{
  "suite": "SX-TXT-1.0",
  "seed": 20261031,
  "corpus_sha256": "699edf9459be5835d4a17f1fd0dad485af176f0dde7d7bb8c323ef3f23e6288e",
  "ground_truth_unit": "Python Unicode code-point offsets; end exclusive",
  "records": 1200,
  "planted_spans": 7960,
  "entity_types": 12,
  "entity_counts": {
    "ACCOUNT_NUMBER": 400,
    "DOB": 200,
    "EMAIL_ADDRESS": 1240,
    "IP_ADDRESS": 400,
    "LOCATION_ADDRESS": 200,
    "NAME_FAMILY": 1440,
    "NAME_GIVEN": 1840,
    "ORGANIZATION": 600,
    "PASSWORD": 400,
    "PHONE_NUMBER": 640,
    "USERNAME": 400,
    "US_SSN": 200
  },
  "families": {
    "email": 200,
    "chat": 200,
    "support": 200,
    "crm_note": 200,
    "engineering_issue": 200,
    "call_transcript": 200
  },
  "fictional_people": 72,
  "utility_probes": 1600,
  "standard_records": 1080,
  "challenge_records": 120,
  "challenge_types": {
    "spaced_email": 40,
    "unicode_confusable": 40,
    "spoken_phone": 40
  },
  "data_origin": "Deterministic templates and fictional profiles. No supplier records. Names may coincide with real names by chance.",
  "limitations": [
    "Template-based extracted English text; not native-file reconstruction or audio/image testing.",
    "Synthetic invalid SSNs and documentation IPs are deliberate fixtures, not representative natural prevalence.",
    "Public challenge corpus is for reproducibility, not a private production holdout.",
    "Repeat-value consistency is narrower than person-level alias resolution."
  ],
  "weights": {
    "NAME_GIVEN": 5,
    "NAME_FAMILY": 5,
    "EMAIL_ADDRESS": 5,
    "USERNAME": 2,
    "PHONE_NUMBER": 5,
    "ORGANIZATION": 2,
    "LOCATION_ADDRESS": 5,
    "DOB": 2,
    "IP_ADDRESS": 2,
    "ACCOUNT_NUMBER": 10,
    "US_SSN": 10,
    "PASSWORD": 10
  }
}