Which of ten middle powers hold something frontier AI developers cannot easily replace, and which small groups of them hold enough of it to set a condition together. Every cell has a source link, a confidence flag and a reason.
The question is not only which middle powers hold important AI resources. It is which of them can realistically turn those resources into coordinated influence over AI governance.
Read before using. This is a first pass built in about a day from public sources. Only two cells reach the top score, and both are chips. All Algorithms cells score 1. One cell has no score because no source was found (hatched). Some figures come from secondary summaries and are marked Single source or Unverified. Treat the coalition ranking as a way to ask questions, not as a finding.
How this works
Choose a view. Strategic coverage shows the original scores. Conditionability-adjusted applies weights for how far each lever can be tied to a condition. Country contribution shows who is strongest in each layer. Evidence confidence shows how well each cell is sourced.
Check the evidence. Select any cell or name to see what the country holds, why it got that score, and the source link.
Look for groups. In the first two views, the coalition finder ranks groups of 3 or 4 countries. Select a row, or press Play tour, to see the group's links and which member holds each layer.
Test the result. Tick the box to score unverified cells as 1 and see whether the top group changes.
The map
Exploratory measure. The adjusted score is a rough way to ask which levers could realistically be turned into a condition. It is not an established index of diplomatic power. The weights below are proposed defaults, the Conditionable flags are first-pass judgments, and both can be wrong.
Coalition finder
Pick the countries that might act together, choose a group size, and see which groups cover the most layers. Coverage of a layer is the highest score any member holds in it.
Select a row to show that group. Each bar is the highest score any member holds in that layer.
What this does and does not show
Sources
Each cell links its own source. Calibration notes record seven score changes made after the first pass so that similar evidence got similar scores. The rubric, rules and change log are in the companion doc.