Attribute Discovery Process

How DataPancake discovers attributes during scanning, including polymorphic detection and recursive parsing of stringified JSON.

Discovery Process

1. Recursively traverse records - Every nested level is visited
2. Track attribute paths - Complete paths from root to leaf (e.g., customer.contact.email)
3. Identify data types - Source data type inferred for each path occurrence
4. Create all 7 polymorphic versions - Proactively created when attribute first discovered
5. Activate matching versions - Only versions matching discovered types set to VERSION_STATUS = 'active'
6. Create attribute records - Records created in core.datasource_attribute and core.datasource_attribute_polymorphic_version

---

What Gets Discovered

• All attribute paths - Complete paths from root to leaf (e.g., customer.contact.email)
• Nested objects - Every level of object nesting
• Nested arrays - Both object arrays (ARRAY_TYPE = 'object') and primitive arrays (ARRAY_TYPE = 'primitive')
• Embedded JSON - JSON stored as strings (HAS_EMBEDDED_CONTENT = TRUE); recursively parsed
• Polymorphic variations - All data type variations for the same path (handled via polymorphic versions)

---

Example: Polymorphic Discovery

Record 1:

{
  "customer_id": "C001",
  "customer": {
    "address": "123 Main St, Anytown, ST 12345",
    "metadata": "{\"source\":\"web\",\"tags\":[\"vip\"]}"
  }
}

Record 2:

{
  "customer_id": "C002",
  "customer": {
    "address": {"street": "456 Oak Ave", "city": "Springfield"},
    "metadata": "{\"source\":\"mobile\",\"tags\":[\"new\"]}"
  }
}

Discovery Results:

• customer.address:

  • Record 1: str → address_str activated
  • Record 2: object → address_object activated (existing version)
• customer.metadata:

  • HAS_EMBEDDED_CONTENT = TRUE
  • Recursively parsed to discover: metadata.source, metadata.tags[]