Level 7 · Debug: more paid-shipping orders than exist

Why the number is impossible

Laura’s check is the best kind of debugging: a bound. 3,977 orders paid any shipping at all, so “web and app orders that paid shipping” must be a smaller number. 6,867 is bigger, so the filter is letting in rows it shouldn’t. Before you touch the code, ask which rows could be sneaking in.

The answer is in how Python groups & and |. Just like * happens before + in arithmetic, & happens before |. The starter’s condition

(orders["channel"] == "web") | (orders["channel"] == "app") & (orders["shipping_fee"] > 0)

is read by Python as

(orders["channel"] == "web") | ((orders["channel"] == "app") & (orders["shipping_fee"] > 0))

That’s “every web order, plus app orders that paid shipping”. There are 5,913 web orders and 954 paid-shipping app orders: 5,913 + 954 = 6,867, exactly the wrong number in the draft.

The fix: group the “or” yourself

Put the “web or app” part in its own parentheses so it’s decided first:

web_or_app = (orders["channel"] == "web") | (orders["channel"] == "app")
paid = orders["shipping_fee"] > 0
orders[web_or_app & paid]

Naming the masks, as above, makes the logic easy to read and impossible to group wrongly. A shorter option avoids | altogether:

orders[orders["channel"].isin(["web", "app"]) & (orders["shipping_fee"] > 0)]

Both give 2,210 orders: 1,256 from web and 954 from the app. Marketplace orders are excluded even though 1,767 of them paid shipping, which is a big share: marketplace fees follow different rules.

Checking a fix

After a fix, test the pieces separately: count the web-or-app orders (10,233), count the paid ones (3,977), and confirm the combined filter is no bigger than either. Two quick numbers like these catch most filter bugs. You can count a mask directly with .sum(), because True counts as 1, so you don’t even need to filter the table to check each piece. Then add the per-channel counts back together and make sure they match the combined number.

Common mistakes

• Trusting the left-to-right reading. a | b & c means a | (b & c).
• Leaving out the parentheses around each comparison: orders["channel"] == "web" | ... fails with a TypeError, because | binds tighter than ==.
• Using and / or on Series, which raises ValueError: The truth value of a Series is ambiguous.
• Returning the filtered DataFrame instead of its row count. The ticket asks for a number.