{
 "model": "chaoliangUNSW/Jev-Style-2B-Decision-v4",
 "note": "Every source in the training rows of this model (88 sources). licence_recorded = the licence field stored with the rows in the training pool (one entry per distinct value; 'generated...' = generated for this project). components = the mixture-table rows of the model card. link = the dataset repository the rows were built from (the mirror when one was used), where one exists. origin = rows reused from the v3 training pool or built for v4. Per-source row counts are not listed. Check each source's own terms before use; see 'Training data and licences' on the model card.",
 "totals": {
  "train_rows": 916923,
  "pool_tokens": 589478732,
  "trained_tokens_per_run": [
   294100000,
   294000000
  ],
  "pool_manifest_sha256": "abf858e8e355e877fced2111135996ae0f9eff39b94102e3d3c114fd65522fcf"
 },
 "sources": [
  {
   "id": "alisawuffles/WANLI",
   "components": [
    "Public reading tasks"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/alisawuffles/WANLI",
   "note": "written by GPT-3 and revised by crowdworkers",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "allenai/ai2_arc",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "cc-by-sa-4.0"
   ],
   "link": "https://huggingface.co/datasets/allenai/ai2_arc",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "share_alike": true,
   "origin": "v4"
  },
  {
   "id": "allenai/aokvqa (train; mirror HuggingFaceM4/A-OKVQA)",
   "components": [
    "Images"
   ],
   "licence_recorded": [
    "apache-2.0"
   ],
   "link": "https://huggingface.co/datasets/HuggingFaceM4/A-OKVQA",
   "note": "images: COCO 2017 (Flickr photos under their own licences, COCO terms of use)",
   "origin": "v4"
  },
  {
   "id": "allenai/qasc",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/allenai/qasc",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "allenai/qasper",
   "components": [
    "Public reading tasks"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/allenai/qasper",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "allenai/ropes",
   "components": [
    "Public reading tasks"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/allenai/ropes",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "allenai/winogrande",
   "components": [
    "Language understanding"
   ],
   "licence_recorded": [
    "cc-by-2.0"
   ],
   "link": "https://huggingface.co/datasets/allenai/winogrande",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "clinc/clinc_oos",
   "components": [
    "Option-format views"
   ],
   "licence_recorded": [
    "cc-by-3.0"
   ],
   "link": "https://huggingface.co/datasets/clinc/clinc_oos",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "clinc150",
   "components": [
    "Intents and yes/no QA"
   ],
   "licence_recorded": [
    "cc-by-3.0"
   ],
   "link": "https://huggingface.co/datasets/clinc/clinc_oos",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "coastalcph/lex_glue",
   "components": [
    "Public reading tasks"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/coastalcph/lex_glue",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "czyssrs/FinQA",
   "components": [
    "Judging and scoring",
    "Public reading tasks"
   ],
   "licence_recorded": [
    "mit"
   ],
   "link": "https://github.com/czyssrs/FinQA",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "deepmind/aqua_rat",
   "components": [
    "Judging and scoring",
    "Public reading tasks"
   ],
   "licence_recorded": [
    "apache-2.0"
   ],
   "link": "https://huggingface.co/datasets/deepmind/aqua_rat",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "derek-thomas/ScienceQA (train)",
   "components": [
    "Images",
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "cc-by-sa-4.0"
   ],
   "link": "https://huggingface.co/datasets/derek-thomas/ScienceQA",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "share_alike": true,
   "origin": "v4"
  },
  {
   "id": "facebook/textvqa (train sample; mirror lmms-lab/textvqa)",
   "components": [
    "Images"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/lmms-lab/textvqa",
   "note": "images: Open Images (CC BY 2.0)",
   "origin": "v4"
  },
  {
   "id": "generated:crux_type_synth",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated for this project; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "generated:v4fam/multi_entity_sentiment",
   "components": [
    "Language understanding"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "texts written by a Claude / GPT model from code assignments; kept where two blind labeler models reproduce them; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "generated:v4fam/narrative_deduction",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "narratives written by a Claude / GPT model from code-built worlds; gold from code, checked by a blind solver model; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "generated:v4fam/stance_sarcasm",
   "components": [
    "Language understanding"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "texts written by Claude / GPT models from code specs; kept where two blind labeler models agree; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "generated:v4fam/synthetic_intents",
   "components": [
    "Retrieval and classification"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "messages written by a Claude / GPT model; kept where a blind verifier model picks the same intent; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "google-deepmind/habermas_machine (train; mirror mjang01011/cs329x-habermas-dataset)",
   "components": [
    "Judging and scoring",
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "apache-2.0"
   ],
   "link": "https://huggingface.co/datasets/mjang01011/cs329x-habermas-dataset",
   "note": "upstream github.com/google-deepmind/habermas_machine (Apache-2.0); the mirror card has no licence field; part of the rows use the dataset's own human rating distributions as soft labels",
   "origin": "v4"
  },
  {
