Validate a join before you trust the numbers

Question

Revenue looked 30% too high after a join because the customer table had duplicate IDs.

Write `safe_merge(orders, customers)` that returns a dict with keys `rows`, `unmatched` and `duplicate_keys`: the row count of a left join on `customer_id`, the number of orders with no matching customer, and the

Accepted Answer

Step 1: Count duplicate keys on the right table *before* merging — that is the number that explains any unexpected row growth. Step 2: `how='left'` keeps every order even when no customer matches, so nothing silently disappears from revenue. Step 3: `indicator=True` adds a `_merge` column; counting `left_only` gives the unmatched orders without a second pass. Step 4: Casting to `int` returns plain Python numbers rather than NumPy scalars, which serialise cleanly into reports.

Validate a join before you trust the numbers

More Pandas — Joins questions

Asked at Razorpay