benchmark_manifest.json

{
  "benchmark_name": "IndicConBench",
  "version": "1.0.0",
  "author": "Vikhram S",
  "organization": "Vikhram Labs",
  "huggingface_url": "https://huggingface.co/datasets/vikhram-labs/IndicConBench",
  "license": "apache-2.0",
  "total_examples": 8408,
  "subsets": {
    "IndicConBench-Lite": {
      "count": 3720,
      "score_range": [
        0,
        35
      ],
      "file": "benchmark_lite.jsonl"
    },
    "IndicConBench-Core": {
      "count": 4624,
      "score_range": [
        36,
        60
      ],
      "file": "benchmark_core.jsonl"
    },
    "IndicConBench-Hard": {
      "count": 36,
      "score_range": [
        61,
        80
      ],
      "file": "benchmark_hard.jsonl"
    },
    "IndicConBench-Expert": {
      "count": 28,
      "score_range": [
        81,
        100
      ],
      "file": "benchmark_expert.jsonl"
    }
  },
  "task_distribution": {
    "qa": 930,
    "retrieval": 930,
    "reasoning": 930,
    "summarization": 1860,
    "entailment": 2790,
    "classification": 930,
    "multihop": 8,
    "linking": 10,
    "gpqa": 20
  },
  "language_distribution": {
    "en": 4204,
    "hi": 4204
  },
  "knowledge_type_distribution": {
    "factual": 3720,
    "interpretive": 3694,
    "procedural": 930,
    "jurisprudential": 54,
    "historical": 10
  },
  "reasoning_depth_distribution": {
    "1": 3720,
    "3": 3694,
    "2": 930,
    "4": 36,
    "5": 28
  },
  "difficulty_score_distribution": {
    "10-19": 2325,
    "100-109": 10,
    "20-29": 1395,
    "30-39": 465,
    "40-49": 1860,
    "50-59": 1847,
    "60-69": 452,
    "70-79": 31,
    "80-89": 5,
    "90-99": 18
  },
  "capability_coverage": {
    "9. Hallucination Resistance": 4670,
    "1. Constitutional Knowledge": 4650,
    "2. Constitutional Reasoning": 3758,
    "7. Adversarial Robustness": 3740,
    "3. Long-Context Reasoning": 1868,
    "10. Legal Argument Quality": 950,
    "5. Retrieval": 930,
    "4. Multi-Hop Legal Reasoning": 38,
    "8. Citation Faithfulness": 28,
    "citation_accuracy": 20,
    "6. Cross-Lingual Transfer": 10
  },
  "benchmark_dimensions": [
    "1. Constitutional Knowledge",
    "2. Constitutional Reasoning",
    "3. Long-Context Reasoning",
    "4. Multi-Hop Legal Reasoning",
    "5. Retrieval",
    "6. Cross-Lingual Transfer",
    "7. Adversarial Robustness",
    "8. Citation Faithfulness",
    "9. Hallucination Resistance",
    "10. Legal Argument Quality"
  ],
  "metrics": [
    "EM",
    "F1",
    "ROUGE-L",
    "BERTScore",
    "Retrieval_Recall@1",
    "Retrieval_Recall@5",
    "MRR",
    "nDCG",
    "Calibration_Error",
    "Hallucination_Rate",
    "Citation_Accuracy"
  ],
  "evaluation_script": "benchmark/evaluate_benchmark.py",
  "leaderboard_script": "benchmark/leaderboard.py"
}