Identity
The same model version, company scope or survey population
What makes a comparison useful
Same question, clear boundaries and a traceable source
Measure
A stated unit and definition, kept separate from other measures
Period
Observation dates and capture dates shown separately
Coverage
Missing values, estimates and uncertainty stay visible
Read the measure before the ranking
Open the source register| Data | What we compare | What it cannot establish |
|---|---|---|
| Models | Same benchmark and configuration | An overlapping score interval is not a clear winner |
| Prices | Same input and output token workload | List-price arithmetic is not a guaranteed bill |
| Companies | Reported company or segment scope, with its own fiscal period | Funding and valuation are not revenue |
| Countries | Same survey population and year | European survey data is not a global ranking |
| Usage | The recorded platform and observation window | A leaderboard cohort is not global market share |
| Studies | The task, population and evidence type | Different experiments do not form one combined AI effect |
A gap is not zero
Missing observations stay missing. Lines connect recorded values; incomplete years are identified. Percentage-point changes and percentage changes keep their distinct units. Percentage change from a zero baseline is undefined.
A snapshot is not a release date
A catalogue observation records what was captured at that time. A change between two snapshots does not establish the exact time it took effect.
Geography describes coverage
Europe and the EU are different groupings. Multinational models can appear in more than one country or region. Country associations are not deployment locations or national AI-strength scores.
Natural Earth · geographic boundariesAccess is not a licence
Each dataset retains its source-specific reuse terms. Public APIs, company marks and original editorial material do not share one blanket licence.
Publication policy