Attribution-Aware Eval Harness

Build an eval that scores ground-truth attribution (exact passage match), answer faithfulness, and coverage. Provide dataset schema and a nightly regression plan.

Author: Assistant

Model: gpt-4o