Methodology

The Warmth Engine Observatory tracks observable coordination and fragmentation dynamics across geopolitical blocs through verified events and mapped coordination connections.

Primary differentiator: The 7-type Coordination Connections taxonomy is the first systematic framework for tracking relationships between verified AI infrastructure events.

All classifications are based on observable facts from official public documentation. Implementation evidence is required, not announcements. Hidden intent, private communications, and future intentions are explicitly outside scope.

Events progress through five sequential stages. Failure at Stage 1 or 2 excludes the event from the database. Stages 3–5 classify qualifying events and document their relationships.

Events must demonstrate geopolitical significance to qualify for database entry. Four assessment pathways apply based on event nature, with an auto-qualification route for multi-nation coordination.

Events are classified by Event Type — the mechanism by which the action occurs and the actor type involved. Event Type is distinct from Domain (which classifies subject matter) and Tier (which classifies coordination scale).

Events are classified into three tiers based on bloc participation and coordination dynamics — whether events represent coordination or fragmentation. Each tier captures a distinct dynamic in AI infrastructure alignment.

Events are assigned to one of four domains based on their primary subject matter. Domain classification captures what the event is primarily about.

Durability ranges reflect typical commitment horizons for events within each domain.

Each event is tagged with the industry whose operational landscape its documented provisions most directly address. Industry tags are distinct from classification — they identify an external relationship rather than an intrinsic property of the event.

Primary/secondary structure: Each event receives one primary tag. Up to two secondary tags document specific pathways to other industries via mechanism statements.

Each qualified event is assessed for environmental intersections across five tag types. Environmental Nexus Tags (ENTs) are an analytical enrichment layer parallel to industry tags — they identify documented relationships between events and environmental systems, not intrinsic properties of the event itself. A tag is applied when the event meets the qualifying threshold for that type.

Each applied tag carries a mechanism statement — a one-sentence document-verifiable description of how the environmental intersection manifests. Events carry 0–5 tags. Tags use a binary qualification model with evidence classified as Documented or Assessed.

The Warmth Engine Ratio is the ratio of environmental resource demand to environmental governance coordination across the event database. A declining ratio indicates governance converging with demand; a rising ratio indicates demand outpacing governance response.

Confidence Levels: CC-V (Verified) requires primary source documentation. CC-E (Established) requires authoritative statement with corroboration. CC-A (Analytical) permits analyst assessment for parallel policy patterns.

Auditability: Every connection can be verified against the evidence cited.

Four designations classify actors by their structural role. Principal AI Actors (PAA) hold full-stack or near-full-stack capability and anchor an ecosystem — each defines a WEO bloc. AI Infrastructure Keystones (AIK) control chokepoint capabilities that both ecosystems depend on. Advanced Capability State (ACS) hold significant capability that collapses without their principal ecosystem. Participants operate within an ecosystem without independent structural leverage.

Actor capability profiles and bloc alignment data are available on the Bloc Membership Register. The interactive capability panel on the Map provides full dimension detail and source evidence for each actor.

Challenge assignment: Commitments are assigned to the Challenge List Reference — a fixed list of societal challenges drawn from the published lists of major institutions, in which a challenge appears only where several institutions name it independently. The Observatory authors none of the categories; the list is refreshed on a set cycle.

Position in the Observatory: The register is a separate layer with its own unit of analysis, the record. It changes no event's classification and meets the event corpus at one point — a Challenge Attribution, recorded where an event's own source is also a register record.

The register is published as dated observations on the Commitments page. Method and first observation: The Compute Commitment Register — Methodology Working Note (Zenodo), also available as a self-hosted edition.

WEO aims to meet the verification rigour of major financial indices. All events require either 2+ primary sources OR 1 primary + 2 corroborating sources. The 6-month verification lag ensures implementation confirmation before database entry. All cited sources are submitted to the Internet Archive, with capture status tracked across the full evidence base. Current preservation rate: — across — cited source records.

Low-confidence events (<70%) are excluded from the database. Only HIGH (90%+) and MEDIUM (70–90%) confidence events appear in WEO.

WEO's documentation-based verification applies consistently across all blocs and domains. Coverage varies because documentation accessibility varies — some regulatory landscapes produce denser public records than others, and classified activity is excluded from scope across all blocs. Event counts in less-documented areas represent conservative lower bounds, not the full extent of AI infrastructure activity.

WEO applies proven institutional methodologies to tracking AI infrastructure coordination dynamics.