Observe the job, not just the final answer

A final answer can look sensible while the job used an outdated record or skipped a required check. Record the job identifier, the input version, the actions requested, their results, and the final state. These facts let an operator explain what happened without storing every conversation indefinitely.

Keep application events separate from transport events. A successful HTTP response says the endpoint answered. A completed business event says the intended work finished. A queue acknowledgment says a message was accepted. These are different observations.

Build an example failure into the first release. Reject one action in a development environment and verify that the job reports failure, keeps the information needed for recovery, and does not silently report completion.

Choose retention deliberately. Store the smallest evidence needed to support the workflow, restrict access, and delete old records on a documented schedule. Logs are operational data, not an excuse to copy every credential and customer message into another system.