# Publication 01 method

## Working title

**From headline megawatts to delivered compute: an evidence audit of Saudi and UAE AI-data-center expansion**

## Question

What can public evidence actually establish about the location, capacity definition, construction, power pathway, energization, and operating status of the largest recent Saudi and UAE AI-data-center announcements?

## Cutoff and inclusion rule

The evidence cutoff is **2026-07-17**. The main audit includes named Saudi or UAE projects and partner programs announced or materially updated from 2024 through the cutoff that report at least 50 MW. It also retains a multi-country HUMAIN-AMD claim because its unallocated geography demonstrates why partnership announcements cannot be summed into a national pipeline.

National targets and multi-country portfolio targets are context, not project observations. Cloud regions and strategic initiatives without a reported MW value are tracked in the wider atlas but excluded from MW calculations. Operating facilities below 50 MW may be used as calibration cases, not mixed into the headline-project denominator.

## Evidence bundle

Each project is tested on separate dimensions. No composite readiness score is used.

| Dimension | Evidence that counts | Evidence that does not count |
|---|---|---|
| Capacity | A stated value with an identified measure such as IT load, facility load, connection MW, or campus-scale MW | A number whose measure is silently inferred |
| Site | A named facility, parcel, permit address, coordinates, or documented land control | Country-level intent or a city name by itself |
| Construction | Dated ground-breaking, contractor notice, construction permit, progress record, or constructed-facility certification | A future completion date |
| Power | Named utility or supplier, then separately a power allocation, connection agreement, substation, firm MW, reinforcement scope, or energization record | National generation capacity or a generic renewable-energy claim |
| Operation | Energization, commissioning, constructed-facility certification, current owner status, or other dated operating evidence | A target date after that date has passed |
| Operability | Disclosed cooling architecture, PUE/WUE, water source and capacity, fiber diversity, discharge arrangements, and relevant permits | Corporate sustainability language without a site-specific metric |

The table records the strongest observed evidence in each dimension. A blank means **not publicly established by the cutoff**, not “no” and never zero.

## Status vocabulary

The project status is the highest dated milestone supported by evidence: `announced`, `site_identified`, `permitted`, `power_contracted`, `financed`, `under_construction`, `energized`, `commissioned`, or `operating`. Parallel dimensions remain visible because a project can have site control without a connection agreement, or construction evidence without a disclosed capacity definition.

## Source rule

The audit uses project owners and counterparties, regulators, utilities, audited reports, and independent certification records. Press coverage is used only to discover leads. Each row stores a primary source ID, precise locator, supporting source IDs, target-date status, and overlap warning.

## Calculations

The analysis reports counts and coverage shares. It displays the gross sum of the seven headline values only as a diagnostic of the announcement universe and labels it **not a market total** because:

1. the HUMAIN-AMD figure spans Saudi Arabia and the United States;
2. public sources do not resolve overlap among HUMAIN partner programs;
3. Khazna project figures can sit inside wider Khazna portfolio targets; and
4. IT load, facility load, connection MW, and undefined “capacity” are not interchangeable.

No deliverable-MW total is calculated until a project has a defined capacity measure, adequate site evidence, a dated construction milestone, and a documented power connection or energization pathway.

## Known public-disclosure limit

Saudi regulation requires registrants to provide CST with the detailed site address and coordinates, land and white-space area, total facility power in MW, total IT load in MW, rack count, and tier. CST's public facility register names providers, facilities, and classification but does not publish most of those submitted fields. This is a disclosure gap, not a data-collection failure, and it should be identified as such in the article.

## Reproduction

Run:

```bash
python3 scripts/acquire_headline_sources.py
python3 analysis/build_headline_evidence.py
python3 scripts/validate_project.py
```

The first command archives accessible source pages and updates hashes. The second writes the source-reviewed factual table and its machine-readable summary. The third enforces table contracts, source joins, dates, numeric fields, and archived-file hashes.