This lesson's interactive features are locked, please to keep using them

Grain Validation

The grain of a table is what one row represents. An orders table might have one row per order, or one row per line item. Getting this wrong leads to double-counting and incorrect numbers in every report that uses the table.

Why Grain Matters

If your orders table has a grain of "one row per order" but accidentally contains duplicate order IDs, every aggregation will be inflated:

# This looks fine...
orders = pd.DataFrame(
    {
        "order_id": ["ORD-7K2M", "ORD-9P4X", "ORD-9P4X", "ORD-3D8V"],
        "revenue": [50, 75, 75, 100],
    }
)

# But total revenue is wrong
print(orders["revenue"].sum())  # 300, but it should be 225

Order ORD-9P4X appears twice, and revenue is overcounted by $75. This is a grain violation.

Validating

Use .duplicated() to check whether your grain is valid:

# Check for duplicate order IDs
duplicates = orders[orders.duplicated(subset=["order_id"], keep=False)]
print(f"Found {len(duplicates)} rows with duplicate order_ids")

If you get zero duplicates, your grain is valid.

Multi-Column Grain

Sometimes multiple columns define the grain. An order line item table might have a grain of "one row per order + product combination":

line_items = pd.DataFrame(
    {
        "order_id": ["ORD-7K2M", "ORD-7K2M", "ORD-9P4X", "ORD-9P4X"],
        "product_id": ["CHL-482", "TST-731", "CHL-482", "OVN-205"],
        "quantity": [2, 1, 3, 1],
    }
)

# Validate: no duplicate order_id + product_id combinations
dupes = line_items.duplicated(subset=["order_id", "product_id"])
print(f"Grain violations: {dupes.sum()}")  # 0, so the grain is valid