{
  "schema_version": 1,
  "revision": "2026-10-04",
  "label_encoding": {
    "1": "harmful/toxic",
    "0": "safe/non-toxic",
    "-1": "unparsed"
  },
  "note": "Contains aggregate counts only; no item text, item identifiers, generations or training data. The named paper reports model names in a separate panel.",
  "runs": [
    {
      "id": "Run 1",
      "source_items_sha256": "46e9be7456c8c4a13168428f7fc7679e33b34660b564d379fecd90f9eca25cb2",
      "recorded_run_utc": "2026-10-02T20:41:25.493821+00:00",
      "tasks": {
        "cultural_prompt": [
          {
            "gold": 0,
            "pred": 0,
            "n": 32
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 7
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 2
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 30
          }
        ],
        "cultural_response": [
          {
            "gold": 0,
            "pred": 0,
            "n": 58
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 2
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 5
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 6
          }
        ],
        "cultural_wild": [
          {
            "gold": 0,
            "pred": 0,
            "n": 211
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 4
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 31
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 184
          }
        ],
        "toxicity": [
          {
            "gold": 0,
            "pred": 0,
            "n": 282
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 218
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 184
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 316
          }
        ]
      },
      "prompt_response_joint": [
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 0,
          "n": 2
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 5
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 32
        },
        {
          "prompt_gold": 1,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 20
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 0,
          "response_correct": 0,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 1,
          "response_correct": 0,
          "n": 4
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 6
        }
      ],
      "settings_recorded": {
        "seed": 0,
        "bs": 8,
        "thinking": false,
        "think_budget": 0
      },
      "task_config_digests_recorded": {
        "cultural_prompt": "83e8b807e4ea",
        "cultural_response": "83e8b807e4ea",
        "cultural_wild": "0fc7d0ba1688",
        "toxicity": "ae35ef84f5df"
      },
      "checkpoint_sha256": null,
      "dataset_revision": null,
      "generation_reproducibility": "incomplete: immutable checkpoint, tokenizer/template, dataset revision and original runtime version not recorded in these files",
      "result": {
        "score": 44.92238891599356,
        "ci95": [
          37.87936950565585,
          51.583686038606096
        ],
        "normalized_legs": {
          "cultural_prompt": 75.80128205128204,
          "cultural_response": 51.21212121212122,
          "cultural_wild": 83.72093023255813,
          "toxicity": 19.599999999999994
        },
        "draws": 20000,
        "seed": 20261004,
        "method": "percentile bootstrap; joint-gold-stratified prompt clusters; class-stratified independent toxicity/wild",
        "uncertainty": "conditional item sampling only; excludes training, generation reruns, model selection and dataset shift"
      }
    },
    {
      "id": "Run 2",
      "source_items_sha256": "c2f792020328ddc728a340d9bc8cd82eacfa9306b7034d570ca26a5654387d07",
      "recorded_run_utc": "2026-10-02T19:39:10.758345+00:00",
      "tasks": {
        "cultural_prompt": [
          {
            "gold": 0,
            "pred": 0,
            "n": 35
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 4
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 3
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 29
          }
        ],
        "cultural_response": [
          {
            "gold": 0,
            "pred": 0,
            "n": 55
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 5
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 2
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 9
          }
        ],
        "cultural_wild": [
          {
            "gold": 0,
            "pred": 0,
            "n": 212
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 3
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 46
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 169
          }
        ],
        "toxicity": [
          {
            "gold": 0,
            "pred": 0,
            "n": 357
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 143
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 283
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 217
          }
        ]
      },
      "prompt_response_joint": [
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 0,
          "n": 3
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 1
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 0,
          "n": 2
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 33
        },
        {
          "prompt_gold": 1,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 20
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 0,
          "response_correct": 0,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 1,
          "response_correct": 0,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 8
        }
      ],
      "settings_recorded": {
        "seed": 0,
        "bs": 8,
        "thinking": false,
        "think_budget": 0
      },
      "task_config_digests_recorded": {
        "cultural_prompt": "83e8b807e4ea",
        "cultural_response": "83e8b807e4ea",
        "cultural_wild": "0fc7d0ba1688",
        "toxicity": "ae35ef84f5df"
      },
      "checkpoint_sha256": null,
      "dataset_revision": null,
      "generation_reproducibility": "incomplete: immutable checkpoint, tokenizer/template, dataset revision and original runtime version not recorded in these files",
      "result": {
        "score": 45.91045675900327,
        "ci95": [
          39.404488546014704,
          51.53846947787355
        ],
        "normalized_legs": {
          "cultural_prompt": 80.36858974358974,
          "cultural_response": 73.48484848484848,
          "cultural_wild": 77.2093023255814,
          "toxicity": 14.79999999999999
        },
        "draws": 20000,
        "seed": 20261004,
        "method": "percentile bootstrap; joint-gold-stratified prompt clusters; class-stratified independent toxicity/wild",
