Methodology

What this is — and is not

What this is — and what it isn't

Think of this as a measuring tape, not a crystal ball. We measure how much a country can access, use, and gain real advantage from cognitive infrastructure — the tools, data, compute, talent, and institutions that help people and machines think and decide. We do not measure "intelligence," and we never rank how smart anyone is.

We show every country through two lenses:

• CPI (absolute power) — the country's total weight in the world, like the size of an engine.
• CC (per-capita) — how much reaches the average person, like miles-per-gallon.

A big country can score high on CPI but low on CC, and that gap matters.

Finally, anything pointing to the future is a scenario — a "what if this trend held?" sketch, not a prediction. All numbers here are model constructs we built, not facts handed down from nature. Read them as a structured way to compare, question, and discuss.

What this is. Cognitive Coefficient is a composite-indicator and scenario platform. It measures, as transparently as possible, the access to, utilization of, and leverage from cognitive infrastructure — AI systems, compute, data, education, digital skills, innovation capacity, and knowledge networks — across economies and occupations.

What this is not. It does not measure intelligence, human worth, or destiny. It does not claim causality. It does not predict outcomes. Every index is a model construct over curated data, and every projection is a scenario with explicit assumptions and uncertainty — not a forecast of actual outcomes.

Measurement vs. prediction. We deliberately separate:

• Measurement — present-day composite indices computed from observed/curated indicators (CC, CDI, CII, CSI, AIR, AIX, CRI).
• Exposure & capability — structural characteristics (e.g. occupational AI exposure for OFI).
• Forecasts & scenarios — bounded-growth and diffusion projections under alternative assumptions (CV/CM are forward

signals; the scenario engine and OFI projections are explicitly scenario-based).

Standing disclaimer. Throughout the platform: these are model outputs and scenarios, not forecasts of actual outcomes.

Framing: measurement, not prediction

This platform operationalizes a single latent construct — a polity's access to, utilization of, and leverage from cognitive infrastructure (compute, data, models, talent pipelines, and the institutions that convert them into capability). It is explicitly not an estimator of "intelligence," national or otherwise; that term is excluded by design to avoid reifying a contested and ethically loaded concept.

Every unit is reported under two complementary aggregations:

• CPI — Cognitive Power Index (absolute): total systemic weight, scaling with population and resource stock. Read it as extensive capacity.
• CC — Cognitive Capita (per-capita): intensity of access for the representative individual. Read it as an intensive margin.

The two routinely diverge; the CPI–CC gap is itself an object of analysis, not noise to be reconciled.

Standing disclaimer. All outputs are model constructs: composite indices contingent on indicator selection, normalization, and weighting choices documented here. Any forward-dated value is a scenario — a conditional projection under stated assumptions — and carries no probability claim. The instrument is built for comparison, stress-testing, and deliberation, not point forecasting.

Data & provenance

Where the v1 numbers come from

This first version (v1) is a starter dataset — carefully assembled, but a draft. Every value carries a tag saying where it came from: some are real reported figures, some are pulled from public sources, and some are estimates we made to fill gaps. Nothing is hidden; if a number is a best guess, it says so.

To compare very different things on one scale, we rescale each indicator to run from 5 to 100. The lowest value becomes 5, the highest becomes 100, and everything else lands in between. That keeps any single raw unit (dollars, gigawatts, headcounts) from dominating.

Some indicators are wildly lopsided — a few giants, a long tail of small players. For those we apply a log transform, which gently compresses the extremes so the giants don't flatten everyone else into a blur.

One honest catch: v1 is a single snapshot in time. So anything that looks like "growth" or a trend over years is a rough heuristic, not a measured history. Treat the time-based pieces as illustrative scaffolding for now.

v1 data is curated and source-tagged. Each indicator value carries a source and year and an estimated flag. Where a precise current value was unavailable, a best-calibrated estimate is used and flagged estimated: true. Anchor indicators (internet use, R&D/GDP, tertiary enrollment, GDP) draw on well-documented public statistics (World Bank, ITU, UNESCO, OECD, IMF); AI-specific indicators draw on the Stanford AI Index, Tortoise Global AI Index, Oxford Insights Government AI Readiness, WIPO, and Top500. This is an illustrative-but-anchored v1 panel, not an official statistical release. A live ETL layer (documented under Architecture) is the roadmap to continuously-ingested official data.

Normalization. Every indicator is min-max normalized across the country panel to a 5–100 range (a 5-point floor avoids a single worst performer collapsing to a degenerate zero). Heavy-tailed indicators (GDP, GDP per capita, private AI investment, patents, researchers, scientific publications, Top500 systems, data-center MW, cloud regions, notable models, AI publication share) are log-transformed (log1p) before normalization. All indicators are oriented so that higher = more cognitive infrastructure. Missing cells are excluded from their pillar's mean rather than imputed.

Cross-sectional, single-vintage. The v1 panel is a single cross-section. There is no historical time series, so the temporal indices (CV, CM, CEV) are forward heuristics derived from present-day structure — not estimated trends. This is stated explicitly in each temporal index below and in the code's own docstrings.

Provenance and v1 transformations

The v1 corpus is a curated, source-tagged cross-section. Each indicator value carries a provenance flag distinguishing reported (primary/official), derived (computed from public sources), and estimated (analyst-imputed to close coverage gaps). Imputation is disclosed at the cell level rather than silently smoothed, so consumers can filter or down-weight estimated inputs.

Normalization. Indicators are mapped to a common min–max scale on [5, 100] — empirical min → 5, max → 100, linear in between. The 5-floor (rather than 0) preserves a nonzero presence for the weakest unit and avoids degenerate zeros propagating through products and ratios.

Heavy-tailed indicators. Quantities with extreme right skew (compute stock, capital flows, model counts) are log-transformed prior to normalization, so that multiplicative differences map to roughly additive distances and a handful of leaders do not compress the remaining distribution to the floor.

Temporal caveat. v1 is a single cross-section. Consequently, any index presented as a trajectory, growth rate, or time series is a heuristic — constructed from assumed dynamics or proxy scaling, not estimated from panel observations. Longitudinal interpretation should await multi-period data.

CPI · Cognitive Power Index

Measures: The Cognitive Power Index is a country's total brainpower in the AI economy — its overall size and muscle, not how efficient it is per person. It's why huge countries like the US and China sit at the top.Absolute national capacity in the intelligence economy — the headline, great-power lens (the US and China lead by a wide margin).An absolute (extensive, non-normalized-to-population) measure of national capacity in the intelligence economy: a scale-sensitive composite latent construct capturing aggregate AI capital, compute stock, frontier-model presence, robotics, total research output, and economic mass. 0–100

We gather several big things that make a country powerful in AI: money invested, computing hardware, top-tier AI models, robots, total research, and the size of the economy. Because these numbers span from tiny to enormous, we first shrink the giant ones down to a fairer scale (so a country isn't ranked first just because one number is astronomically large). Then we score each piece and add them up, giving more weight to the pieces that matter most, to get one combined power score.

