Choosing small test data for a workflow

Build a safe test set with a normal input, a messy but valid input, and a missing or malformed value. Write the expected output or failure behavior for each before running the workflow.

Add a duplicate submission test if the workflow sends messages or creates records. Use invented data with the same structure as the real source, and inspect both successful output and error routing. A huge production file is a poor first diagnostic case.

Which edge case has caused trouble before?