        "uncertainty": "conditional item sampling only; excludes training, generation reruns, model selection and dataset shift"
      }
    },
    {
      "id": "Run 3",
      "source_items_sha256": "d0c322d986d2c0bc802deec28747325de3127a3ef3a3dab93d8612d2fb06ec18",
      "recorded_run_utc": "2026-10-02T21:13:39.631452+00:00",
      "tasks": {
        "cultural_prompt": [
          {
            "gold": 0,
            "pred": 0,
            "n": 35
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 4
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 2
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 30
          }
        ],
        "cultural_response": [
          {
            "gold": 0,
            "pred": 0,
            "n": 57
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 3
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 3
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 8
          }
        ],
        "cultural_wild": [
          {
            "gold": 0,
            "pred": 0,
            "n": 212
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 3
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 49
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 166
          }
        ],
        "toxicity": [
          {
            "gold": 0,
            "pred": 0,
            "n": 352
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 148
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 278
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 222
          }
        ]
      },
      "prompt_response_joint": [
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 0,
          "n": 2
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 2
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 0,
          "n": 1
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 34
        },
        {
          "prompt_gold": 1,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 20
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 0,
          "response_correct": 0,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 1,
          "response_correct": 0,
          "n": 2
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 8
        }
      ],
      "settings_recorded": {
        "seed": 0,
        "bs": 8,
        "thinking": false,
        "think_budget": 0
      },
      "task_config_digests_recorded": {
        "cultural_prompt": "83e8b807e4ea",
        "cultural_response": "83e8b807e4ea",
        "cultural_wild": "0fc7d0ba1688",
        "toxicity": "ae35ef84f5df"
      },
      "checkpoint_sha256": null,
      "dataset_revision": null,
      "generation_reproducibility": "incomplete: immutable checkpoint, tokenizer/template, dataset revision and original runtime version not recorded in these files",
      "result": {
        "score": 45.23913599320575,
        "ci95": [
          38.60081180205454,
          51.24152217049746
        ],
        "normalized_legs": {
          "cultural_prompt": 83.49358974358974,
          "cultural_response": 67.72727272727272,
          "cultural_wild": 75.81395348837209,
          "toxicity": 14.79999999999999
        },
        "draws": 20000,
        "seed": 20261004,
        "method": "percentile bootstrap; joint-gold-stratified prompt clusters; class-stratified independent toxicity/wild",
        "uncertainty": "conditional item sampling only; excludes training, generation reruns, model selection and dataset shift"
      }
    },
    {
      "id": "Run 4",
      "source_items_sha256": "c89732081de164b599a1ed05f4ba1e3fe3fbefc3c96d434358a179b684e4aa8e",
      "recorded_run_utc": "2026-10-03T00:15:24.383129+00:00",
      "tasks": {
        "cultural_prompt": [
          {
            "gold": 0,
            "pred": 0,
            "n": 35
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 4
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 4
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 28
          }
        ],
        "cultural_response": [
          {
            "gold": 0,
            "pred": 0,
            "n": 54
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 6
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 2
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 9
          }
        ],
        "cultural_wild": [
          {
            "gold": 0,
            "pred": 0,
            "n": 211
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 4
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 43
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 172
          }
        ],
        "toxicity": [
          {
            "gold": 0,
            "pred": 0,
            "n": 397
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 103
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 342
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 158
          }
        ]
      },
      "prompt_response_joint": [
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 0,
          "n": 3
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 1
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 0,
          "n": 2
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 33
        },
        {
          "prompt_gold": 1,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 2
        },
        {
          "prompt_gold": 1,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 0,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 18
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 0,
          "response_correct": 0,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 1,
          "response_correct": 0,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 8
        }
      ],
      "settings_recorded": {
        "seed": 0,
        "bs": 8,
        "thinking": false,
        "think_budget": 0
      },
      "task_config_digests_recorded": {
        "cultural_prompt": "83e8b807e4ea",
        "cultural_response": "83e8b807e4ea",
        "cultural_wild": "0fc7d0ba1688",
        "toxicity": "ae35ef84f5df"
      },
      "checkpoint_sha256": null,
      "dataset_revision": null,
      "generation_reproducibility": "incomplete: immutable checkpoint, tokenizer/template, dataset revision and original runtime version not recorded in these files",
      "result": {
        "score": 43.36688440758209,
        "ci95": [
          36.922425682360284,
          48.95255751838601
        ],
        "normalized_legs": {
          "cultural_prompt": 77.24358974358974,
          "cultural_response": 71.81818181818183,
          "cultural_wild": 78.13953488372091,
          "toxicity": 11.00000000000001
        },