CPI is a weighted, log-normalized composite of absolute MASS. Per-capita inputs (researchers, publications, patents) are first multiplied by population to recover national totals; every component is log1p-transformed and min-max normalized to 5–100 across the panel, then combined with these weights:

Component	Weight
AI capital mobilized (private + state + infrastructure)	2.5
Compute / data-center capacity	1.5
Frontier AI models	1.5
Robotics & automation (industrial-robot stock)	1.5
Supercomputing (Top500)	1.0
Semiconductor capability	1.0
Researchers (total)	1.0
Research output (total)	1.0
Patents (total)	1.0
Economic mass (GDP)	0.5

CPI = Σ(wᵢ·normᵢ) / Σwᵢ. The heavy weights on AI capital, compute, frontier models and robotics mean the two economies that mobilize the most of each — the US and China — dominate and separate sharply from the rest of the field.

For each indicator x_i, apply a log transform (e.g. ln(1+x_i)) to compress heavy right-skew, then min-max or z-normalize the logged values to a common 0-100 scale; aggregate as a weighted arithmetic mean CPI = Σ w_i · ñ_i with Σw_i = 1, weights reflecting the relative importance of capital/compute/frontier/robotics/research/GDP. Properties: log-normalization yields an approximately multiplicative-to-additive mapping and dampens (but does not remove) the dominance of scale, so populous/large-GDP states still rank highest by construction; the index is compensatory (strength in one pillar offsets weakness in another) and sensitive to weight and normalization-bound choices.

Inputs: AI capital mobilized, Data-center capacity, Frontier models, Industrial-robot stock, Top500 systems, Semiconductor capability, Total researchers / publications / patents, GDP
Sources: Stanford AI Index, IFR World Robotics, Top500, WIPO, World Bank, state & infrastructure capex (curated)

Assumptions

• Absolute mass (not per-capita) is the correct lens for great-power capacity.
• 'AI capital' is deliberately broadened beyond private VC to include state programs and infrastructure/semiconductor capex, which is where the US and China are both enormous.

Caveats

• Rewards scale, so large economies rank high regardless of per-capita depth — use CC for the per-person view.
• 'AI capital mobilized' aggregates heterogeneous public/private/infrastructure sources and is partly estimated.
• Robotics uses operational industrial-robot stock (IFR); service/military robotics are not yet included.

ADI · Agentic Deployment Index

Measures: The Agentic Deployment Index shows how widely and powerfully a country actually puts autonomous AI 'agents' to work — software that can act on its own — across ordinary people, businesses, and government.How widely and powerfully a nation deploys autonomous, tool-using AI AGENTS — across citizens, business and government. The platform's central forward lever of cognitive advantage.A deployment-and-diffusion construct quantifying the breadth and depth of autonomous-agent fielding across the citizen, enterprise, and public sectors — combining a capability (supply-side) lens with adoption/reach (demand-side) lenses. 0–100

We measure five things: how capable the agents are, how easily citizens can access them, how much freedom and reach those agents have, how much businesses use them, and how much government uses them. Each is scored on the same scale, then we take an average that leans more heavily on the parts judged most important to real deployment.

ADI is a weighted mean of five 0–100 agentic indicators (already on a common scale, so no panel normalization):

Component	What it captures	Weight
Agent capability	the 'intelligence' / reliability of agents nationally available (frontier agentic reasoning, planning, tool-use)	1.5
Access — citizens	share of the workforce/population that can actually field capable autonomous agents	1.5
Autonomy & reach	what agents are permitted and able to DO — tool/API/MCP ecosystems, payment/identity/data rails, regulatory permission	1.2
Business deployment	enterprises running agents in real workflows	1.0
Government deployment	public-sector agentic deployment	0.8

ADI also feeds the headline indices: agent access × population enters CPI (the absolute 'agent army' scale), and an agentic pillar enters CC (per-capita).

Normalize each of the five components (agent capability, citizen access, autonomy & reach, business adoption, government adoption) to a common 0-100 scale, then aggregate as a weighted mean ADI = Σ w_k · ñ_k, Σw_k = 1, with weights tilted toward the components most diagnostic of effective deployment. Properties: compensatory weighted-arithmetic aggregation; capability acts as a supply ceiling while the three adoption/reach terms capture realized diffusion, so the index can flag capability-rich but adoption-poor states; sensitive to the weight vector, which encodes a normative stance on what 'powerful deployment' means.

Inputs: Agent capability, Agent access (citizens), Agent autonomy & reach, Business agent adoption, Government agent adoption
Sources: Curated forward estimates (anchored to frontier-lab/agent-ecosystem, enterprise-adoption and gov-AI signals)

Assumptions

• An automated army of capable agents acting on one's behalf is a — perhaps the — central near-term lever of cognitive advantage.
• Capability, access, and autonomy compound: many agents that are smart AND permitted to act matter more than any one alone.

Caveats

• This is a forward-looking, partly-subjective construct — the most estimate-heavy index here; treat the ordering as indicative, not precise.
• Agent deployment is moving fast; values are a 2025 snapshot.

CC · Cognitive Coefficient

Measures: The Cognitive Coefficient measures how intensely AI is woven into a country per person — like income-per-person instead of total income. Small, advanced countries do very well here even though they're not huge.Overall access to, utilization of, and leverage from cognitive infrastructure for an economy.An intensive (per-capita / quality-of-system) composite analogous to GDP-per-capita or the HDI, characterizing the depth of cognitive-economy penetration independent of national scale, across nine normalized pillars (AI access, adoption, compute, data, education, digital literacy, innovation, knowledge infrastructure, agentic capability). 0–100

We look at nine different ingredients of an AI-ready society — things like access to AI, how much people actually use it, computing power, data, education, digital skills, innovation, knowledge infrastructure, and use of AI agents. Each ingredient is scored on the same 0-100 scale so they're comparable, and then we simply take the plain average of all nine. Every pillar counts equally.

CC is the arithmetic mean of 8 normalized pillars: AI Access, AI Adoption, Compute, Data, Education, Digital Literacy, Innovation, and Knowledge Infrastructure.

Each pillar is a weighted arithmetic mean of its member indicators (member weights shown in the data layer; e.g. R&D/GDP and data-center capacity carry weight 1.5). Each indicator is first min-max normalized across the panel to 5–100 (log1p-transformed first for heavy-tailed indicators). The pillars are then combined with equal weight:

CC = (1/8) · Σ pillar_p

Missing indicators are dropped from their pillar's mean; missing pillars are dropped from CC.

