{
  "verified_date": "2026-10-02",
  "scope": "Read-only source verification; no model calls or pilot edits",
  "source_identity": {
    "requested": "google/super_glue",
    "verified_current_card": "aps/super_glue",
    "note": "Retrieved aps/super_glue rows API; observed dataset repository revision from live Hugging Face metadata. Rows API is not itself revision-pinned; normalized downloaded row hashes are authoritative for reproduction.",
    "hf_revision_observed": "3de24cf8022e94f4ee4b9d55a6f539891524d646"
  },
  "rte": {
    "validation_count": 277,
    "fields": {
      "source_text_omitted": "RTE source text omitted from public export; retrieve the original row from the linked source.",
      "retrieve_from": "https://huggingface.co/datasets/aps/super_glue/viewer/rte/validation",
      "source_idx": null
    },
    "labels": {
      "0": "entailment",
      "1": "not_entailment"
    },
    "yes_mapping": "For the question whether the premise entails the hypothesis, target_yes = 1 - label.",
    "definition": "Decide whether the premise entails the hypothesis. Non-entailment includes unsupported hypotheses as well as contradictions.",
    "provenance": "SuperGLUE reuses GLUE RTE; merges RTE1, RTE2, RTE3, RTE5 and converts to two classes.",
    "license": "No unified permissive license verified. HF card marks other and refers to original dataset licenses, with a research-context use/redistribution statement. Exact upstream RTE terms remain unverified."
  },
  "wic": {
    "validation_count": 638,
    "fields": {
      "word": "string",
      "sentence1": "string",
      "sentence2": "string",
      "start1": "int32",
      "start2": "int32",
      "end1": "int32",
      "end2": "int32",
      "idx": "int32",
      "label": "ClassLabel"
    },
    "labels": {
      "0": "False",
      "1": "True"
    },
    "yes_mapping": "For the question whether the marked word has the same meaning in both sentences, target_yes = label.",
    "definition": "Binary classification of whether two occurrences of a target noun or verb share the same sense.",
    "provenance": "Pilehvar and Camacho-Collados (2019); examples drawn from WordNet, VerbNet and Wiktionary.",
    "license": "CC BY-NC 4.0 explicitly stated by dataset authors. Do not relabel as CC BY 4.0 or Apache/MIT.",
    "offset_caveat": "Field types verified; character offset end-exclusive convention was not independently verified in this short pass. Check sentence[start:end] against intended word occurrences before rendering markers."
  },
  "sources": [
    {
      "url": "https://huggingface.co/datasets/aps/super_glue/raw/main/README.md",
      "supports": "RTE/WiC exact fields, counts, class label ordering; card license other and dataset-specific licensing statement"
    },
    {
      "url": "https://www.tensorflow.org/datasets/catalog/super_glue#super_gluerte",
      "supports": "Independent primary implementation counts and RTE provenance"
    },
    {
      "url": "https://www.tensorflow.org/datasets/catalog/super_glue#super_gluewic",
      "supports": "WiC counts, fields and source resources"
    },
    {
      "url": "https://pilehvar.github.io/wic/",
      "supports": "Author task definition, True/False examples, explicit CC BY-NC 4.0 license"
    },
    {
      "url": "https://super.gluebenchmark.com/diagnostics/",
      "supports": "Entailment vs not-entailment interpretation; this page describes diagnostics, not RTE validation provenance"
    }
  ],
  "benchmark_caveats": [
    "These are old public evaluation sets; label agreement does not establish unseen-data generalization or absence of training contamination.",
    "Keep the published labels and exact source row IDs frozen; any label audit exclusions must be separately reported.",
    "Report RTE and WiC results separately; neither is an exact event-probability ground truth task.",
    "Do not claim a blanket commercial redistribution license from the SuperGLUE wrapper."
  ]
}