   "id": "google-research-datasets/mbpp",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/google-research-datasets/mbpp",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "google-research-datasets/schema_guided_dstc8",
   "components": [
    "Retrieval and classification"
   ],
   "licence_recorded": [
    "cc-by-sa-4.0"
   ],
   "link": "https://huggingface.co/datasets/google-research-datasets/schema_guided_dstc8",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "share_alike": true,
   "origin": "v4"
  },
  {
   "id": "google/boolq",
   "components": [
    "Intents and yes/no QA"
   ],
   "licence_recorded": [
    "cc-by-sa-3.0"
   ],
   "link": "https://huggingface.co/datasets/google/boolq",
   "share_alike": true,
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "google/civil_comments",
   "components": [
    "Safety themes"
   ],
   "licence_recorded": [
    "cc0-1.0"
   ],
   "link": "https://huggingface.co/datasets/google/civil_comments",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "hardfam:F1",
   "components": [
    "Hard cases"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "hard-case family written by an OpenAI GPT model; labels computed by code",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "hardfam:F2",
   "components": [
    "Hard cases"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "hard-case family generated by code",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "hardfam:F3",
   "components": [
    "Hard cases"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "hard-case family generated by code",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "hardfam:F6",
   "components": [
    "Hard cases"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "hard-case family written by an OpenAI GPT model; labels computed by code",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "hardfam:F9",
   "components": [
    "Hard cases"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "hard-case family generated by code",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "hover-nlp/hover",
   "components": [
    "Retrieval and classification"
   ],
   "licence_recorded": [
    "cc-by-sa-4.0"
   ],
   "link": "https://huggingface.co/datasets/hover-nlp/hover",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "share_alike": true,
   "origin": "v4"
  },
  {
   "id": "hugosousa/TimeQA",
   "components": [
    "Public reading tasks"
   ],
   "licence_recorded": [
    "bsd-3-clause-clear"
   ],
   "link": "https://huggingface.co/datasets/hugosousa/TimeQA",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "iabufarha/iSarcasmEval task A en (train; mirror tlam25/iSarcasm_taskA)",
   "components": [
    "Language understanding"
   ],
   "licence_recorded": [
    "mit"
   ],
   "link": "https://huggingface.co/datasets/tlam25/iSarcasm_taskA",
   "note": "upstream github.com/iabufarha/iSarcasmEval (MIT); the mirror card says Apache-2.0; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "jackhhao/jailbreak-classification",
   "components": [
    "Safety themes"
   ],
   "licence_recorded": [
    "apache-2.0"
   ],
   "link": "https://huggingface.co/datasets/jackhhao/jailbreak-classification",
   "note": "its jailbreak prompts come from the jailbreak_llms collection (research purposes only); part of the benign prompts come from GPTeacher (generated by GPT-4)",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "jmhessel/newyorker_caption_contest",
   "components": [
    "Arts and taste"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/jmhessel/newyorker_caption_contest",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "Lakera/gandalf_ignore_instructions",
   "components": [
    "Safety themes"
   ],
   "licence_recorded": [
    "mit"
   ],
   "link": "https://huggingface.co/datasets/Lakera/gandalf_ignore_instructions",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "lasha-nlp/CONDAQA",
   "components": [
    "Public reading tasks"
   ],
   "licence_recorded": [
    "apache-2.0"
   ],
   "link": "https://huggingface.co/datasets/lasha-nlp/CONDAQA",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "legacy:generated_rules:approved_earliest",
   "components": [
    "Mac agent checks"
   ],
   "licence_recorded": [
    "generated:round-1 project collectors and rule generator (mac_laya.tasks / trajectory_v2); no third-party text"
   ],
   "link": null,
   "note": "generated for this project",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "legacy:generated_rules:cost_under_limit",
   "components": [
    "Mac agent checks"
   ],
   "licence_recorded": [
    "generated:round-1 project collectors and rule generator (mac_laya.tasks / trajectory_v2); no third-party text"
   ],
   "link": null,
   "note": "generated for this project",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "legacy:generated_rules:priority_max",
   "components": [
    "Mac agent checks"
   ],
   "licence_recorded": [
    "generated:round-1 project collectors and rule generator (mac_laya.tasks / trajectory_v2); no third-party text"
   ],
   "link": null,
   "note": "generated for this project",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "legacy:generated_rules:team_min_amount",
   "components": [
    "Mac agent checks"
   ],
   "licence_recorded": [
    "generated:round-1 project collectors and rule generator (mac_laya.tasks / trajectory_v2); no third-party text"
   ],
   "link": null,
   "note": "generated for this project",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "legacy:mac_v2:filesystem_policy",
   "components": [
    "Mac agent checks"
   ],
   "licence_recorded": [
    "generated:round-1 project collectors and rule generator (mac_laya.tasks / trajectory_v2); no third-party text"
   ],
   "link": null,
   "note": "generated for this project",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "Lichess/chess-puzzles",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "cc0-1.0"
   ],
   "link": "https://huggingface.co/datasets/Lichess/chess-puzzles",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:api_catalog",