Each pillar p_j is normalized to [0,100] (min-max against the cross-country distribution), then aggregated as an unweighted arithmetic mean CC = (1/9) Σ_{j=1..9} ñ_j. Properties: equal weighting is a transparent, assumption-light default but implies full substitutability across pillars and equal marginal value; the arithmetic mean is fully compensatory, so a country can mask a near-zero pillar with high scores elsewhere — contrast with the geometric aggregation used in CDI. Being intensive, it favors small high-capability economies and is roughly scale-invariant.

Inputs: AI Access (internet, broadband, cloud regions), AI Adoption (AI-skill penetration, national AI capacity, AI talent), Compute (Top500, data-center MW, semiconductor), Data (data governance, sci. publications, digital skills), Education (tertiary, schooling, human capital, STEM share), Digital Literacy (digital skills, AI-skill penetration, internet), Innovation (R&D/GDP, researchers, patents, GII, AI investment), Knowledge Infrastructure (publications, notable models, AI-pub share, gov AI readiness)
Sources: World Bank WDI, ITU, UNESCO UIS, OECD, WIPO GII, Stanford AI Index, Tortoise Global AI Index, Oxford Insights, Top500

Assumptions

• Equal weighting across the 8 pillars is a transparent default, not an optimized weighting.
• Min-max (5–100) normalization makes CC a relative, panel-dependent measure.
• Log transforms tame heavy-tailed indicators before normalization.

Caveats

• CC is relative to the 48-economy panel, not an absolute capacity.
• Sensitivity to weighting choices is not yet published.
• Curated v1 data includes flagged estimates.

CDI · Cognitive Development Index

Measures: The Cognitive Development Index is a single 0-to-1 scorecard of how developed a country is for the AI age, built in the same spirit as the UN's Human Development Index. A balanced country scores better than a lopsided one.Balanced cognitive development, penalizing imbalance across dimensions (HDI-style).An HDI-style bounded [0,1] composite of cognitive-economy development across five normalized dimensions (AI readiness, digital, education, innovation, knowledge), deliberately using non-compensatory aggregation to penalize cross-dimensional imbalance. 0–1

We take five areas — AI readiness, digital, education, innovation, and knowledge — each scored on the same scale. Instead of a plain average, we multiply them together and take a root (a 'geometric mean'), which is a way of averaging that punishes imbalance. If a country is great in four areas but very weak in one, that weak area drags the whole score down, so you can't paper over a gap.

CDI is the geometric mean of 5 normalized dimension indices, each scaled to [0,1]:

CDI = (D_ai · D_digital · D_education · D_innovation · D_knowledge)^(1/5)

Each dimension is the mean of its normalized (0–100) components ÷ 100. The geometric mean (as in the HDI since 2010) penalizes uneven development: a low score in any dimension drags the whole index down. Component floor of 0.001 prevents zero-collapse.

Normalize each dimension d_m to [0,1], then aggregate via the geometric mean CDI = (Π_{m=1..5} ñ_m)^{1/5}. Properties: the geometric mean is sub-compensatory — it is bounded above by the arithmetic mean (AM-GM inequality) and collapses toward 0 if any dimension approaches 0, so poor balance is penalized and substitutability across dimensions is limited; it is undefined/zero-dominated at boundary values, typically handled by flooring normalized scores above 0. Mirrors the post-2010 HDI methodology.

Inputs: AI readiness (gov AI readiness, national AI capacity, AI governance), Digital capability (internet, digital skills, mobile broadband), Education (tertiary, mean schooling, human capital), Innovation (R&D/GDP, patents, GII), Knowledge production (sci. publications, AI-pub share, notable models)
Sources: UNDP HDI (method), World Bank, UNESCO, OECD, Stanford AI Index

Assumptions

• Five equally-weighted dimensions.
• Geometric aggregation across dimensions; arithmetic within a dimension.

Caveats

• Panel-relative (0–1 is relative to observed min/max).
• Missing dimensions are dropped, which can affect cross-country comparability.

CII · Cognitive Inequality Index

Measures: The Cognitive Inequality Index measures how unfairly access to AI and cognitive tools is spread across the world's people — not across countries equally, but counting people, so a populous country weighs more.Inequality in access to cognitive infrastructure across the world's people.A global, population-weighted inequality statistic: the Gini coefficient of the cross-country Cognitive Coefficient (CC) distribution, treating people (not states) as the unit so it reflects inequality of cognitive-economy access across the world's population. 0–1

We take each country's per-person AI intensity score and ask: if you lined up all the people in the world by how much AI capacity they have, how uneven is that line? We use a standard inequality yardstick (the Gini measure) where 0 means everyone is equal and 1 means one group has everything. Because we weight by population, big-population countries shape the result more than tiny ones.

CII is the population-weighted Gini coefficient of the cross-country CC distribution. Each economy contributes its CC value weighted by population:

Gini = Σᵢ Σⱼ wᵢ wⱼ |CCᵢ − CCⱼ| / (2 · W² · μ_w)

where wᵢ is population, W = Σ wᵢ, and μ_w is the population-weighted mean CC. We also report the Theil-T index and the Lorenz curve (cumulative share of people vs. cumulative share of cognitive capital).

Treating each country's CC as the 'income' attached to its population mass, compute the population-weighted Gini G = (Σ_i Σ_j p_i p_j |CC_i − CC_j|) / (2 μ̄), where p_i is population share and μ̄ the population-weighted mean CC. Properties: bounded [0,1], 0 = perfect equality, →1 = maximal concentration; satisfies scale invariance and the Pigou-Dalton transfer principle but, being between-country only (CC is a national mean), it ignores within-country dispersion and so understates true person-level inequality; sensitive to the assumption that every resident shares the national CC.

Inputs: CC per economy, Population per economy
Sources: Gini (1912), Theil (1967), Lorenz (1905)

Assumptions

• Each economy's CC is treated as the cognitive-capital level of its people (no within-country variation in v1).

Caveats

• Measures inequality BETWEEN economies, not within them.
• Sensitive to the set of economies included.

CSI · Cognitive Separation Index

Measures: The Cognitive Separation Index is a simple gap measure: how much more AI capacity the world's most-equipped people have compared with the bottom half. A bigger number means a wider gulf between the top and the masses.How far the cognitively richest pull ahead of the rest.A top-to-bottom ratio measure of cognitive stratification: the mean Cognitive Coefficient of the top decile of the world population relative to the mean of the bottom half — a percentile-share inequality metric complementary to the Gini-based CII. ratio (e.g. 2.9×)

We rank all of the world's people by their AI/cognitive capacity. We take the average capacity of the richest 10% of people and divide it by the average of the poorest 50%. If the top group has, say, eight times as much, the index is 8. The higher the ratio, the more separated the haves are from the have-nots.

