Establish what is happening
Identify the preview’s coverage
Record the number and positions of displayed or analyzed records, any stated sampling limit, and the validation status shown to users. Distinguish a visual sample from a full-file validation. A preview can legitimately show fewer rows while still validating the entire file, so verify rather than assume its scope.
Find the first later exception
Locate a controlled record beyond the sampled region that differs in type, length, structure, or required value. Compare it with an earlier accepted record and note the exact property that changes. Keep the original ordering because moving the exception can alter the inference that caused the failure.
Compare preview and execution rules
Capture the selected schema, parser settings, and final error or changed value. Determine whether execution uses the same rules and simply encounters new data, or whether settings change between stages. Those explanations require different follow-up even when both present as a preview that looked fine.
What to carry forward
Finish with the preview coverage, first excluded exception, relevant setting, and observed execution result. State whether full-file validity was actually claimed. The next owner should know whether to investigate sampling, inference, validation scope, or a separate handoff mismatch between preview and import.
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