Implement the data-fetching logic for an autocomplete input whose UI is already provided. The component requests suggestions for the current query and must support debouncing, retries, latest-result wins, and caching. ### Part 1: Debounce Requests Wait for a short quiet period before sending a request. Empty input should clear suggestions and must not trigger a fetch. #### What This Part Should Cover - Timer cleanup on every query change and unmount - A clear distinction between debouncing input and delaying rendering - Testability with an injected clock or fake timers ### Part 2: Latest Result Wins Responses may arrive out of order. Ensure an older request cannot replace suggestions for a newer query. #### What This Part Should Cover - Cancellation when supported - A request generation or query identity check - Safe loading and error state transitions ### Part 3: Retry Transient Failures Retry only failures classified as transient, with a bounded attempt count and backoff. A new query cancels the old retry sequence. #### What This Part Should Cover - Error classification - Bounded exponential backoff with jitter - Cancellation during sleep and fetch ### Part 4: Cache Suggestions Reuse recent successful results by normalized query. Define cache size, freshness, and how cached data interacts with a background refresh. #### What This Part Should Cover - A normalized key - TTL and bounded eviction - Avoiding shared mutable results - Stale-while-revalidate as an explicit option ### What a Strong Answer Covers A strong answer treats every timer, request, and retry as cancelable lifecycle state, prevents stale writes after unmount, and keeps cache policy separate from UI rendering. ### Follow-up Questions - Preserve accessibility announcements while results update. - Deduplicate identical in-flight queries across components. - Add client-side ranking without blocking typing.