CSI is the ratio of the population-weighted mean CC of the top decile of people to that of the bottom half:

CSI = mean_CC(top 10% of people) / mean_CC(bottom 50% of people)

Economies are sorted by CC; population is accumulated to form the slices (a partial country is split proportionally), and the top and bottom slices are drawn from disjoint ends so no person is double-counted. We also report top-1% ÷ bottom-50% and top-10% ÷ bottom-40%.

Order the global population by per-person CC; CSI = mean(CC | top 10% of people) / mean(CC | bottom 50% of people), a P90-group / P50-group style inter-group ratio. Properties: unbounded above (≥1), interpretable as a 'how many times' multiple and more legible than a Gini; like CII it is built on national-mean CC weighted by population, so it captures between-country separation only and is insensitive to redistribution occurring entirely within the top decile or within the bottom half; robust to the middle of the distribution by construction.

Inputs: CC per economy, Population per economy
Sources: Top/bottom share ratios (income-distribution literature)

Assumptions

• People within an economy share its CC.

Caveats

• A ratio, not a 0–1 index.
• Cross-country only.

CV · Cognitive Velocity

Measures: Cognitive Velocity is a forward-looking guess at how fast a country is likely to improve each year. It's an estimate of speed, not a measured track record of past growth.A forward heuristic for an economy's rate of cognitive-infrastructure improvement.A forward-looking heuristic projected rate of improvement (% per year) of cognitive-economy standing, constructed from momentum and convergence (catch-up) headroom — explicitly a model-based expectation, not an estimated historical trend. %/yr (model heuristic)

We combine two ideas into a yearly percentage. First, momentum: a country already investing heavily and building pipelines is set to keep moving. Second, catch-up room: a country that is behind has more easy room to grow quickly, while a country near the top has less headroom. We blend these into a single 'percent per year' estimate. It's a rule-of-thumb projection, so treat it as a heuristic rather than a fact.

CV is a forward heuristic, not a measured trend (v1 has no historical panel). It scales a baseline rate by momentum and catch-up headroom:

CV = 2.0 · (0.5 + 1.6·CM/100) · (1 + 0.9·(1 − CC/100))

Higher Cognitive Momentum (CM) raises the rate; lower CC adds catch-up headroom. The code's own docstring states this is not a measured time series.

CV ≈ f(momentum, headroom), e.g. CV = α·(momentum signal) + β·(frontier_gap), where headroom is a decreasing function of current level (frontier − level), embedding a conditional-convergence prior (laggards have larger β·gap). Properties: it is a constructed forecast, not a regression slope or realized CAGR, so it carries no statistical confidence interval and is sensitive to the chosen α, β and to how 'frontier' and momentum are operationalized; the catch-up term mechanically inflates expected velocity for low-level countries, which should be read as potential, not guaranteed, growth.

Inputs: Cognitive Momentum (CM), CC (for headroom)
Sources: Internal heuristic

Assumptions

• Momentum and headroom proxy near-term improvement in lieu of a fitted trend.

Caveats

• Not estimated from history; no standard error.
• A structural heuristic, not a forecast.

CM · Cognitive Momentum

Measures: Cognitive Momentum is a snapshot of the forces pushing a country forward — the fuel in the tank for future AI growth, rather than where it stands today.Forward signal that improvement is likely to be sustained or accelerate.A leading-indicator composite of growth potential in the cognitive economy, aggregating forward-looking flow/pipeline signals (AI-investment intensity, education pipeline, research intensity, infrastructure pipeline) as distinct from stock/level measures of current standing. 0–100

We combine four forward-pointing signals: how intensely the country is investing in AI, how strong its education pipeline is, how much it spends on research, and how fast it's building infrastructure. Each is scored on the same scale and blended into one momentum number. A high score means lots of forward push is already in motion.

CM is the mean of four normalized forward signals:

1. Investment intensity — private AI investment ÷ GDP (log-scaled, normalized)
2. Education pipeline — tertiary enrollment, STEM share, AI-skill penetration
3. Research intensity — R&D/GDP, researchers per million
4. Infrastructure pipeline — data-center capacity, cloud regions

CM = mean(invest, pipeline, research, infrastructure)

Normalize each of the four signals to a common 0-100 scale and aggregate as a (weighted) arithmetic mean CM = Σ w_q · ñ_q. Properties: compensatory aggregation of intensity/flow variables chosen for their predictive rather than contemporaneous content; it is an input/driver index, so it feeds forward-looking constructs (e.g. it is a primary component of Cognitive Velocity) and should not be interpreted as realized growth; validity hinges on the assumption that these four flows are leading indicators of future cognitive-economy gains.

Inputs: AI investment ÷ GDP, Tertiary / STEM / AI-skill, R&D/GDP, researchers, Data-center, cloud regions
Sources: Stanford AI Index, OECD, UNESCO

Assumptions

• Equal weight across the four signals.

Caveats

• A present-day composite of forward-leaning inputs, not a fitted acceleration.

AIR · AI Readiness Index

Measures: AI Readiness measures how prepared a country's institutions and infrastructure are to actually take advantage of AI — the rules, skills, and plumbing that need to be in place before AI can deliver.Institutional and infrastructural preparedness to deploy AI.An institutional-and-infrastructural preparedness construct capturing enabling-environment capacity for AI uptake across governmental readiness, governance/regulatory frameworks, data governance, talent supply, and compute. 0–100

We score five things: how ready the government is, whether good governance and rules exist, how data is managed, the supply of skilled talent, and available computing power. Each gets a 0-100 score, and we combine them into one readiness score. A high score means the foundations are in place.

AIR averages a governance/readiness block with an infrastructure block:

• Governance/readiness = weighted mean of gov AI readiness (×1.5), AI governance framework, data governance, AI talent.
• Infrastructure = mean of normalized Top500, data-center MW, semiconductor capability.

AIR = mean(governance_readiness, infrastructure)

Normalize each of the five sub-pillars (government readiness, governance frameworks, data governance, talent, compute) to 0-100 and aggregate as a (weighted) arithmetic mean AIR = Σ w_r · ñ_r. Properties: compensatory composite mixing institutional/soft factors (governance, data rules) with hard factors (talent, compute), so it characterizes potential-to-adopt rather than realized utilization (cf. AIX); sensitive to weighting between institutional and infrastructural components and to the measurability/proxy quality of governance constructs.

Inputs: Gov AI readiness, AI governance framework, Data governance, AI talent, Top500 / data-center / semiconductor
Sources: Oxford Insights, OECD.AI, Top500

Assumptions

• Readiness and compute infrastructure weighted equally at the block level.

Caveats

• 'Governance framework' scores maturity, not restrictiveness.

AIX · Augmentation Index

Measures: The Augmentation Index measures how much AI is actually being used to boost people's work — real usage on the ground, not just whether the tools are available.Actual utilization of AI — augmentation depth — rather than mere access.A realized-utilization construct measuring actual AI augmentation of human work — workforce AI-skill use and talent depth — deliberately distinguished from access/availability and readiness measures. 0–100

We focus on what's actually happening: how much the workforce uses AI skills in their jobs and the depth of AI talent. These are scored and combined into a single number. The key point is that this counts real use, so a country with lots of access but little actual use scores lower than one where people genuinely put AI to work.

