Converting callbacks to async/await is a semantics change — not just a syntax rewrite. Cancellation, lifetimes, and concurrency boundaries can change runtime behavior and allow network work to continue past a view’s lifetime.

Tie UI requests to lifecycle-scoped Task (cancel on dismiss) or use TaskGroup for bounded fan-out; avoid spawning anonymous Task from view controllers when you expect strict cancellation.

Propagate Task cancellation into the underlying URLSessionTask using withTaskCancellationHandler or explicit cancel calls, and consider both cooperative timeouts and URLSessionConfiguration timeouts to bound request duration.

Use TaskGroup or a bounded worker pool pattern to limit in-flight requests and collect partial results as Result values so successes are preserved while failures are reported.

Make shared caches actor-owned to reduce data races and publish UI state via observable patterns so views can react to changes safely.

Choose detached work (e.g., retries that must outlive UI dismissal) and lifecycle-bound Task when you require strict cancellation on teardown — pick the model that matches lifetime requirements.

Validate changes with async XCTest cancellation tests and add instrumentation around URLSession calls and withCheckedThrowingContinuation boundaries to observe cancellation and resource usage.

Question for the thread: how have you validated cancellation and resource limits when migrating networking paths to structured concurrency?