Contrast the prefill and decode phases and explain why prefill is compute-bound while decode is memory-bandwidth-bound.

Prefill processes all prompt tokens in parallel and is usually matrix-multiply compute bound. Decode adds one token at a time while reading large weights and KV state, so memory bandwidth and batching dominate.

Make interfaces and ownership explicit; add versioning, access control, monitoring, canary rollout, rollback, and a plan for delayed labels or human review.