AIX is a weighted mean of utilization proxies (NOT an O*NET task pipeline — that is used for OFI):

AIX = wmean( AI-skill penetration ×2, AI talent ×1, national AI capacity ×1 )

Weighting AI-skill penetration most heavily reflects that AIX is about use, not readiness or access.

Normalize the utilization components (workforce AI-skill usage, talent) to 0-100 and aggregate as a (weighted) mean AIX = Σ w_s · ñ_s. Properties: an outcome/usage index rather than an enabling-environment index, so it can diverge sharply from AIR (high readiness, low augmentation indicates an adoption gap); validity depends on usage proxies (e.g. skill-penetration or tool-usage signals) that may under-measure informal or unobserved use and may be biased toward digitally legible sectors.

Inputs: AI-skill penetration, AI talent, National AI capacity score
Sources: LinkedIn / Stanford AI Index, Tortoise

Assumptions

• Workforce AI-skill penetration is the best available proxy for utilization depth in v1.

Caveats

• A proxy for utilization; direct per-firm adoption telemetry is a roadmap item.
• Distinct from AIR (readiness) by construction.

CRI · Cognitive Resilience Index

Measures: Cognitive Resilience measures how well a country could stand on its own in AI — how much it relies on its own capabilities rather than depending on AI infrastructure controlled by other countries.Resilience to cognitive concentration — self-reliance plus a broad capability base.A self-sufficiency/robustness construct measuring resilience to external dependency on frontier AI infrastructure, combining domestic AI capability (self-reliance) with diversification/breadth factors (education, data governance). 0–100

We combine two ideas: self-reliance (how much capable AI a country can build and run itself) and breadth (the supporting strengths like education and good data rules that let it adapt). Each part is scored and blended into one resilience number. A high score means a country is less likely to be left stranded if it loses access to outside frontier AI.

CRI blends domestic capability with breadth:

CRI = 0.6 · domestic_capability + 0.4 · breadth

• domestic_capability = mean normalized [national AI capacity, notable models, semiconductor, AI talent]
• breadth = mean normalized [tertiary enrollment, human capital, data governance, internet]

Higher CRI = more able to rely on home-grown capability across a broad base, hence less fragile to externally-concentrated frontier infrastructure. (There is no HHI or shared Monte-Carlo machinery — CRI is this deterministic blend.)

Aggregate a self-reliance term (domestic AI capability) with a breadth term (normalized education and data-governance scores), e.g. CRI = w_1·(self-reliance) + w_2·(breadth), each component on 0-100. Properties: frames resilience as low external-dependency plus adaptive capacity; the breadth term proxies the capacity to substitute or recover if external frontier access is cut. Limitation: it infers resilience from capability/breadth stocks rather than from observed shock response, so it is a structural-vulnerability proxy, not a stress-tested measure; weights encode how much self-reliance vs. breadth is deemed protective.

Inputs: National AI capacity, notable models, semiconductor, AI talent, Tertiary, human capital, data governance, internet
Sources: Stanford AI Index, World Bank HCI, OECD

Assumptions

• Self-reliance (60%) and breadth (40%) jointly proxy resilience.

Caveats

• A structural proxy; does not model specific supply-chain or vendor dependencies.

CEV · Cognitive Escape Velocity

Measures: Cognitive Escape Velocity describes a country's trajectory relative to the cutting edge — whether it's pulling ahead, keeping pace, or slipping behind. It sorts countries into bands from 'Frontier' down to 'Falling behind'.An economy's trajectory relative to the frontier — closing the relative gap, holding, or slipping.A categorical trajectory classification relative to the moving frontier, binning countries into ordinal states (Frontier / Accelerating / Advancing / Maintaining / Falling behind) by their proportional growth pace versus the frontier's pace. category + signed %

We look at how fast a country is improving compared with how fast the leaders are moving. If it's growing faster than the frontier, it's 'Accelerating' or at the 'Frontier'; if it's growing about as fast, it's 'Maintaining'; if it's growing slower, it's 'Falling behind'. We turn that pace comparison into one of a few labeled categories.

CEV classifies on proportional growth pace vs. the frontier (the highest-CC economy):

ratio = CV(country) / CV(frontier)

• Frontier — the highest-CC economy itself
• Accelerating — ratio ≥ 1.05
• Advancing — ratio ≥ 0.97
• Maintaining — ratio ≥ 0.82
• Falling behind — ratio < 0.82

The reported score is (ratio − 1)·100 — percent pace relative to the frontier. Because CV is a heuristic, CEV is a relative-trajectory classification, not a probability.

Compute a relative growth rate (own proportional improvement pace vs. the frontier's pace) and apply threshold cutoffs to assign an ordinal band: pace > frontier ⇒ Accelerating/Frontier, pace ≈ frontier ⇒ Maintaining, pace < frontier ⇒ Falling behind (with 'Advancing' an intermediate band). Properties: a discretization of a continuous relative-velocity signal, so it is interpretable and convergence-aware (it answers 'is the gap to the frontier closing or widening?') but is threshold-sensitive and loses within-band resolution; depends on how the frontier and proportional pace are defined and on the placement of the band cutoffs.

Inputs: CV(country), CV(frontier)
Sources: Internal (derived from CV)

Assumptions

• Relative position improves only if you grow faster in %/yr than the frontier.

Caveats

• Inherits CV's heuristic nature; not fitted to history.

OFI · Occupational Future Index

Measures: The Occupational Future Index measures how much AI is likely to reshape the tasks within a specific job — how much of the work gets automated, how much gets AI-assisted, and how much stays human. It is about how the job changes, not about whether the job disappears.Task-transformation pressure on an occupation from AI — not job loss.A task-level transformation-pressure construct for an occupation: the degree to which AI shifts an occupation's task bundle, decomposed into automate / augment / remain-human shares — explicitly a measure of task reconfiguration, not employment displacement. 0–100

We break a job down into its individual tasks, then sort each task into one of three buckets: tasks AI can take over, tasks AI helps a human do better, and tasks that stay fully human. We then summarize how much of the job falls into each bucket to gauge the overall transformation pressure. A high score means lots of the job's tasks are set to change, not that the job is being lost.

For each occupation:

pressure = exposure · (0.4·adoption_risk + 0.3·economic_pressure + 0.3·agentic_exposure) friction = (1 − 0.5·regulatory_friction) · (1 − 0.45·trust_requirement) OFI = 100 · clamp₀¹(pressure · friction)

Task shares (automate / augment / human) come from the occupation panel and are projected to 2030/35/40 via Bass diffusion of the agentic-automation ceiling, damped by regulatory/trust friction. Scores are grounded in published exposure frameworks.

