{
 "body": "\n## What it measures\n\nHistoires Morales evaluates moral alignment in French. Each story follows a seven-sentence structure: a norm, situation, and intention, followed by a normative action and consequence and a norm-divergent action and consequence.\n\nThe harness presents the dataset\u2019s query and choices to a multiple-choice model. The dataset is adapted from Moral Stories through translation and manual refinement.\n\n## How it is scored\n\nThe lm-evaluation-harness task reports mean `acc` and `acc_norm`, both higher-is-better. The task uses the dataset\u2019s train split as its test split and applies a custom `process_docs` function. No human baseline is stated.\n\n## Dataset and licence\n\nThe harness README describes 12,000 stories. It identifies the Hugging Face dataset `LabHC/histoires_morales`, with French narratives following the seven-sentence template. The Hugging Face dataset card lists an MIT licence. The dataset ships one `train` split, which the harness uses directly as its evaluation split with visible choice and label fields, so answers are public rather than held out.\n\n## Who publishes it\n\nThe dataset is maintained by LabHC and integrated into EleutherAI\u2019s lm-evaluation-harness. The README cites the paper \u201cHistoires Morales: A French Dataset for Assessing Moral Alignment,\u201d arXiv:2501.17117. It says the work was accepted to NAACL 2025; no current leaderboard was found.\n\n## Lineage\n\nHistoires Morales is adapted from the English Moral Stories dataset. The French collection is a language adaptation rather than an interchangeable alias. No successor was established.\n\n## Saturation and contamination\n\nSaturation is unknown. The dataset and task definition are public, and its source material derives from an established benchmark, creating medium exposure risk. Moral judgments are also culturally and linguistically situated.\n\n## How to run it\n\nUse lm-evaluation-harness task `histoires_morales`. It loads the train split, applies the repository\u2019s document processor, and reports accuracy and normalized accuracy. Record harness revision and preprocessing because the task uses a custom function.\n\n## Reading the numbers\n\nA high score indicates agreement with the dataset\u2019s normative distinctions in French. It does not define universal morality, cultural robustness, or safe policy behavior. Compare language versions only with care, and inspect disagreement cases rather than relying on one aggregate accuracy.\n\nThe seven-sentence structure separates the goal from the norm and consequences. That supports controlled comparison but may differ from spontaneous moral reasoning in conversation.\n",
 "build": {
  "built_at": "2026-09-09T16:56:50+00:00",
  "commit": "0a599558854c0e238c03a0f0d725239cb28f9d11",
  "eligibility_as_of": "2026-09-09"
 },
 "disposition": {
  "canonical_id": "histoires_morales",
  "reasons": [],
  "status": "unassessed",
  "verified_results": []
 },
 "models_covered": [],
 "page": {
  "aliases": [],
  "category": "safety",
  "contamination": {
   "note": "The dataset is publicly hosted and adapted from Moral Stories; exposure is not established.",
   "risk": "medium"
  },
  "dataset": {
   "languages": [
    "French"
   ],
   "license": "MIT",
   "modalities": [
    "text"
   ],
   "public_test_set": true,
   "size": 12000,
   "size_note": "The harness README describes 12,000 stories, each with seven sentences.",
   "splits": "train",
   "url": "https://huggingface.co/datasets/LabHC/histoires_morales"
  },
  "freshness": {
   "luna-batch-048": null,
   "luna-batch-048 (Codex coordinated)": null,
   "researched": "2026-09-08",
   "researched_by": "GPT-5.6 Luna",
   "reviewed": "2026-09-08",
   "reviewed_by": "Claude Sonnet 5 independent review"
  },
  "harness": {
   "bigbench": "",
   "helm": "",
   "inspect_evals": "",
   "lm_eval": "histoires_morales",
   "opencompass": "",
   "other": ""
  },
  "id": "histoires_morales",
  "last_updated": "2025-07",
  "leaderboard_url": "",
  "lineage": {
   "family": "",
   "predecessor": "",
   "successors": [],
   "variants": []
  },
  "measures": "The dataset presents French narratives describing an intention, a norm-observing action and consequence, and a norm-divergent action and consequence. It evaluates moral alignment judgments in French.",
  "metric": {
   "baseline_note": "The harness defines accuracy and normalized accuracy but no baseline.",
   "direction": "higher_is_better",
   "human_baseline": null,
   "max_score": 100,
   "name": "accuracy",
   "random_baseline": null,
   "unit": "percent"
  },
  "name": "Histoires Morales",
  "page_kind": "benchmark",
  "paper": {
   "arxiv": "2501.17117",
   "title": "Histoires Morales: A French Dataset for Assessing Moral Alignment",
   "url": "https://arxiv.org/abs/2501.17117",
   "year": 2025
  },
  "publisher": {
   "authors": [],
   "org": "LabHC",
   "url": "https://huggingface.co/datasets/LabHC/histoires_morales"
  },
  "released": "2025",
  "repo_url": "https://github.com/EleutherAI/lm-evaluation-harness",
  "saturation": {
   "as_of": "",
   "note": "No current authoritative leaderboard was established.",
   "status": "unknown",
   "top_score": null
  },
  "sources": [
   {
    "accessed": "2026-09-08",
    "title": "lm-evaluation-harness Histoires Morales README",
    "url": "https://raw.githubusercontent.com/EleutherAI/lm-evaluation-harness/main/lm_eval/tasks/histoires_morales/README.md"
   },
   {
    "accessed": "2026-09-08",
    "title": "Histoires Morales harness task",
    "url": "https://raw.githubusercontent.com/EleutherAI/lm-evaluation-harness/main/lm_eval/tasks/histoires_morales/histoires_morales.yaml"
   },
   {
    "accessed": "2026-09-08",
    "title": "Histoires Morales paper",
    "url": "https://arxiv.org/abs/2501.17117"
   },
   {
    "accessed": "2026-09-08",
    "title": "Hugging Face dataset API record (license mit, single train split parquet file)",
    "url": "https://huggingface.co/api/datasets/LabHC/histoires_morales"
   }
  ],
  "status": "active",
  "subcategory": "moral alignment",
  "summary": "Histoires Morales is a French moral-alignment dataset of normative and norm-divergent stories.",
  "tags": [
   "french",
   "moral-alignment",
   "multiple-choice"
  ],
  "task_format": "Multiple-choice classification over moral-story narratives."
 }
}