A safeguard turned out to be stronger than we expected

We wanted to verify that cross-tenant learning could not accidentally become approved global learning. The running system turned out to have no functioning apply-path at all.

What we learned: Sometimes the safest architecture is the absence of a capability until governance exists.