Decompose an occupation into its constituent tasks, classify/score each task on its AI exposure into automate, augment, or remain-human, then aggregate (typically a task-importance- or time-weighted share) to a transformation-pressure score for the occupation. Properties: a bottom-up, task-based exposure measure in the O*NET-style tradition, separating exposure from outcome — high transformation pressure implies task restructuring (and possible productivity/skill shifts), not net job loss, since augmentation and reallocation can offset automation; sensitive to the task taxonomy, the per-task exposure scoring, and the weighting used to roll tasks up to the occupation level.

Inputs: exposure, adoption_risk, economic_pressure, agentic_exposure, regulatory_friction, trust_requirement, task_shares
Sources: Felten–Raj–Seamans AIOE, Eloundou et al. 2023 (GPTs are GPTs), OECD AI exposure, O*NET

Assumptions

• Exposure measures task overlap with AI capability, not realized automation.
• Bass diffusion governs adoption of the agentic ceiling.

Caveats

• Models task transformation, NOT employment outcomes.
• Occupation-level, US-anchored employment/wage context.

MPI · Metro Power

Measures: A metro area's ABSOLUTE cognitive capacity — its total weight in the AI economy.A metro area's ABSOLUTE cognitive capacity — its total weight in the AI economy.A metro area's ABSOLUTE cognitive capacity — its total weight in the AI economy. 0–100

Weighted mean of log-normalized absolute-mass components (per-capita inputs are multiplied by population first, then log-scaled and min-max normalized to 5–100):

Component	Weight
AI investment	2.0
Notable AI orgs	1.5
Talent mass	1.5
Tech workforce	1.2
Economic mass (GDP)	1.0
Compute mass	1.0

The metro analogue of CPI.

Weighted mean of log-normalized absolute-mass components (per-capita inputs are multiplied by population first, then log-scaled and min-max normalized to 5–100):

Component	Weight
AI investment	2.0
Notable AI orgs	1.5
Talent mass	1.5
Tech workforce	1.2
Economic mass (GDP)	1.0
Compute mass	1.0

The metro analogue of CPI.

Weighted mean of log-normalized absolute-mass components (per-capita inputs are multiplied by population first, then log-scaled and min-max normalized to 5–100):

Component	Weight
AI investment	2.0
Notable AI orgs	1.5
Talent mass	1.5
Tech workforce	1.2
Economic mass (GDP)	1.0
Compute mass	1.0

The metro analogue of CPI.

Inputs: AI investment, notable AI orgs, AI talent, tech employment, GDP, data centers, population
Sources: Curated v1; EU metros: Eurostat metropolitan datasets

Caveats

• Curated estimates for most metros; EU metro GDP/population are real (Eurostat).

MCC · Metro Coefficient

Measures: A metro's PER-CAPITA cognitive intensity — depth per resident.A metro's PER-CAPITA cognitive intensity — depth per resident.A metro's PER-CAPITA cognitive intensity — depth per resident. 0–100

The arithmetic mean of the five metro pillars (Talent, Capital, Research, Infrastructure, Agentic), each itself a weighted mean of normalized indicators. The metro analogue of CC.

The arithmetic mean of the five metro pillars (Talent, Capital, Research, Infrastructure, Agentic), each itself a weighted mean of normalized indicators. The metro analogue of CC.

The arithmetic mean of the five metro pillars (Talent, Capital, Research, Infrastructure, Agentic), each itself a weighted mean of normalized indicators. The metro analogue of CC.

Inputs: Talent, Capital, Research, Infrastructure, Agentic pillars
Sources: Curated v1

Caveats

• Per-capita intensity — rewards dense hubs over large ones.

MDI · Metro Agentic

Measures: How much a metro deploys autonomous AI agents.How much a metro deploys autonomous AI agents.How much a metro deploys autonomous AI agents. 0–100

The Agentic pillar: a weighted mean of agent adoption (×1.5) and AI-talent depth (×1.0).

The Agentic pillar: a weighted mean of agent adoption (×1.5) and AI-talent depth (×1.0).

The Agentic pillar: a weighted mean of agent adoption (×1.5) and AI-talent depth (×1.0).

Inputs: agent adoption, AI talent
Sources: Curated v1

Caveats

• Curated estimate.

SPI · State Power

Measures: A US state's ABSOLUTE cognitive capacity.A US state's ABSOLUTE cognitive capacity.A US state's ABSOLUTE cognitive capacity. 0–100

Same construction as Metro Power (MPI) over the 50 states + DC: weighted mean of log-normalized absolute mass (AI investment ×2.0, notable AI orgs ×1.5, talent mass ×1.5, tech workforce ×1.2, GSP ×1.0, compute ×1.0).

Same construction as Metro Power (MPI) over the 50 states + DC: weighted mean of log-normalized absolute mass (AI investment ×2.0, notable AI orgs ×1.5, talent mass ×1.5, tech workforce ×1.2, GSP ×1.0, compute ×1.0).

Same construction as Metro Power (MPI) over the 50 states + DC: weighted mean of log-normalized absolute mass (AI investment ×2.0, notable AI orgs ×1.5, talent mass ×1.5, tech workforce ×1.2, GSP ×1.0, compute ×1.0).

Inputs: AI investment, AI orgs, talent, tech employment, GSP, population
Sources: Curated v1; population: US Census ACS

Caveats

• Population is real (Census ACS); GSP & AI fields curated.

SCC · State Coefficient

Measures: A US state's PER-CAPITA cognitive intensity.A US state's PER-CAPITA cognitive intensity.A US state's PER-CAPITA cognitive intensity. 0–100

Mean of the five state pillars (Talent, Capital, Research, Infrastructure, Agentic). The state analogue of CC/MCC.

Mean of the five state pillars (Talent, Capital, Research, Infrastructure, Agentic). The state analogue of CC/MCC.

Mean of the five state pillars (Talent, Capital, Research, Infrastructure, Agentic). The state analogue of CC/MCC.

Inputs: five state pillars
Sources: Curated v1; population: US Census ACS

Caveats

• Per-capita intensity.

SDI · State Agentic

Measures: How much a US state deploys autonomous AI agents.How much a US state deploys autonomous AI agents.How much a US state deploys autonomous AI agents. 0–100

The Agentic pillar (agent adoption ×1.5 + AI talent ×1.0).

The Agentic pillar (agent adoption ×1.5 + AI talent ×1.0).

The Agentic pillar (agent adoption ×1.5 + AI talent ×1.0).

Inputs: agent adoption, AI talent
Sources: Curated v1

Caveats

• Curated estimate.

