{
  "design": "Purposively selected endpoint anchors, not a systematic or representative literature sample.",
  "unit": "One paper per row; classification refers to the contrasts described in the row, not every experiment in the paper.",
  "fields": {
    "endpoint_class": "Human judgment about what the selected contrast actually measures. The categories are descriptive and not ordered evidence grades.",
    "normal_primary_cells_at_start": "True only for the selected experimental program that starts with normal freshly explanted cells before introducing oncogenic signaling. False does not assert that all cells or tissue in a paper are malignant.",
    "clinical_incident_cancer_endpoint": "True only if a human intervention evaluates occurrence of new cancer, not tissue markers, animal tumors, or transplanted malignant cells."
  },
  "coding": "One GPT-family reader extracted the table and checked source locators. No independent second coding; counts reproduce the extraction rather than verify its correctness.",
  "uncertainty": "No sampling intervals for paper counts: selection is purposive. Source judgments are contestable and locators permit correction."
}