{
  "schemaVersion": 1,
  "generatedAt": "2026-09-06",
  "title": "English-name collisions among Chinese establishments in the US FDA device registration listing",
  "source": "openFDA device/registrationlisting count aggregations (public, no authentication)",
  "frame": "Every device registration-listing record whose registration.iso_country_code is CN. No sampling. City is a complete partition of the frame: 971 city buckets covering 42098 records.",
  "method": "Counted distinct registration numbers and distinct establishment names by city batch, then took the union. Names normalised with the same rule as the NHTSA vPIC collision census so the two frames are comparable: uppercase, punctuation to spaces, collapse whitespace, drop CO/LTD/LIMITED/INC/CORP/CORPORATION/COMPANY/GROUP/MANUFACTURING/MFG. No synonym or romanisation merging.",
  "batches": 45,
  "listingRecords": 42098,
  "cityBuckets": 971,
  "establishments": 5012,
  "distinctRawNames": 5010,
  "distinctNormalisedNames": 4993,
  "establishmentsInACollisionGroupLowerBound": 19,
  "shareInACollisionGroupLowerBoundPercent": 0.4,
  "comparison": {
    "vpicFrame": "assets/datasets/vpic-china-name-collisions-2026-09-06.json",
    "note": "Same normalisation rule. The vPIC frame is China-registered vehicle manufacturers; this frame is China-registered medical-device establishments. Two unrelated US registration processes."
  },
  "limits": [
    "聚合接口拿不到「名字 ↔ 注册号」的配对,所以这里给的是碰撞的**下界**:E 家企业归一化后得到 N 个不同名字,处在碰撞组里的至少 E − N 家。真实碰撞数只会更高,不会更低。",
    "openFDA 里的名字是企业自报的英文名,不是中国登记簿上的法定名称。这里的碰撞不证明登记簿里也碰撞。",
    "归一化只做大小写、标点、空白与常见后缀。拼写差一个字母就算两个名字,所以这是碰撞率的下界之下界。",
    "框架是向美国 FDA 注册过器械的中国主体,不代表中国制造商总体。",
    "一家企业若在 openFDA 里有多个注册号,会被计成多家;registration_number 是这套数据的企业主键。"
  ]
}