{
  "schema_version": 1,
  "disclosure": "AI-assisted editorial teaching records, not model predictions or an automated evidence judge. Human-review fields remain null. Verdicts apply only to the checked abstract.",
  "prepared_on": "2026-09-15",
  "records": [
    {
      "schema_version": 1,
      "record_type": "handwritten_teaching_example",
      "claim": "Generation and citation retrieval are optimized jointly.",
      "source_url": "https://arxiv.org/abs/2504.00824v2",
      "source_version": "arXiv v2, 2025-04-03",
      "evidence_location": "Abstract: training description",
      "evidence_paraphrase": "The source describes joint optimization of generation and citation.",
      "scope": {
        "checked_material": "Abstract only; not full-paper verification",
        "task_or_population": "The paper's scholarly writing and reference retrieval setting",
        "metric": "Do not equate retrieval accuracy with claim factuality",
        "conditions": "No new model run or benchmark measurement"
      },
      "source_identity_status": "arxiv_title_and_version_checked",
      "support_status": "supported_in_checked_source",
      "decision": "keep",
      "revised_claim": "Generation and citation retrieval are optimized jointly.",
      "unresolved_questions": [],
      "human_reviewer": null,
      "human_reviewed_at": null
    },
    {
      "schema_version": 1,
      "record_type": "handwritten_teaching_example",
      "claim": "The reported 40.1% is a factual-correctness rate for generated claims.",
      "source_url": "https://arxiv.org/abs/2504.00824v2",
      "source_version": "arXiv v2, 2025-04-03",
      "evidence_location": "Abstract: retrieval evaluation result",
      "evidence_paraphrase": "The source labels this number top-1 retrieval accuracy on its evaluation dataset.",
      "scope": {
        "checked_material": "Abstract only; not full-paper verification",
        "task_or_population": "The paper's scholarly writing and reference retrieval setting",
        "metric": "Do not equate retrieval accuracy with claim factuality",
        "conditions": "No new model run or benchmark measurement"
      },
      "source_identity_status": "arxiv_title_and_version_checked",
      "support_status": "metric_mismatch",
      "decision": "revise_metric_description",
      "revised_claim": "The paper reports 40.1% top-1 retrieval accuracy on its evaluation dataset.",
      "unresolved_questions": [],
      "human_reviewer": null,
      "human_reviewed_at": null
    },
    {
      "schema_version": 1,
      "record_type": "handwritten_teaching_example",
      "claim": "Every generated sentence is guaranteed to have complete evidence.",
      "source_url": "https://arxiv.org/abs/2504.00824v2",
      "source_version": "arXiv v2, 2025-04-03",
      "evidence_location": "Abstract, inspected as a whole",
      "evidence_paraphrase": "No universal sentence-support guarantee is established in this checked abstract.",
      "scope": {
        "checked_material": "Abstract only; not full-paper verification",
        "task_or_population": "The paper's scholarly writing and reference retrieval setting",
        "metric": "Do not equate retrieval accuracy with claim factuality",
        "conditions": "No new model run or benchmark measurement"
      },
      "source_identity_status": "arxiv_title_and_version_checked",
      "support_status": "not_established_by_checked_source",
      "decision": "inspect_further_or_remove_guarantee",
      "revised_claim": null,
      "unresolved_questions": [],
      "human_reviewer": null,
      "human_reviewed_at": null
    }
  ]
}