This package now includes a runnable execution loop:
- choose an adapter
- run benchmark prompts
- score outputs
- promote a candidate into the scoreboard
- generate a starter quantization-retention report
It is still a starter path, not a finished frontier evaluation harness.