{
  "source_url": "https://archive.ics.uci.edu/static/public/228/sms+spam+collection.zip",
  "source_doi": "10.24432/C5CC84",
  "source_zip_sha256": "1587ea43e58e82b14ff1f5425c88e17f8496bfcdb67a583dbff9eefaf9963ce3",
  "source_rows": 5574,
  "retained_unique_messages": 5086,
  "removed_duplicate_or_empty_rows": 488,
  "conflicting_normalized_text_groups": 0,
  "seed": 42,
  "vocabulary_size": 2500,
  "split": "70% train, 15% validation, 15% held-out test; stratified after normalized-text deduplication",
  "metrics": {
    "train": {
      "n": 3560,
      "spam": 414,
      "legitimate": 3146,
      "accuracy": 0.9977528089887641,
      "spam_precision": 0.9833333333333333,
      "spam_recall": 0.9975845410628019,
      "spam_f1": 0.9904076738609112,
      "confusion_matrix": [
        [
          3139,
          7
        ],
        [
          1,
          413
        ]
      ]
    },
    "validation": {
      "n": 763,
      "spam": 89,
      "legitimate": 674,
      "accuracy": 0.9803407601572739,
      "spam_precision": 0.9204545454545454,
      "spam_recall": 0.9101123595505618,
      "spam_f1": 0.9152542372881356,
      "confusion_matrix": [
        [
          667,
          7
        ],
        [
          8,
          81
        ]
      ]
    },
    "test": {
      "n": 763,
      "spam": 88,
      "legitimate": 675,
      "accuracy": 0.9737876802096985,
      "spam_precision": 0.8777777777777778,
      "spam_recall": 0.8977272727272727,
      "spam_f1": 0.8876404494382022,
      "confusion_matrix": [
        [
          664,
          11
        ],
        [
          9,
          79
        ]
      ]
    }
  },
  "threshold_policy": {
    "spam_review": 0.8,
    "normal_queue": 0.2,
    "else": "manual triage"
  },
  "limitations": [
    "Historical English SMS, not a modern business inbox or Nigerian customer-service benchmark.",
    "Class probabilities are model scores, not independently calibrated certainty.",
    "Normalized exact duplicates removed; near-duplicates and shared message templates can remain.",
    "Random split measures within-corpus performance, not chronological or cross-domain generalization.",
    "Workflow effort and financial values are user assumptions, not measured dataset outcomes."
  ]
}