Back to indexComparisonMar 30, 20264 min

Comparing Countries Without Getting Lost in 3D

A compact routine for keeping comparisons readable, moving between countries, and grounding visual impressions in real constitutional text.

Open atlas Official ConstitutionalMap AI tutorial playlist Country Selection and Semantic Map View

Keep the comparison small at first

Large country sets usually make the first reading harder, not better. A pair or a small group is enough to see whether similarity is broad or concentrated in a few thematic clusters.

Use one country as an anchor and change only one comparison variable at a time.

A reliable comparison loop

The statistics panel and the article detail panel are there to slow you down in a good way. Use them to check whether the visual pattern is supported by the underlying text.

Select two countries and review their coverage and entropy.
Open one dense cluster and inspect two or three articles.
Use the comparison flow to validate the initial visual impression.
Swap one country and repeat the same sequence.

How to read the visualization

What You Are Seeing

Each point in the 3D view represents a constitutional article or other meaningful legal unit. Nearby points are not nearby because they come from the same country, but because the language of those passages is semantically similar. Use country selection to compare how different constitutions occupy the same semantic terrain.

The map and the country list are selection tools. They decide which constitutions are loaded into the scene. Selecting more countries does not change the geometry of the embedding itself; it changes which parts of that global semantic space you can inspect.

Semantic Space

The embedding turns legal text into vectors, and the clustering step groups vectors that tend to discuss related constitutional themes. In country mode, color shows political origin. In cluster mode, color shows thematic neighborhood.

Large, dense clouds usually indicate recurring constitutional ideas such as rights, institutions, emergency powers, elections, or amendment rules. Isolated points often mark unusual provisions, rare wording, or country-specific constitutional design choices.

The platform offers two types of search: keyword search finds literal term occurrences, while semantic search retrieves conceptually nearby passages even without matching terms. Search results highlight regions of the semantic space in the 3D canvas, linking what you read to where it sits.

How to Read the Metrics

In the country statistics, Coverage measures how much of the global cluster landscape a constitution reaches. Entropy measures how evenly its segments are distributed across that landscape: high entropy suggests a broader semantic spread, while low entropy suggests concentration in fewer themes.

In the article detail panel, Global Cluster is the identifier of the thematic group assigned to that segment in the worldwide clustering. When the value is -1, it means the segment was left outside the defined thematic groupings in that global step. Probability indicates how confidently the clustering model placed that segment in that group: higher values mean a cleaner fit, while lower values usually mark more ambiguous or boundary cases.