Seems that the Google-style approach would be a good fit: mostly one codebase with a well-defined dependency graph, and build/test infrastructure that supports fast and comprehensive validation. This would seem to obviate the need for the catalog system described, but probably requires more investment in the build system.

Author here. The earnings call quote gets all the attention, but the engineering blog tells a different story. Their staff engineers talk about faster migrations, not replacement. The gap between the investor pitch and the eng reality is where the interesting bits are.