Keep the lesson for the next incident
Document each coverage claim
State which steps sample, scan, validate, and execute. Explain what ready means at each point and when users learn about later exceptions. Keep these meanings aligned with the interface so a future wording change does not accidentally turn a sampled preview into an unsupported whole-file assurance.
Retain a position-sensitive example
Keep one exceptional record that can move across the relevant sampling boundary. Record expected inference, validation, and execution outcomes for each placement. Preserve the property that caused the original difference, rather than retaining a large source file whose decisive late value becomes difficult for reviewers to identify.
Review inference and readiness changes
Assign review when sample size, schema inference, validation sequencing, or readiness copy changes. Compare the position matrix and final state. Track late failures by meaningful classification where available so a recurrence does not disappear into a generic import error count that cannot distinguish sampling from unrelated execution faults.
What to carry forward
Close follow-up with explicit coverage definitions, the movable exception fixture, and an owner for inference and readiness changes. Preserve any incomplete validation boundary in guidance. The next team should be able to explain exactly what a clean preview establishes and what remains unknown until later records are checked.
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