# AI editions

This directory contains linear, retrieval-friendly editions of the State of Brain Emulation 2025 publication set:

- [`report.md`](report.md): the full report, with stable records for all 30 numbered figures.
- [`report-sections/index.md`](report-sections/index.md): bounded chapter files for selective retrieval when the complete report is too large for a useful context window.
- [`at-a-glance.md`](at-a-glance.md): a condensed AI adaptation of the final three-page summary.
- [`asimov.md`](asimov.md): **Article: Building Brains on a Computer**, recovered from the supplied final-draft export with local plot data attached where available. The filename remains stable for durable links.
- [`guesstimator.md`](guesstimator.md): the interactive Guesstimator's model map, assumptions, exact input tables, calculation functions, and source files for agent-driven scenario analysis.
- [`manifest.json`](manifest.json): document versions, source hashes, figure IDs, assets, data status, source paths, and generator functions.
- [`manifest.schema.json`](manifest.schema.json): JSON Schema for the manifest's document, section, table, and figure records.

## Source precedence

1. The published PDFs and live Asimov page are the publication authorities.
2. The Google Docs exports supply recoverable hierarchy, links, references, and long-form text.
3. Exact published-PDF crops are primary where material PDF/export differences were identified; otherwise final-circulation export rasters are retained when the current generator output differs or the exact input snapshot is unavailable.
4. Consumed data files are vendored under [`data/`](data/), and exact snapshots of both generators are vendored under [`code/`](code/); all are hashed in the manifest. Inline numeric data comes from those snapshots or, where disclosed, literal arrays in plotting code.
5. Missing plot inputs are marked unavailable; they are never inferred from an image.

For the Guesstimator, the live interactive app controls user-facing behavior. The AI artifact exposes the committed implementation and its exact inputs; the report PDFs remain authoritative for claims made by the report.

Pandoc substitutes a generic placeholder for eleven complex Google Docs tables. The generator bypasses that lossy conversion and emits the source cells directly as Markdown tables, including the model-organism overview and gap-analysis tables.

## Regeneration

From the website repository root, with the three supplied ZIP exports in the workspace `tmp/` directory:

```sh
python3 scripts/generate-ai-editions.py
```

Regeneration requires Python with `beautifulsoup4`, Pandoc, and Poppler (`pdftoppm`). Generated Markdown, exact publication/export assets, and consumed data snapshots are committed so the deployed website has no runtime dependency on those tools, the source archives, or a moving external-data branch.