   "components": [
    "Tools and automation"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated by this project's code (no third-party text); part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:aspect_review",
   "components": [
    "Language understanding"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "reviews written by Claude / GPT models from code blueprints; labels from code, confirmed by a blind second model; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:bpomp",
   "components": [
    "Arts and taste"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "limericks written by Claude / GPT models; the defects and labels are made by code; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:device_simulator",
   "components": [
    "Tools and automation"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated by this project's code (no third-party text); part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:gsm_templates",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated for this project; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:link_safety",
   "components": [
    "Retrieval and classification"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated for this project; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:long_needle",
   "components": [
    "Long documents"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated by this project's code (no third-party text); part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:model_selection",
   "components": [
    "Tools and automation"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated by this project's code (no third-party text); part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:new_domain",
   "components": [
    "New-domain decisions"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "documents and questions written by Claude and GPT models; answers kept where two teachers agreed; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:palette_pref",
   "components": [
    "Arts and taste"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated by this project's code (no third-party text); part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:sata_type",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated for this project; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:tool_catalog",
   "components": [
    "Tools and automation"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated by this project's code (no third-party text); part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev-synth:tool_retrieval_type",
   "components": [
    "Tools and automation"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated by this project's code (no third-party text); part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "macjev/long_table",
   "components": [
    "Long documents"
   ],
   "licence_recorded": [
    "generated:cc0 (programmatic synthetic tables)"
   ],
   "link": null,
   "note": "generated for this project",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "macjev_sim:goal_done",
   "components": [
    "Mac agent checks"
   ],
   "licence_recorded": [
    "generated:macjev-sim (synthetic states written by project code; no third-party text)"
   ],
   "link": null,
   "note": "the project's own simulator; part of the goal wording was paraphrased by an OpenAI GPT model",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "massive-1.1:intent",
   "components": [
    "Intents and yes/no QA"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/AmazonScience/massive",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "mteb/banking77",
   "components": [
    "Retrieval and classification"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/mteb/banking77",
   "note": "BANKING77 by PolyAI: CC BY 4.0 upstream (https://huggingface.co/datasets/PolyAI/banking77); rows taken from the MTEB mirror, whose card says MIT; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "music-x-lab/POP909-Dataset (chord_midi labels)",
   "components": [
    "Arts and taste"
   ],
   "licence_recorded": [
    "mit"
   ],
   "link": "https://github.com/music-x-lab/POP909-Dataset",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "naver-clova-ix/cord-v2 (train)",
   "components": [
    "Images"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/naver-clova-ix/cord-v2",
   "origin": "v4"
  },
  {
   "id": "neuralchemy/Prompt-injection-dataset",
   "components": [
    "Safety themes"
   ],
   "licence_recorded": [
    "apache-2.0",
    "cc-by-4.0",
    "mit"
   ],
   "link": "https://huggingface.co/datasets/neuralchemy/Prompt-injection-dataset",
   "note": "the upstream rows its card marks research-only were removed",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "nvidia/HelpSteer2",
   "components": [
    "Judging and scoring"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/nvidia/HelpSteer2",
   "note": "responses mostly written by NVIDIA Nemotron models and Mixtral-8x7B-Instruct; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API); part of the rows use the dataset's own human rating distributions as soft labels",
   "origin": "v4"
  },
  {
   "id": "nvidia/HelpSteer3",
   "components": [
    "Judging and scoring"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/nvidia/HelpSteer3",
   "note": "responses written by language models (rated by humans); part of the rows use the dataset's own human rating distributions as soft labels",
   "origin": "v4"
  },
  {
   "id": "nvidia/When2Call",
   "components": [
    "Tools and automation"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/nvidia/When2Call",
   "note": "synthetic tool-use conversations generated with language models by NVIDIA; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "openai/gsm8k",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "mit"
   ],
