CMS’s RADV extrapolation appeal is still unresolved. Here’s why risk-adjustment coding teams need agentic AI built for defensibility, not just speed.