Fix cartesian product in agg_monthly_loans causing inflated metrics

[Copilot]

agg_monthly_loans was producing 1,162 rows (expected 10–30) and inflated loan values due to a LEFT JOIN back to the raw loans table on only loan_type_name. Since monthly_originations is already aggregated at month + loan_type grain, this join multiplied every aggregated row by every individual loan of that type.

Changes

• models/marts/agg_monthly_loans.sql: Remove the LEFT JOIN loans ON orig.loan_type_name = loans.loan_type_name from the combined CTE and drop the loans.customer_id column from the SELECT

-- Before (combined CTE)
from monthly_originations orig
full outer join monthly_payments pay
    on orig.month_start = pay.month_start
left join loans                                     -- ❌ cartesian product
    on orig.loan_type_name = loans.loan_type_name

-- After
from monthly_originations orig
full outer join monthly_payments pay
    on orig.month_start = pay.month_start

The loans CTE is retained — it's still used by monthly_originations for the initial aggregation.

Original prompt

The agg_monthly_loans model has a SQL bug that causes duplicate rows and inflated metrics in the Risk Analytics Reporting Dashboard.

Problem:
The model performs a LEFT JOIN loans ON orig.loan_type_name = loans.loan_type_name which creates a cartesian product. Since monthly_originations is already aggregated by month + loan_type, this join matches each aggregated row to EVERY individual loan of that type, creating duplicates.

Impact:

• Data quality assertion failing: Expected 10-30 rows, actual 1,162 rows
• February 2026 has 81 duplicate rows instead of ~5-10
• Loan values artificially inflated (e.g., $339M for Feb 2026)
• Risk Analytics Reporting Dashboard showing incorrect, inflated loan origination values

Root Cause:
PR #3 added this join to include customer_id, but the join logic is incorrect for an aggregated table.

Required Fix:
Remove the problematic LEFT JOIN to the loans table in models/marts/agg_monthly_loans.sql. The aggregation should remain at the month + loan_type grain without attempting to join back to individual loan records.

If customer-level detail is truly needed, the model needs to be redesigned to aggregate at the customer + month + loan_type grain, but based on the current model structure and data quality assertion (expecting 10-30 rows), the intended grain is monthly aggregation only.

Files to modify:

• models/marts/agg_monthly_loans.sql: Remove the LEFT JOIN loans and the customer_id column from the final select

---

🔒 GitHub Advanced Security automatically protects Copilot coding agent pull requests. You can protect all pull requests by enabling Advanced Security for your repositories. Learn more about Advanced Security.