV are provided for convenience. `data/sources.json` identifies original public URLs, periods and source locators. Claims based on reported, selected-period or incomplete evidence retain their limitations. `analysis/original-story.md` preserves the original ten-point thesis, including the old $65B wording; the public page and current claim record separately explain the corrected $70B references.