Module 08 Summary
The idea this module was built around
Date filtering is where careful analysts lose data without noticing, because a missing day looks exactly like a slow day. Nothing errors. The chart just dips.
What you can now do
- Write date ranges that include every intended row and no unintended ones
- Compare periods with boundaries stated explicitly rather than implied
- Prove a period split is complete by checking that the parts sum to the total
The trap this module removed
BETWEEN '2026-06-01' AND '2026-06-15' on a TIMESTAMP column includes 15 June only at 00:00:00. No order was placed at exactly midnight, so all 33 orders on 15 June disappear: BETWEEN returns 472 where the half-open form returns 505. An entire day vanished from a two-week figure, and the output looked perfectly normal.
The convention to adopt permanently
Write every date range as >= start AND < next_start. It reads slightly worse. It is correct for every period length, needs no special handling for the final day, and behaves the same whether the column stores a date or a timestamp.
Figures worth remembering
The dataset spans 2026-06-01 to 2026-06-30 - 30 days, 1,000 orders, averaging 33.33 per day. The first three days hold 102 orders; 30 June alone holds 33.
Before you move on
Find a date filter in something you maintain and check whether it uses BETWEEN on a timestamp column. If it does, you have very likely been losing the last day of every period it reports.