Level 6 · Stack two quarterly exports

concat stacks, merge matches

Both functions combine tables, but they answer different questions. merge lines rows up side by side by a key, like attaching a customer’s name to each order. concat doesn’t look at the contents at all: it puts one table under the other. Use it when the tables have the same columns and hold different rows, such as two quarters of the same export.

h1 = pd.concat([q1, q2], ignore_index=True)

The tables go in a list, in the order you want them stacked. Columns are matched by name; a column that exists in only one table is filled with NaN in the other’s rows, which is a useful warning sign that the exports don’t have the same shape.

Why ignore_index=True

q1 and q2 are filtered slices of orders, so they keep their original row labels. Without ignore_index=True, the stacked table carries those old labels, and in general two tables can share labels, which makes .loc lookups return two rows instead of one. ignore_index=True throws the old labels away and numbers the result 0, 1, 2, … from scratch. For two exports from different files, where both start at 0, it’s essential.

Check the length

The safest one-line test after a concat is that no rows were lost or added:

len(q1), len(q2), len(h1)          # (857, 1022, 1879)
assert len(h1) == len(q1) + len(q2)

assert stops the code with an error if the condition is False, so a broken stack can’t slip into the report. Q2 has about 19% more orders than Q1, which fits the store’s spring and summer pattern.

The summary itself is a groupby you know: over the first half of 2025, web brings 817 orders, the app 796 and the marketplace 266. Marketplace orders are only about one in seven, but their shipping adds up to $1,559.59, close to web’s $1,817.55, because marketplace fees follow different rules.

Common mistakes

• Passing the tables without a list: pd.concat(q1, q2) raises a TypeError.
• Forgetting ignore_index=True, so labels can repeat after stacking.
• Using merge to stack rows; it matches keys instead of appending.
• Skipping the length check, which is the quickest proof that the stack is complete.