Retries at three layers: give the decision one owner
The library retries a request, a background job retries it too, and the user clicks again. One connection problem now produces several independent retry decisions. In a SimAggregator integration, map every place that can resend an action. Do not assume a repeat is safe without checking the operation's documented meaning.
Separate reads from operations that may change state or incur costs. A convenient universal transport policy may be unsuitable for both. The decision should consider whether the request was sent and whether its outcome is known.
Choose the layer that makes the final decision, and have other layers report attempt information. Link repeats to their original local task and preserve the actual transmission count. This does not promise exactly-once execution by an external service; it makes your own behavior observable.
Test with a simulated connection interruption in a test environment. You should be able to answer who chose to retry and what evidence justified that choice. If the answer is “every layer did a little,” simplify ownership before tuning waiting intervals. Include the interface button in the review, because user-driven repeats are still repeats.