hard20 pts · first answer countsprompt-chainingevaluation

Errors flowing down a chain

A 4-step chain extracts data, transforms it, writes a report and translates it. Bad extractions in step 1 produce confident but wrong reports. What design change helps most?

A.Merge all four steps into one promptB.Make step 4 more carefulC.Add validation between steps, such as schema and sanity checks after step 1, and stop or flag the run when checks failD.Use the largest model only for translation

← Tiny wording, big swing Scores change between runs →