This competition evaluates Retrieval-Augmented Generation (RAG) systems on their ability to produce grounded answers supported by evidence citations.

Participants must retrieve relevant document chunks from a provided corpus and generate answers that are supported by those retrieved passages.

This dataset supports benchmark-style evaluation of Retrieval-Augmented Generation (RAG) systems on:
- **Grounded answering generation** (answer must match evidence)
- **Evidence Citation accuracy** (cite the correct chunks)
- **Hallucination control** (penalize unsupported content)

You will be given:
1) a **knowledge corpus** split into chunks, and  
2) questions that must be answered **with citations to chunk IDs**.

The dataset consists of a synthetic but structured knowledge corpus designed to simulate real-world operational documents such as incident reports, architecture documentation, and security investigations.

The goal is to benchmark how well RAG systems can answer questions while citing the correct supporting evidence.

---

## Files Included

### 1) `corpus.csv` - Knowledge Base Chunks
Each row is a chunk from a document.
```
**Columns**
- `chunk_id` (string): unique chunk identifier (example: `doc_0012_chunk_2`)
- `doc_id` (string): document identifier
- `title` (string): document title
- `doc_type` (string): document category (e.g., incident_report, policy, architecture, security_advisory)
- `chunk_text` (string): the evidence text participants can retrieve
```

---

### 2) `train.csv` - Training Questions (with gold labels)
Use this file to build and validate your approach.
```
**Columns**
- `question_id` (string): unique question identifier (example: `q_1021`)
- `question` (string): the question text
- `answer` (string): gold reference answer
- `gold_chunk_ids` (string/list-like): minimal set of evidence chunk IDs supporting the answer
```

---

### 3) `test.csv` - Test Questions (no labels)
Use this file to generate submissions.
```
**Columns**
- `question_id` (string)
- `question` (string)
```

---

### 4) `sample_submission.csv` - Submission Template
Your submission must contain:

```
- `question_id`
- `answer`
- `cited_chunk_ids`
```

`cited_chunk_ids` should be a list of chunk IDs from `corpus.csv`.

Example:
```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]"
```

### 5) metaData.csv
Dataset schema and column definitions.
```
**columns**
- field (column name)
- type (string/int/list)
- description
- example
```

Example rows:
```
chunk_id,string,Unique ID of corpus chunk,doc_0012_chunk_2
gold_chunk_ids,list[chunk_id],Gold evidence chunks (train only),[doc_0012_chunk_2]
```