Training Orchestration Inventory
This page records the current state of the training boundary between Python and
Rust.
The inventory is intentionally conservative:
- the Rust core owns the portable
ModelSpec evaluation surface
- Python routes supported training and replay through Rust first and only
owns fallback preprocessing or unsupported cases when Rust cannot handle a
request
- the remaining parity gaps stay listed here until they are fully closed
See Rust Core and Training Core Migration.