**custom benchmark metric**
Submissions are scored using a custom metric that combines:

1) **Answer Quality**  
   Token-level similarity against the gold answer (robust to minor phrasing differences)

2) **Citation Precision**  
   Are your cited chunks actually correct?

3) **Citation Recall**  
   Did you cite the necessary gold evidence?

4) **Grounding Consistency**  
   Does your answer stay faithful to the evidence you cited?

5) **Over-Citation Penalty**  
   Citing too many chunks reduces score (to discourage “cite everything” strategies)

**Final Score:** Weighted combination of the above (higher is better).  
This makes it possible to rank solutions that are not only accurate, but also **verifiably grounded**.

---

## Submission Format
Submit a CSV with these columns:

```csv
question_id,answer,cited_chunk_ids
q_3001,"Your answer here","[doc_0012_chunk_2, doc_0012_chunk_3]"
q_3002,"Another answer","[doc_0101_chunk_1]"
```