Establish what is happening
Reconcile one bounded import
Use a controlled file with known record identities and inspect the final outcome for each row. Separate accepted, rejected, skipped, and unknown records. Compare these with the summary counts and destination state; do not use a percentage or job status alone as proof that every source row is accounted for.
Inspect what retry selects
Record whether retry reuses the whole file, selects failed rows, or creates a corrected-file import. Compare the selected source identities with the outcome record. Check what the interface tells users about accepted rows before they confirm another attempt, especially when the corrected file retains its original filename.
Trace one already accepted record
Follow a known accepted row into a controlled retry and inspect whether it is skipped, updated, or created again under the documented rules. Keep the retry attempt and original import linked. If identity evidence is missing, classify the repeat outcome as unknown rather than assuming a safeguard prevented duplication.
What to carry forward
End with a row-outcome reconciliation, actual retry selection, and one traced accepted record. State any unknown durable outcomes explicitly. The next owner should know which work may repeat before investigating retry behavior or recommending that a user upload the entire original file again.
Technical background: Google Cloud 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