Data Model¶

This section documents the data schemas used throughout the RWA calculator, including input requirements, intermediate structures, and output formats.

Overview¶

The calculator uses Polars schemas to define and validate data at each stage:

flowchart LR
    A[Input Schemas] --> B[Intermediate Schemas] --> C[Output Schemas]

Schema Categories¶

Input Schemas¶

Define the structure of data loaded from external sources:

• Input Schemas
• Counterparty Schema
• Facility Schema
• Loan Schema
• Contingent Schema
• Collateral Schema
• Guarantee Schema
• Provision Schema
• Rating Schema
• FX Rates Schema
• Specialised Lending Schema
• Equity Exposure Schema
• Mapping Schemas (Facility, Org, Lending)

Data Validation¶

Validation rules and troubleshooting:

• Data Validation Guide
• Schema Validation Functions
• Business Rule Validators
• Common Errors and Fixes
• Debugging Tips

Intermediate Schemas¶

Define internal data structures used during processing:

• Intermediate Schemas
• Resolved Hierarchy Schema
• Classified Exposure Schema
• CRM Adjusted Schema

Output Schemas¶

Define the structure of calculation results:

• Output Schemas
• SA Result Schema
• IRB Result Schema
• Slotting Result Schema
• Aggregated Result Schema

Regulatory Tables¶

Lookup tables for risk weights and parameters:

• Regulatory Tables
• Risk Weight Tables
• CCF Tables
• Haircut Tables
• Slotting Tables
• F-IRB LGD Tables

Schema Usage¶

Validation¶

Schema shape is enforced by the loader edge seal, not by a validator you call: a missing required column becomes a typed null plus a DQ001 error, and declared columns are cast to their declared dtype. What you call explicitly is the input-domain gate.

from rwa_calc.contracts.validation import validate_bundle_values

# Categorical and numeric column domains across the whole raw bundle
errors = validate_bundle_values(bundle)

for error in errors:
    print(f"{error.code} {error.field_name}: {error.message}")

Both pipeline entries run this already — the file loader attaches its errors to the returned RawDataBundle, and PipelineOrchestrator.run_with_data runs it for in-memory bundles. Call it directly only when validating a bundle outside a run.

Type Checking¶

import polars as pl
from rwa_calc.data.schemas import FACILITY_SCHEMA

# Create DataFrame with explicit schema
facilities = pl.DataFrame({
    "facility_reference": ["F001", "F002"],
    "counterparty_reference": ["C001", "C001"],
    "limit": [1_000_000.0, 500_000.0],
}).cast(FACILITY_SCHEMA)

Data Types¶

Nullable Fields¶

Optional fields are marked with | None in schemas:

COUNTERPARTY_SCHEMA = {
    "counterparty_reference": pl.String,  # Required
    "annual_revenue": pl.Float64,         # Optional (can be null)
    "entity_type": pl.String,             # Required
}

Next Steps¶

• Input Schemas - Required input data formats
• Data Validation Guide - Validation and troubleshooting
• Output Schemas - Result data formats
• Regulatory Tables - Lookup table reference