ITI · Industry Transformation Index

Measures: How strongly AI is reshaping an industry.How strongly AI is reshaping an industry.How strongly AI is reshaping an industry. 0–100

Weighted mean of four 0–100 drivers:

Driver	Weight
AI exposure	1.4
AI adoption	1.2
Productivity uplift	1.0
Growth outlook	0.8

Each industry also carries an automate / augment / human task split that sums to 100.

Weighted mean of four 0–100 drivers:

Driver	Weight
AI exposure	1.4
AI adoption	1.2
Productivity uplift	1.0
Growth outlook	0.8

Each industry also carries an automate / augment / human task split that sums to 100.

Weighted mean of four 0–100 drivers:

Driver	Weight
AI exposure	1.4
AI adoption	1.2
Productivity uplift	1.0
Growth outlook	0.8

Each industry also carries an automate / augment / human task split that sums to 100.

Inputs: AI exposure, adoption, productivity uplift, growth outlook
Sources: Curated; grounded in published AI-exposure frameworks

Caveats

• Models task transformation, not employment loss.

CPX · Corporate Power Index

Measures: A company's ABSOLUTE power to shape global affairs.A company's ABSOLUTE power to shape global affairs.A company's ABSOLUTE power to shape global affairs. 0–100

Weighted mean of log-normalized absolute mass plus a linear strategic-leverage term:

Component	Weight	Transform
Market cap	2.5	log, min-max
Revenue	1.2	log, min-max
Employees	0.8	log, min-max
Strategic leverage	1.5	linear 0–100

The corporate analogue of CPI.

Weighted mean of log-normalized absolute mass plus a linear strategic-leverage term:

Component	Weight	Transform
Market cap	2.5	log, min-max
Revenue	1.2	log, min-max
Employees	0.8	log, min-max
Strategic leverage	1.5	linear 0–100

The corporate analogue of CPI.

Weighted mean of log-normalized absolute mass plus a linear strategic-leverage term:

Component	Weight	Transform
Market cap	2.5	log, min-max
Revenue	1.2	log, min-max
Employees	0.8	log, min-max
Strategic leverage	1.5	linear 0–100

The corporate analogue of CPI.

Inputs: market cap, revenue, employees, strategic leverage
Sources: Public financials; market caps refreshed nightly (Yahoo)

Caveats

• Not investment advice.

CAI · Corporate AI Index

Measures: A company's internal AI posture — how AI-mature it is.A company's internal AI posture — how AI-mature it is.A company's internal AI posture — how AI-mature it is. 0–100

Weighted mean of five 0–100 fields:

Field	Weight
AI talent	1.3
AI in products	1.2
AI-stack independence	1.2
AI adoption	1.1
AI governance	0.6

AI-stack independence rewards owning your own frontier models + AI silicon + cloud (e.g. Google: Gemini + TPU + GCP).

Weighted mean of five 0–100 fields:

Field	Weight
AI talent	1.3
AI in products	1.2
AI-stack independence	1.2
AI adoption	1.1
AI governance	0.6

AI-stack independence rewards owning your own frontier models + AI silicon + cloud (e.g. Google: Gemini + TPU + GCP).

Weighted mean of five 0–100 fields:

Field	Weight
AI talent	1.3
AI in products	1.2
AI-stack independence	1.2
AI adoption	1.1
AI governance	0.6

AI-stack independence rewards owning your own frontier models + AI silicon + cloud (e.g. Google: Gemini + TPU + GCP).

Inputs: AI adoption, AI talent, AI products, AI-stack independence, AI governance
Sources: Public disclosures; curated

Caveats

• Curated judgement of AI posture.

AII · AI Influence Index

Measures: How much a company shapes the AI world (the 'Power AI 200' ranking) — size is only one factor.How much a company shapes the AI world (the 'Power AI 200' ranking) — size is only one factor.How much a company shapes the AI world (the 'Power AI 200' ranking) — size is only one factor. 0–100

A deliberately size-de-emphasized composite:

Factor	Weight
Frontier capability	1.5
Ecosystem criticality	1.5
Leadership influence	1.3
AI momentum	1.2
AI-stack independence	1.0
Strategic leverage	0.8
Market cap (log-norm)	0.7

Ecosystem criticality captures the chokepoints (TSMC, ASML, Nvidia) and the foundational models everyone builds on (OpenAI, Anthropic). Leadership influence credits the company's key AI figures.

A deliberately size-de-emphasized composite:

Factor	Weight
Frontier capability	1.5
Ecosystem criticality	1.5
Leadership influence	1.3
AI momentum	1.2
AI-stack independence	1.0
Strategic leverage	0.8
Market cap (log-norm)	0.7

Ecosystem criticality captures the chokepoints (TSMC, ASML, Nvidia) and the foundational models everyone builds on (OpenAI, Anthropic). Leadership influence credits the company's key AI figures.

A deliberately size-de-emphasized composite:

Factor	Weight
Frontier capability	1.5
Ecosystem criticality	1.5
Leadership influence	1.3
AI momentum	1.2
AI-stack independence	1.0
Strategic leverage	0.8
Market cap (log-norm)	0.7

Ecosystem criticality captures the chokepoints (TSMC, ASML, Nvidia) and the foundational models everyone builds on (OpenAI, Anthropic). Leadership influence credits the company's key AI figures.

Inputs: frontier capability, ecosystem criticality, leadership influence, AI momentum, AI-stack, strategic leverage, market cap
Sources: Curated, synthesised from public information

Caveats

• Judgement-heavy; not a precise measurement.

PII · Power & Influence Index

Measures: A person's overall ability to shape the course of civilization.A person's overall ability to shape the course of civilization.A person's overall ability to shape the course of civilization. 0–100

A weighted blend of five power axes, then scaled by entrenchment:

Axis	Weight
Political power	1.3
Economic power	1.2
AI power	1.2
Institutional power	1.0
Platform power	0.9

base = Σ(wᵢ·axisᵢ)/Σwᵢ. An entrenchment factor then scales it within a ±25% band based on tenure security (0–100): tf = 0.75 + 0.25·(tenure/100) — so a term-limited leader (low tenure) is discounted and an entrenched one (Xi ≈100) is not. PII = base · tf.

A weighted blend of five power axes, then scaled by entrenchment:

Axis	Weight
Political power	1.3
Economic power	1.2
AI power	1.2
Institutional power	1.0
Platform power	0.9

base = Σ(wᵢ·axisᵢ)/Σwᵢ. An entrenchment factor then scales it within a ±25% band based on tenure security (0–100): tf = 0.75 + 0.25·(tenure/100) — so a term-limited leader (low tenure) is discounted and an entrenched one (Xi ≈100) is not. PII = base · tf.

A weighted blend of five power axes, then scaled by entrenchment:

Axis	Weight
Political power	1.3
Economic power	1.2
AI power	1.2
Institutional power	1.0
Platform power	0.9

base = Σ(wᵢ·axisᵢ)/Σwᵢ. An entrenchment factor then scales it within a ±25% band based on tenure security (0–100): tf = 0.75 + 0.25·(tenure/100) — so a term-limited leader (low tenure) is discounted and an entrenched one (Xi ≈100) is not. PII = base · tf.

