llvm-exegesis measures hardware behavior to validate LLVM's scheduling models. However, its dependency on libpfm's symbolic counter repository silently blocks entire classes of measurement. Counters that are undocumented, hidden by the vendors, or newly shipped remain unreadable through the standard pipeline, cutting off validation for the exact hardware used by the scheduling models. This talk presents new raw PMU encoding infrastructure in TableGen's PfmCounter class that lets per-CPU .td entries specify event and unit mask values directly. The presentation walks through the design, the demonstrated end-to-end path on Zen 2, and the roadmap this foundation opens toward a libpfm-independent llvm-exegesis and future cross-platform support.