Infrastructure Index Update. Oct 2026

This week we built our prototype evaluation framework for the GINC Infrastructure Index. We've developed an initial dataset of roughly 50,000 infrastructure assets across a broad range of categories for our experimental work and have now scored all of these assets on our emerging criticality rubric. The rubric has seven measures in two groups: severity (how bad would a loss of this asset be) and exposure (how far would a loss of the asset be felt). Together they produce three criticality scores (domestic, regional and global), which now drive the prototype indices at infra.ginc.org.

Scoring Philosophy. Every measure in the rubric starts from the same framing of criticality: what is the impact if this asset stops working tomorrow, with today’s alternatives and stocks in place. Each is answered on a five-point scale, from 1 (least) to 5 (most).

Severity: how bad is the loss? Four questions measure severity:

Essential. How essential is the service/outcome that would stop? The scale runs from 1, optional or discretionary use, through economic activity, everyday services (communications, travel, payments, fuel) and vital services (power, heat, health, public safety), to 5, survival needs such as water, food and emergency services within days.
Unique. How much of its role could other assets, routes or suppliers take over within a month? The scale runs from 1, almost all of it, to 5, nothing else can deliver the same outcomes at meaningful scale.
Downtime. How long might it take until the service is back to normal? The scale runs from 1, days, through weeks, months and a year or two, to 5, several years or never.
Ripple. What else might fail because of it? The scale runs from 1, nothing else, through minor effects, one critical system degrading and several failing with it, to 5, a cascade.

An asset’s severity score is the average of the four, from 1 to 5.

Exposure: how far is the loss felt? Three questions use the same five-point scale: 1 not noticed; 2 local (one city, district or industry); 3 widespread (several regions or sectors); 4 major (much of the economy or population); and 5 near-total (normal life and business cannot continue).

Domestic exposure covers the asset’s own country. This is the geographic location of the asset, not the ownership structure/location.
Regional exposure covers neighbouring countries: those sharing a land border, plus any country within approx 500 km, with a minimum of three. These are geographic neighbours, not institutional groupings.
Global exposure covers the rest of the world.

Criticality: domestic, regional and global. Multiplying severity by each exposure score gives three criticality scores, each from 1 to 25. Domestic criticality shows how much the host country depends on the asset, regional criticality how much its neighbours do and global criticality how much the rest of the world does. We never add the three together. A strait that would barely be noticed at home but would halt world trade cannot be described by one number, and combining the scores would hide exactly that difference.

What this gives us. This emerging framework allows us to rank order the ‘Most Critical Assets’ by category and by nation on a broadly consistent basis, noting all of the obvious limitations across categories. These emerging measures help explain what each country depends on, what its region depends on and what the world depends on. The world list includes every asset with global exposure of 2 or more, currently almost 5,000 assets. Keeping the domestic, regional and global scopes separate also produces a new list of custodial assets, where global criticality exceeds domestic by five points or more. These are assets such as the Panama Canal, ASML and Carl Zeiss by example, whose host country bears less of the cost of losing them relative to neighbours the world.

How the scores are made. Each answer is the level the rating model judges most likely. For this initial dataset, we have been trialling System One Models like Jev from TypeSafe. System One models only provide probabilistic estimates of choices (like the five point rubric), they don't provide text responses and hence don't hallucinate the way LLMs can. we provide the model with the name, location, a description and in-house 'why it matters' assessment of the asset and the model provides a probabilistic assessment of each rubric option presented. We take the dominant response as primary and where the top two levels are within five percentage points, we take the lower one. Every asset also carries a confidence band (high, medium or low) based on the variability of the response.

Expect the infra index to evolve rapidly. The prototype index is experimental and the rubric is new, so expect scores to move as we further review and evaluate assets and take in feedback. If a score looks wrong, or you know something about an asset that we have missed, please share feedback. We're also looking for open-source asset lists to ingest and include in our analysis for any country or asset category.