{
  "schema": "horizon.ifeval-publication.v1",
  "editionId": "olmoe-0924-ifeval541-v1",
  "state": {
    "validationStatus": "valid",
    "evidenceLevel": "E1",
    "experimentType": "standalone_benchmark",
    "publicationStatus": "public",
    "reproductionStatus": "not_attempted"
  },
  "ownerApproval": "2026-09-16: owner explicitly authorized publication on horizonrunmap.com",
  "sourceManifestSha256": "26a7973d6d6047e2fa129284d736c78006b2755a0ee68da6bd6f6312e5332e9a",
  "contractSha256": "b230f0b5e7e9155023c42145880035a83d6da26e27599ef4e7e325d7b969c2b6",
  "sourceContractNote": "The immutable experiment contract retains its original internal_only status. This successor records the later public admission.",
  "model": "allenai/OLMoE-1B-7B-0924-Instruct",
  "revision": "7f1c97f440f06ce36705e4f2b843edb5925f4498",
  "scope": "Absolute instruction-following observations for the complete 541-prompt population. External values are contextual, non-controlled references.",
  "preservationClaimAllowed": false,
  "seedPolicy": {
    "python_random": 0,
    "langdetect": 0
  },
  "verificationDependencies": {
    "immutabledict": "4.2.2",
    "langdetect": "1.0.9",
    "nltk": "3.9.2",
    "packaging": "26.2",
    "six": "1.17.0",
    "click": "8.4.2",
    "joblib": "1.6.0",
    "regex": "2026.7.19",
    "tqdm": "4.70.0"
  },
  "externalReferences": [
    {
      "value": 48.1,
      "label": "Original 0924 model card",
      "metric": "Loose Acc",
      "url": "https://huggingface.co/allenai/OLMoE-1B-7B-0924-Instruct#evaluation-snapshot"
    },
    {
      "value": 45.29,
      "label": "0924 checkpoint in the 0125 model card",
      "metric": "IFEval (variant unspecified in table)",
      "url": "https://huggingface.co/allenai/OLMoE-1B-7B-0125-Instruct#performance"
    }
  ]
}