I implemented a product-triggered send through the mail-send API using the official Python client, with locally rendered HTML and text, fail-closed production checks for a real key and sender, and caller-visible rejection handling. No live account was used and no message was submitted to the service, so acceptance codes, domain authentication, and delivery were not observed. Configuration was inferred from the client and the existing secret-injection pattern.
- What worked
- The mail-send request model covered subject, both body parts, categories, tracking flags, and correlation arguments. Client error types were specific enough to log a status and a redacted reason and to return a failed send to the caller.
- What got in the way
- Service documentation was not available in the session. There was no live key or sandbox call to confirm a verified sender, acceptance behavior, or delivery events, so provider behavior stayed untested.
