Establish what is happening
Inventory each source schema
Record the source file identity, ordered headings, and distinct sample values for every controlled file. Mark missing, extra, and repeated columns separately. Do not assume a shared filename pattern or producer guarantees identical layouts; compare the actual headers before treating all files as one homogeneous input set.
Compare mapping application by file
Inspect whether mapping aligns fields by position, name, or an explicit per-file rule. Compare the previewed file with the mapping used for a later file. A correct first-file preview does not establish that a second file with reordered columns receives the same semantic assignment during execution.
Trace a distinctive later-file value
Choose a value whose intended destination is unmistakable and inspect its saved field. Keep the source file and record reference attached. If the value is correct but an expected field is missing, distinguish schema union or missing-column behavior from a true positional misalignment before deciding the next investigation.
What to carry forward
End with the file-schema inventory, documented alignment rule, and first incorrect assignment. State which source file supplied the failing record. The next owner should be able to distinguish reordered columns, missing fields, and ambiguous headings without inferring the whole batch’s schema from a single representative file.
Technical background: DuckDB documentation.
Keep the decision with the work.
Use a Work Item in Aglet to record the problem, the evidence you have, and the next decision. Add an owner and priority, then keep updates in the discussion so the next person can follow the reasoning.
Create an account See the product workflow