CrossER / data /schema.md
smurthy5's picture
Upload data/schema.md with huggingface_hub
87106bc verified

CrossER v2 Data Schema

entities.json

{
  "entity_id": "tc2_prod_0001",
  "source_system": "SAP_TC2",
  "entity_type": "product",
  "name": "Maltodextrin DE20 Grade A",
  "attributes": {
    "category": "starches",
    "sub_category": "maltodextrin",
    "cn_code": "CN 1702.90",
    "dextrose_equivalent": 20,
    "grade": "A",
    "region": "NA",
    "cost_center": "CC-412",
    "status": "active"
  }
}

Source Systems

System ID Prefix Naming Style Completeness
SAP_TC2 tc2_ Formal English 95%
SAP_CFIN cfin_ Internal codes 90%
SAP_APAC apac_ Abbreviated 60%
LEGACY_ERP leg_ Cryptic codes 45%
SHAREPOINT sp_ Authoritative 85%

Entity Types

  • product — Physical products (starches, sugars, acids, proteins, oils)
  • supplier — Vendor/supplier companies
  • tax_code — Tax classification codes (CN/HS codes)
  • legal_entity — Legal entities / subsidiaries

pairs.json

{
  "pair_id": "pair_0001",
  "entity_a": "tc2_prod_0001",
  "entity_b": "cfin_prod_0001",
  "label": "match",
  "string_similarity": 0.234,
  "attribute_overlap": 0.500,
  "difficulty_tier": "hard",
  "entity_type": "product",
  "system_pair": ["SAP_CFIN", "SAP_TC2"]
}

Labels

  • match — Same real-world entity
  • no_match — Different entities
  • ambiguous — Requires human judgment

Difficulty Tiers

  • easy — String sim >= 0.70
  • medium — String sim 0.30-0.70
  • hard — String sim < 0.30
  • adversarial_negative — High similarity but NOT a match
  • medium_negative / obvious_negative — No-match tiers

Splits & Subsets

  • splits/{train,val,test}.json — 40/20/40 stratified split
  • subsets/crosser_easy.json — Easy-tier test pairs
  • subsets/crosser_hard.json — Hard + adversarial test pairs

Oracle Context

  • context/structured/oracle_context.json — Complete pairwise migration records
  • context/structured/alias_table.json — Canonical entity alias groups

Demo Subset

  • demo/*.csv — Per-system entity CSVs
  • demo/ground_truth.csv — Labeled pairs
  • demo/rules.yaml — Business rules
  • demo/knowledge.yaml — Domain knowledge artifacts

Extended Pipelines (--pipeline full)

Email Chains (email_chains/)

// chains.json — array of email threads
{
  "chain_id": "email_chain_0001",
  "domain": "commodity_trading",
  "messages": [
    {
      "message_id": "email_chain_0001_msg_001",
      "sender": "s.chen@globalfoods.com",
      "sender_name": "Sarah Chen",
      "sender_role": "Operations Manager, North America",
      "recipients": ["j.morrison@globalfoods.com"],
      "subject": "RE: Is Maltodextrin DE20 Grade A in SAP_TC2 the same as record in SAP_CFIN?",
      "body": "...",
      "timestamp": "2025-07-15T09:00:00"
    }
  ]
}

Email Prediction Format

[
  {
    "chain_id": "email_chain_0001",
    "extracted_knowledge": [{"type": "entity_mapping", "description": "..."}],
    "extracted_relationships": [{"type": "discussed_in_context", "entities_involved": 2}],
    "extracted_corrections": [{"type": "expert_correction", "source_message_ids": ["..."]}]
  }
]

Support Tickets (tickets/)

// tickets.json
{
  "ticket_id": "TKT-0001",
  "domain": "commodity_trading",
  "title": "Product ID mismatch between SAP_TC2 and SAP_CFIN",
  "priority": "high",
  "resolution": {
    "resolution_type": "entity_mapping",
    "verdict": "confirmed_match",
    "confidence": 0.95,
    "reasoning": "..."
  },
  "referenced_precedents": ["TKT-0003"]
}

Ticket Prediction Format

[
  {
    "ticket_id": "TKT-0001",
    "predicted_resolution": "confirmed_match",
    "referenced_precedent_ids": ["TKT-0003"],
    "extracted_knowledge": [{"type": "entity_mapping", "description": "..."}]
  }
]

ERP Events (erp_events/)

// events.json
{
  "event_id": "EVT-000001",
  "timestamp": "2025-01-15T10:30:00",
  "source_system": "SAP_TC2",
  "event_type": "purchase_order_created",
  "entity_refs": [
    {"entity_type": "product", "id": "TC2-1234", "name": "Maltodextrin DE20 Grade A"},
    {"entity_type": "supplier", "id": "SUP-12345", "name": "Cargill Inc."}
  ],
  "attributes": {"amount": 50000.00, "currency": "USD"},
  "related_events": ["EVT-000002"]
}

ERP Prediction Format

[
  {
    "event_id": "EVT-000001",
    "linked_entity_ids": ["EVT-000002"],
    "normalized_event_type": "purchase_order_created"
  }
]

Cases (cases/)

// cases.json
{
  "case_id": "CASE-0001",
  "case_type": "entity_disambiguation",
  "domain": "commodity_trading",
  "linked_entities": [...],
  "resolution": {"verdict": "approved", "confidence": 0.9, "evidence": [...]},
  "cited_precedents": ["CASE-0003"]
}

Case Prediction Format

[
  {
    "case_id": "CASE-0001",
    "predicted_precedent_ids": ["CASE-0003", "CASE-0007"],
    "predicted_resolution": "approved",
    "context_entities": ["entity_a", "entity_b"]
  }
]

Corrections (corrections/)

// sequences.json
{
  "sequence_id": "CORR-0001",
  "pattern_id": "entity_mapping_fix",
  "domain": "commodity_trading",
  "steps": [
    {"step_index": 0, "step_type": "system_proposes_mapping", ...},
    {"step_index": 1, "step_type": "expert_reviews", ...},
    {"step_index": 2, "step_type": "expert_rejects", ...},
    {"step_index": 3, "step_type": "expert_provides_correct_mapping", ...}
  ]
}

Correction Prediction Format

[
  {
    "sequence_id": "CORR-0001",
    "generated_rules": [{"type": "override_mapping", "condition": "...", "action": "..."}],
    "updated_mappings": [{"entity_a": "...", "entity_b": "...", "is_match": true}]
  }
]

Knowledge Documents (knowledge_docs/)

// manifest.json
{
  "doc_id": "KDOC-0001",
  "filename": "KDOC-0001_concept_commodity_trading.md",
  "title": "Concept: Dextrose Equivalent and Its Impact on Classification",
  "doc_type": "concept",
  "domain": "commodity_trading"
}

Knowledge Prediction Format

[
  {
    "doc_id": "KDOC-0001",
    "extracted_artifacts": [
      {"type": "concept", "content": "...", "entities": [...], "confidence": 0.95}
    ]
  }
]