   "link": "https://huggingface.co/datasets/openai/gsm8k",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "openlifescienceai/medmcqa",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "apache-2.0"
   ],
   "link": "https://huggingface.co/datasets/openlifescienceai/medmcqa",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "ParticleMedia/RAGTruth (train; mirror wandb/RAGTruth-processed)",
   "components": [
    "Retrieval and classification"
   ],
   "licence_recorded": [
    "mit"
   ],
   "link": "https://huggingface.co/datasets/wandb/RAGTruth-processed",
   "note": "upstream github.com/ParticleMedia/RAGTruth (MIT); responses written by language models (hallucination spans annotated by humans); part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "rajpurkar/squad_v2",
   "components": [
    "Safety themes"
   ],
   "licence_recorded": [
    "cc-by-sa-4.0"
   ],
   "link": "https://huggingface.co/datasets/rajpurkar/squad_v2",
   "share_alike": true,
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "Rowan/hellaswag",
   "components": [
    "Language understanding"
   ],
   "licence_recorded": [
    "mit"
   ],
   "link": "https://huggingface.co/datasets/Rowan/hellaswag",
   "note": "MIT only in the card text (no licence field); the original GitHub repository is blocked after a DMCA notice from wikiHow; only the ActivityNet-caption items were used; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "stanfordnlp/contract-nli",
   "components": [
    "Language understanding"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://github.com/stanfordnlp/contract-nli",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "stanfordnlp/snli",
   "components": [
    "Language understanding"
   ],
   "licence_recorded": [
    "cc-by-sa-4.0"
   ],
   "link": "https://huggingface.co/datasets/stanfordnlp/snli",
   "note": "SNLI train pairs plus adversarial hypotheses written by a Claude / GPT model, kept where two blind labeler models agree; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "share_alike": true,
   "origin": "v4"
  },
  {
   "id": "synth-v4:algo_reasoning",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated for this project; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "synth:causal_graph_scm",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated for this project; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "synth:clinical_trial_reports",
   "components": [
    "Language understanding"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated for this project; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "synth:reasoning_relevance",
   "components": [
    "Retrieval and classification"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "generated for this project; part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "tasksource/esci",
   "components": [
    "Retrieval and classification"
   ],
   "licence_recorded": [
    "apache-2.0"
   ],
   "link": "https://huggingface.co/datasets/tasksource/esci",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "tau/commonsense_qa",
   "components": [
    "Knowledge and reasoning"
   ],
   "licence_recorded": [
    "mit"
   ],
   "link": "https://huggingface.co/datasets/tau/commonsense_qa",
   "note": "part of the rows use soft labels distilled from TypeSafe Jev 1.14 (via its public API)",
   "origin": "v4"
  },
  {
   "id": "teacher:themes_jailbreak",
   "components": [
    "Safety themes"
   ],
   "licence_recorded": [
    "generated:claude-opus-5.5"
   ],
   "link": null,
   "note": "jailbreak and toxicity prompts written and labelled by an Anthropic Claude model",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "theatticusproject/maud",
   "components": [
    "Public reading tasks"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/theatticusproject/maud",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "tonytan48/TempReason",
   "components": [
    "Public reading tasks"
   ],
   "licence_recorded": [
    "cc-by-sa-3.0"
   ],
   "link": "https://huggingface.co/datasets/tonytan48/TempReason",
   "share_alike": true,
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "TrustAIRLab/in-the-wild-jailbreak-prompts",
   "components": [
    "Safety themes"
   ],
   "licence_recorded": [
    "mit"
   ],
   "link": "https://huggingface.co/datasets/TrustAIRLab/in-the-wild-jailbreak-prompts",
   "note": "prompts from the jailbreak_llms collection, which states it is for research purposes only",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "typed_synthetic",
   "components": [
    "Typed decisions (v3)"
   ],
   "licence_recorded": [
    "generated"
   ],
   "link": null,
   "note": "model-written business workflows: designed by OpenAI GPT and Anthropic Claude models, labelled by Claude models",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "ucinlp/drop",
   "components": [
    "Public reading tasks"
   ],
   "licence_recorded": [
    "cc-by-sa-4.0"
   ],
   "link": "https://huggingface.co/datasets/ucinlp/drop",
   "share_alike": true,
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "UCLNLP/sharc",
   "components": [
    "Public reading tasks"
   ],
   "licence_recorded": [
    "cc-by-sa-3.0"
   ],
   "link": "https://huggingface.co/datasets/UCLNLP/sharc",
   "share_alike": true,
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "VQAv2 (train sample; mirror lmms-lab/VQAv2)",
   "components": [
    "Images"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/lmms-lab/VQAv2",
   "note": "images: COCO train2014 (Flickr photos under their own licences, COCO terms of use)",
   "origin": "v4"
  },
  {
   "id": "wenhu/tab_fact",
   "components": [
    "Public reading tasks"
   ],
   "licence_recorded": [
    "cc-by-4.0"
   ],
   "link": "https://huggingface.co/datasets/wenhu/tab_fact",
   "origin": "v3 training pool (reused)"
  },
  {
   "id": "yanismiraoui/prompt_injections",
   "components": [
    "Safety themes"
   ],
   "licence_recorded": [
    "apache-2.0"
   ],
   "link": "https://huggingface.co/datasets/yanismiraoui/prompt_injections",
   "origin": "v3 training pool (reused)"
  }
 ]
}