What ordered_at actually contains in retail-orders

Working through retail-orders again. ordered_at is the column people trip over, so here is what it actually looks like.

7,000 rows, no nulls, 4,159 distinct values. Order placement time, UTC.

It runs from 2023-01-01 08:28:00 to 2024-12-30 21:58:00, covering 727 distinct days.

It is a timestamp, not a date, which means every comparison against a bare date is a comparison against midnight. ordered_at <= date '2023-06-30' silently excludes almost all of 30 June. Half-open ranges — >= start and < end — avoid the whole class of off-by-one-day bugs and read no worse.

select
  count(*)                     as rows,
  count(ordered_at)            as present,
  count(*) - count(ordered_at) as nulls,
  count(distinct ordered_at)   as distinct_values
from retail_orders;

Check the range before filtering on it. Half the "the dashboard is empty" reports we have seen are a date filter outside the data's actual range, and the query is not wrong so nothing errors.

Where this bites: 727 populated days is what any window function over this column has to work with. A seven-day lag counts rows, not days — so if a day is missing, lag(7) quietly compares against eight days ago and the week-over-week number is wrong in a way that looks plausible.

The grain is one row per order line, which is the context every one of those numbers depends on. None of them survive a change of grain, which is why "profile the column" and "profile the table" are the same job. Full schema, and the CSV, on the dataset page.