        "draws": 20000,
        "seed": 20261004,
        "method": "percentile bootstrap; joint-gold-stratified prompt clusters; class-stratified independent toxicity/wild",
        "uncertainty": "conditional item sampling only; excludes training, generation reruns, model selection and dataset shift"
      }
    },
    {
      "id": "Run 5",
      "source_items_sha256": "e19f51bf7cccf7168818922ae8891bb238e9b9bb478337b592496cbb677ce97f",
      "recorded_run_utc": "2026-10-03T11:01:59.956666+00:00",
      "tasks": {
        "cultural_prompt": [
          {
            "gold": 0,
            "pred": 0,
            "n": 31
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 8
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 2
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 30
          }
        ],
        "cultural_response": [
          {
            "gold": 0,
            "pred": 0,
            "n": 54
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 6
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 3
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 8
          }
        ],
        "cultural_wild": [
          {
            "gold": 0,
            "pred": 0,
            "n": 211
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 4
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 26
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 189
          }
        ],
        "toxicity": [
          {
            "gold": 0,
            "pred": 0,
            "n": 397
          },
          {
            "gold": 0,
            "pred": 1,
            "n": 103
          },
          {
            "gold": 1,
            "pred": 0,
            "n": 337
          },
          {
            "gold": 1,
            "pred": 1,
            "n": 163
          }
        ]
      },
      "prompt_response_joint": [
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 0,
          "n": 5
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 3
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 0,
          "n": 1
        },
        {
          "prompt_gold": 0,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 30
        },
        {
          "prompt_gold": 1,
          "response_gold": 0,
          "prompt_correct": 0,
          "response_correct": 1,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 0,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 20
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 0,
          "response_correct": 0,
          "n": 1
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 1,
          "response_correct": 0,
          "n": 2
        },
        {
          "prompt_gold": 1,
          "response_gold": 1,
          "prompt_correct": 1,
          "response_correct": 1,
          "n": 8
        }
      ],
      "settings_recorded": {
        "seed": 0,
        "bs": 8,
        "thinking": false,
        "think_budget": 0
      },
      "task_config_digests_recorded": {
        "cultural_prompt": "83e8b807e4ea",
        "cultural_response": "83e8b807e4ea",
        "cultural_wild": "0fc7d0ba1688",
        "toxicity": "ae35ef84f5df"
      },
      "checkpoint_sha256": null,
      "dataset_revision": null,
      "generation_reproducibility": "incomplete: immutable checkpoint, tokenizer/template, dataset revision and original runtime version not recorded in these files",
      "result": {
        "score": 43.00182730705987,
        "ci95": [
          36.06705870195696,
          49.22269135902856
        ],
        "normalized_legs": {
          "cultural_prompt": 73.23717948717947,
          "cultural_response": 62.72727272727274,
          "cultural_wild": 86.04651162790698,
          "toxicity": 12.00000000000001
        },
        "draws": 20000,
        "seed": 20261004,
        "method": "percentile bootstrap; joint-gold-stratified prompt clusters; class-stratified independent toxicity/wild",
        "uncertainty": "conditional item sampling only; excludes training, generation reruns, model selection and dataset shift"
      }
    }
  ],
  "aggregation_reference": {
    "id": "aggregation_reference",
    "source_items_sha256": "f6a04b6c77e9c28aa95d69d99f56caba722bab740dedb40f8fc1f901f01ab6ef",
    "tasks": {
      "cultural_prompt": [
        {
          "gold": 0,
          "pred": 0,
          "n": 30
        },
        {
          "gold": 0,
          "pred": 1,
          "n": 9
        },
        {
          "gold": 1,
          "pred": 0,
          "n": 1
        },
        {
          "gold": 1,
          "pred": 1,
          "n": 31
        }
      ],
      "cultural_response": [
        {
          "gold": 0,
          "pred": 0,
          "n": 59
        },
        {
          "gold": 0,
          "pred": 1,
          "n": 1
        },
        {
          "gold": 1,
          "pred": 0,
          "n": 5
        },
        {
          "gold": 1,
          "pred": 1,
          "n": 6
        }
      ],
      "cultural_wild": [
        {
          "gold": 0,
          "pred": 0,
          "n": 208
        },
        {
          "gold": 0,
          "pred": 1,
          "n": 7
        },
        {
          "gold": 1,
          "pred": 0,
          "n": 32
        },
        {
          "gold": 1,
          "pred": 1,
          "n": 183
        }
      ],
      "general_prompt": [
        {
          "gold": 0,
          "pred": 0,
          "n": 184
        },
        {
          "gold": 0,
          "pred": 1,
          "n": 67
        },
        {
          "gold": 1,
          "pred": 0,
          "n": 28
        },
        {
          "gold": 1,
          "pred": 1,
          "n": 321
        }
      ],
      "general_response": [
        {
          "gold": 0,
          "pred": 0,
          "n": 298
        },
        {
          "gold": 0,
          "pred": 1,
          "n": 51
        },
        {
          "gold": 1,
          "pred": -1,
          "n": 1
        },
        {
          "gold": 1,
          "pred": 0,
          "n": 43
        },
        {
          "gold": 1,
          "pred": 1,
          "n": 207
        }
      ],
      "toxicity": [
        {
          "gold": 0,
          "pred": 0,
          "n": 277
        },
        {
          "gold": 0,
          "pred": 1,
          "n": 223
        },
        {
          "gold": 1,
          "pred": 0,
          "n": 167
        },
        {
          "gold": 1,
          "pred": 1,
          "n": 333
        }
      ]
    },
    "results": {
      "3_safeguard_balanced": 45.756221653023985,
      "5_safeguard_balanced": 45.167811919656046,
      "3_safeguard_plain": 50.46500709684463,
      "5_safeguard_plain": 48.345670924773444
    }
  }
}