In a research system, a document-fetch subagent sometimes gets HTTP 503 responses from a publisher's site. Which two practices match how this failure should be handled? Select two.

Answer: A and C — Retry the 503 within the subagent before involving the coordinator, since it is a transient failure / If retries don't resolve it, return the error category, what was attempted, and any partial results

Transient failures should be recovered locally where possible (a), so the coordinator isn't pulled into every hiccup. Errors the subagent can't resolve should go up with context: the category, what was tried, and partial results (c), so the coordinator can retry differently, switch sources, or continue. Escalating every 503 (b) adds needless round trips. A bare 'unavailable' status (d) discards the context the coordinator needs.