Build a useful investigation brief
Build a movable exception fixture
Create ordinary records plus one value that changes a relevant type or validation condition. Place that record early, just beyond the sample, and near the end. Specify expected behavior independently and preserve each ordering, since the point of the comparison is whether position changes the inferred contract.
Capture inference and validation scope
For each placement, record the inferred schema, displayed preview, readiness message, and whether the entire file was checked. Trace which component samples and which validates. A label saying ready does not establish a full scan, while a limited visual preview does not prove validation was also limited.
Compare execution with identical settings
Run the controlled variants through the final import path and compare errors, accepted values, and any partial outcomes. Keep explicit type choices fixed where supported. If position no longer matters under the same schema, investigate inference; if execution changes settings, retain that separate handoff defect in the brief.
What to carry forward
Deliver a placement matrix, observed sampling boundary, inferred settings, and final outcomes. Identify whether the supported change concerns inference, complete validation, or honest readiness messaging. Keep resource and recovery tradeoffs visible without assuming that displaying every row is necessary to establish whole-file validity.
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