Worked exercise: debug with an assistant

Debugging rounds test whether you form hypotheses and check them, rather than asking the assistant to "fix the bug".

The report

Users report two problems with the "last 7 days" revenue chart: the bar for today is always empty, and some late-evening orders appear under the following day. The function is:

def daily_revenue(orders, today, days=7):
    start = today - timedelta(days=days)
    totals = defaultdict(int)
    for o in orders:
        if start < o.created_at.date() < today:
            totals[o.created_at.date()] += o.amount_cents
    return [totals[start + timedelta(days=i)] for i in range(1, days + 1)]

A strong session

1. Reproduce it with a test before touching the code:

def test_includes_today():
    today = date(2026, 3, 10)
    orders = [Order(created_at=datetime(2026, 3, 10, 9), amount_cents=500)]
    assert daily_revenue(orders, today)[-1] == 500

2. State hypotheses: the filter excludes today; time zones shift late orders to the next day; the output window is off by one.
3. Check them one at a time. The test fails because < today excludes today, while the returned window ends at start + 7 days = today.
4. Fix the comparison to start < day <= today and rerun. That explains the empty bar but not the second symptom: created_at is stored in UTC while "today" is the user's local date, so an order at 21:00 in New York is dated the next day.
5. Propose converting timestamps to the user's zone before taking .date(), and add a test for an order at 23:30 local time.

An assistant asked to "fix the bug" will usually change the comparison and stop; the interview signal is finding the second cause.