Same MAR band. Dramatically different credit ratings.
Within every MAR band that holds enough rated institutions to compute a distribution, blended credit ratings span almost the entire ratings universe. A bank scored B for AI governance can sit anywhere from A-tier credit to BB-tier. A C-band peer can run from A-tier to CCC. The medians barely move across MAR bands. The dispersion within each band is enormous.
A B-rated bank by governance can be anywhere on the credit spectrum.
Each coloured rectangle is the inter-quartile range of blended credit ratings inside that MAR band. The horizontal line through each box is the median. The whiskers extend to the band minimum and maximum. The grand mean (B-tier credit) crosses the plot as a dashed gold reference. Hover or tap any box for the underlying statistics. The flat medians signal that governance posture does not push credit rating up or down on average. The wide spreads signal that any individual institution in a given MAR band could be rated almost anywhere by the agencies. The most populated band, D (n = 52), spans 16.5 notches: from the high-grade tier all the way down to CCC. Band C (n = 184) spans 15. Band B (n = 74) spans 10.5. These are not narrow distributions with a few outliers. The dispersion is the distribution.
The independence is structural, not directional.
Two variables can be uncorrelated in two ways. They can be reliably divergent: high MAR predicts low credit, or vice versa. Or they can be genuinely unrelated: a high-MAR institution could be anywhere on the credit spectrum, and the same in reverse. Within-band dispersion tells us which kind of uncorrelated MAR and credit ratings are. It is the second kind. Knowing an institution's MAR band gives you essentially no information about where it sits on the credit spectrum. Knowing its credit notch gives you essentially no information about its MAR band.
For buyers of governance assurance, this means a credit rating cannot be repurposed as a proxy for AI governance maturity, and an absence of credit downgrade cannot be read as a positive signal on governance posture. For supervisors, it means existing rating data carries no shadow signal on the AI-governance question even after careful stratification. For institutions seeking to demonstrate governance maturity, it means a strong credit rating does not relieve them of the obligation to demonstrate governance posture directly. The two ratings answer different questions and the answers are uncorrelated by construction, not by accident.