Three years of daily sales, with weekends and seasons in them, that regenerate byte for byte.

Testing a forecasting job, a dashboard, or a data pipeline needs a sales series long enough to have seasons and honest enough to have bad weeks. Synthetic time series tools give you a sine wave plus noise. Overwater’s daily sales are the sum of individual orders from individual customers, so weekends are lower because fewer people shop, spring is higher because plants sell in spring, and the level wanders between years because the business had good and bad quarters. The whole thing is a pure function of one seed. Regenerate it and every order id and timestamp comes back the same, so a test that passed yesterday is testing the same data today.

What you should see

Roughly a thousand daily rows. Saturdays and Sundays run below the weekdays around them. Every March to May runs far above the winter before it, and the weekly average ends each year at a different level. Run the download again next month and the historical rows have not changed. Only new days have been appended.