Split versus sample
Splitting is a partition: the files together contain every row once. The example splits a sales file by region into North, South, East and West files, packed into an uncompressed ZIP built in your browser.
Sampling is different: it keeps a subset. The recipe includes a disabled “Sample rows” step that draws a 50% sample stratified by region with the seed “q1-audit”. Enable it and the same rows are chosen every time. Sample output is labelled as a sample wherever it appears.
Questions
How are file names chosen?
From the column value, with characters that are unsafe in file names replaced by underscores. Blank and missing values get their own files, named (blank) and (missing).
Is there a limit on the number of files?
A preflight limit (200 by default, adjustable) stops the split before it creates thousands of files from a high-cardinality column.
Can I split the output of a recipe?
Yes. Select a step in the recipe and the split uses that step’s output.