from dataclasses import dataclass
from enum import Enum


@dataclass
class Task:
    benchmark: str
    metric: str
    col_name: str


# Select your tasks here
# ---------------------------------------------------
class Tasks(Enum):
    # task_key in the json file, metric_key in the json file, name to display in the leaderboard
    task0 = Task("overall", "anls", "ANLS (Overall)")
    task1 = Task("single_evidence", "anls", "ANLS (Single Evidence)")
    task2 = Task("multi_evidence_same_doc", "anls", "ANLS (Multi-Evidence, Same Doc)")
    task3 = Task("multi_evidence_multi_doc", "anls", "ANLS (Multi-Evidence, Multi Doc)")


# Your leaderboard name
# Static files are served relative to the static path set in app.py
TITLE = ""  #  """<h1 align="center" id="space-title">Agentic Document AI Benchmark</h1>"""

# What does your leaderboard evaluate?
INTRODUCTION_TEXT = """
Welcome to the **Agentic Document AI Leaderboard**! This benchmark evaluates the performance of AI agents on complex document understanding tasks that require multi-step reasoning and evidence gathering across documents.

The benchmark uses **ANLS (Average Normalized Levenshtein Similarity)** as the primary evaluation metric, measuring how well models can extract and synthesize information from documents. We evaluate performance on:
- **Overall accuracy** across the entire dataset
- **Single evidence** questions (information from one source)
- **Multi-evidence, same document** questions (combining information within a document)
- **Multi-evidence, multi-document** questions (synthesizing across multiple documents)

We also track **inference costs** in terms of agent steps and USD to help understand the efficiency of different approaches.
"""

# Which evaluations are you running? how can people reproduce what you have?
LLM_BENCHMARKS_TEXT = """
## How it works

The Agentic Document AI benchmark evaluates AI systems on their ability to:
1. Navigate and understand complex document structures
2. Extract relevant evidence to answer questions
3. Synthesize information from multiple sources
4. Perform multi-step reasoning with document context

## Metrics

### Performance Metrics (ANLS-based)
- **ANLS (Overall)**: Main score - Average Normalized Levenshtein Similarity across the entire dataset
- **ANLS (Single Evidence)**: Performance on questions requiring single evidence extraction
- **ANLS (Multi-Evidence, Same Doc)**: Performance when combining evidence within one document
- **ANLS (Multi-Evidence, Multi Doc)**: Performance when synthesizing across multiple documents

### Efficiency Metrics
- **Agent Steps**: Total number of reasoning/action steps taken by the agent
- **Cost (USD)**: Estimated inference cost in US dollars

## Reproducibility

To submit your results:

1. Run your model/agent on the benchmark dataset
2. Generate a JSONL file where each line contains one prediction:

```json
{"question": "What is Dr. McElhaney's position?", "answer": ["Senior Scientist"], "citations": [{"file": "1307326.pdf", "page": 1}], "iterations": 1, "id": "q_4"}
{"question": "Who is the CEO?", "answer": ["John Smith"], "citations": [{"file": "report.pdf", "page": 3}], "iterations": 2, "id": "q_5"}
```

**Required fields:**
- `question`: The question text (string)
- `answer`: List of answer strings
- `citations`: List of citation dicts with "file" and "page"
- `iterations`: Number of agent iterations/steps (integer)
- `id`: Unique question identifier (string)

3. Submit your JSONL file through the submission tab
4. The system will evaluate your predictions against the gold standard and compute ANLS scores

See `submission_template.jsonl` for a complete example.
"""