Inputs: economic / political / AI / platform / institutional power, tenure security
Sources: Curated, synthesised from public information

Caveats

• A model of relative power, not an endorsement or precise measurement.

AAI · AI Advancement Index

Measures: How much a person is advancing artificial intelligence itself — the researchers, lab leaders and thought leaders driving where the field goes. Distinct from PII (political/economic power); this is the 'AI Minds 500' ranking.How much a person is advancing artificial intelligence itself — the researchers, lab leaders and thought leaders driving where the field goes. Distinct from PII (political/economic power); this is the 'AI Minds 500' ranking.How much a person is advancing artificial intelligence itself — the researchers, lab leaders and thought leaders driving where the field goes. Distinct from PII (political/economic power); this is the 'AI Minds 500' ranking. 0–100

A weighted blend of five contribution dimensions:

Dimension	Weight
Research influence	1.4
Frontier role	1.3
Thought leadership	1.1
Momentum	1.1
Field-building	1.0

AAI = Σ(wᵢ·dimᵢ)/Σwᵢ, each dimension 0–100. Research influence is weighted highest so the index rewards advancing the field, not wealth or office — which is why working researchers (Hinton, Bengio, LeCun) rank alongside frontier-lab leaders (Hassabis, Amodei). Momentum keeps it current: a foundational pioneer who has stepped back from active work ranks below where raw research influence alone would place them.

A weighted blend of five contribution dimensions:

Dimension	Weight
Research influence	1.4
Frontier role	1.3
Thought leadership	1.1
Momentum	1.1
Field-building	1.0

AAI = Σ(wᵢ·dimᵢ)/Σwᵢ, each dimension 0–100. Research influence is weighted highest so the index rewards advancing the field, not wealth or office — which is why working researchers (Hinton, Bengio, LeCun) rank alongside frontier-lab leaders (Hassabis, Amodei). Momentum keeps it current: a foundational pioneer who has stepped back from active work ranks below where raw research influence alone would place them.

A weighted blend of five contribution dimensions:

Dimension	Weight
Research influence	1.4
Frontier role	1.3
Thought leadership	1.1
Momentum	1.1
Field-building	1.0

AAI = Σ(wᵢ·dimᵢ)/Σwᵢ, each dimension 0–100. Research influence is weighted highest so the index rewards advancing the field, not wealth or office — which is why working researchers (Hinton, Bengio, LeCun) rank alongside frontier-lab leaders (Hassabis, Amodei). Momentum keeps it current: a foundational pioneer who has stepped back from active work ranks below where raw research influence alone would place them.

Inputs: research influence, frontier role, thought leadership, field-building, momentum
Sources: Curated, synthesised from public information

Caveats

• A model of relative contribution to AI's advancement, not a precise measurement.
• Scores are estimates; the long tail of working researchers is necessarily incomplete.

Limitations & ethics

Honest limits and ethics

Please read these scores as relative, not absolute. They tell you how countries stack up against each other in this dataset — not some true, fixed measure of worth. Change the set of countries and the numbers shift.

A few things to keep in mind:

• We weight most pieces equally. That's a deliberate, simple starting choice, not a claim that everything matters the same. Different weights would give different rankings.
• Scenarios aren't odds. A future scenario shows "what if," not "how likely." Don't read a 60-ish number as a 60% chance of anything.
• Averages hide gaps inside a country. A high national score can sit on top of deep inequality — a few hubs thriving while many people are left out. The single number can't show that.
• The "aristocracy" framing is a warning, not a forecast. When we flag the risk of a small cognitive elite pulling away, we're naming something to watch and prevent, not predicting it will happen.

Use these numbers to ask better questions, not to settle them.

Relative, not absolute. CC, AIR, AIX, CRI and the pillars are min-max normalized within the current panel. A country's score is its standing relative to the other 47 economies, not an absolute physical quantity. Adding or removing economies can shift scores. CDI (geometric mean of normalized dimensions, 0–1) is likewise panel-relative.

Equal weights are a choice. Pillar weights within CC are an explicit, equal-ish default (member-indicator weights are shown per pillar). We do not claim weights are empirically optimal or derived from PCA/expert elicitation. Sensitivity to weighting is a known limitation; a published weight-sensitivity analysis is on the roadmap, not in v1.

Temporal indices are heuristics. CV, CM and CEV are computed from a single cross-section. They encode plausible forward structure (momentum, headroom, pace vs. the frontier) but are not fitted to historical data and carry no standard errors. Historical trajectories shown in charts are explicitly labelled illustrative reconstructions.

Scenarios are not probabilities. The 8 scenarios are parameterized assumption-sets. The Monte-Carlo bands quantify parameter uncertainty within a scenario; they are not calibrated probabilities of real-world futures, and the scenarios are not assigned likelihoods.

Occupational estimates are structural, not predictive. OFI and its task-share projections are grounded in published AI-exposure frameworks (AIOE; Eloundou et al. 2023; OECD), but exposure is not job loss. Adoption depends on cost, workflow, regulation, and trust — which we model coarsely.

Aggregation hides within-country inequality. CII/CSI use cross-country, population-weighted distributions of CC. They capture inequality between economies, not within them (v1 lacks subnational microdata).

Ethics. The platform exists to make cognitive-infrastructure inequality visible so it can be addressed. The "cognitive aristocracy" framing names a risk to monitor, not a desired or predicted outcome. Personal assessments are reflective instruments over self-reported inputs, computed in-session with no storage — never a judgement of a person.

Limitations and ethics

Relativity. All indices are relative, defined with respect to the sampled set of units and the chosen normalization bounds. They support ordinal and comparative claims, not absolute or cardinal welfare statements; adding or removing units re-anchors the scale.

Weighting. v1 uses equal weights within and across components. This is a transparent prior, not a substantive claim of equal importance; it sidesteps unidentified weight estimation but is consequential — rankings are sensitive to it, and alternative weightings (expert-elicited, PCA-derived, or policy-specific) can materially reorder units. Sensitivity analysis is encouraged.

Scenarios are not probabilities. Forward values are conditional constructs under explicit assumptions and carry no probability mass. Treating a scenario score as a likelihood, expectation, or risk estimate is a category error.

Aggregation masks within-unit dispersion. National aggregates — especially per-capita CC — average over potentially extreme internal inequality; high scores can coexist with severe exclusion concentrated in a few hubs. The platform measures between-unit structure, not within-unit distribution.

The aristocracy framing. Naming the risk of an emergent cognitive aristocracy is a normative flag for monitoring, identifying a failure mode worth governing against — not a prediction that it will materialize.