{
  "schemaVersion": 1,
  "study": "cold-email-reply-classification-study",
  "title": "Reply classification study: study protocol",
  "status": "awaiting-real-data",
  "scope": "Collection template, not a dataset or completed field study. All null values are intentionally uncollected.",
  "unitOfObservation": "first-human-reply",
  "requiredInputs": "Redacted reply excerpts and two independent human reviewers.",
  "controls": "Hide campaign totals and the other reviewer label; retain ambiguous and automated categories.",
  "primaryAnalysis": "Label disagreement matrix before adjudication; this is not model accuracy.",
  "registration": {
    "owner": null,
    "protocolVersion": "1",
    "registeredAt": null,
    "population": null,
    "eligibilityRules": null,
    "primaryOutcomeDefinition": null,
    "denominator": null,
    "observationWindow": null,
    "targetSampleAndJustification": null,
    "assignmentProcedureAndSeed": null,
    "exclusionRules": null,
    "safetyStopConditions": null,
    "missingDataPolicy": null,
    "retentionAndAccessOwner": null
  },
  "collectionRules": [
    "Assign pseudonymous IDs; do not store mailbox passwords, access tokens or unnecessary identifiers.",
    "Freeze the outcome and exclusions before collection; retain exclusions with reasons.",
    "Do not use repeated observations as independent people. Record clustering by person, company or mailbox.",
    "Do not treat a software test, simulated reply or researcher-written example as participant or campaign evidence.",
    "Record complaints and opt-outs and follow the actual suppression workflow.",
    "For interviews obtain permission before recording and respect withdrawal.",
    "Publish aggregate or redacted evidence only after review; keep original private records outside the public website."
  ],
  "fieldDictionary": {
    "thread_id": {
      "required": true,
      "type": "string, number or boolean as specified by the registered protocol; null until collected",
      "meaning": "thread id",
      "missingValue": null
    },
    "reply_excerpt": {
      "required": true,
      "type": "string, number or boolean as specified by the registered protocol; null until collected",
      "meaning": "reply excerpt",
      "missingValue": null
    },
    "reviewer_a_label": {
      "required": true,
      "type": "string, number or boolean as specified by the registered protocol; null until collected",
      "meaning": "reviewer a label",
      "missingValue": null
    },
    "reviewer_b_label": {
      "required": true,
      "type": "string, number or boolean as specified by the registered protocol; null until collected",
      "meaning": "reviewer b label",
      "missingValue": null
    },
    "adjudicated_label": {
      "required": true,
      "type": "string, number or boolean as specified by the registered protocol; null until collected",
      "meaning": "adjudicated label",
      "missingValue": null
    },
    "disagreement_reason": {
      "required": true,
      "type": "string, number or boolean as specified by the registered protocol; null until collected",
      "meaning": "disagreement reason",
      "missingValue": null
    },
    "automated_flag": {
      "required": true,
      "type": "string, number or boolean as specified by the registered protocol; null until collected",
      "meaning": "automated flag",
      "missingValue": null
    }
  },
  "blankRecord": {
    "observation_id": null,
    "thread_id": null,
    "reply_excerpt": null,
    "reviewer_a_label": null,
    "reviewer_b_label": null,
    "adjudicated_label": null,
    "disagreement_reason": null,
    "automated_flag": null,
    "excluded": false,
    "exclusion_reason": null,
    "protocol_deviation": null
  },
  "records": [],
  "analysisStatus": "not-run",
  "results": null,
  "interpretationLimits": "Do not force automated replies into a positive or negative category. Keep opt-outs distinct from ordinary declines. Publish results only after the evidence, method and limitations have been reviewed. This protocol provides no benchmark, expected lift or completed-study claim.",
  "procedure": [
    "Draft category definitions, boundary examples and an ambiguous option. Remove unnecessary personal data before annotation.",
    "Have two reviewers label independently without campaign performance totals. Adjudicate disagreements only after preserving both original labels.",
    "Before collection, write the primary outcome, observation window, exclusion rules and stopping conditions. Preserve excluded observations with a reason rather than quietly removing them.",
    "Pilot the procedure with fictional or owned test data, resolve ambiguous fields, and freeze a dated protocol version before the main run."
  ],
  "interviewPrompts": []
}