# Diagnose Cold-Load and Massive-Table Performance in a Web App You are reviewing two frontend performance problems. Explain how you would measure each one before changing code, then propose a prioritized design. ### Clarifying Questions to Ask - Which user devices, networks, browsers, and geographic regions matter? - Which user-visible metric is failing: first content, largest content, interaction latency, or memory stability? - Does the table need all records locally, or only search, aggregation, and a visible window? - Which filtering and sorting operations must be exact and interactive? ### Part 1: Cold Page Load A user opens a large commerce-style site in a private browsing session with no warm application cache. How would you improve loading speed without hiding important content? #### What This Part Should Cover - Measurement of the network and rendering critical path - Server, CDN, compression, caching, image, font, JavaScript, and CSS choices - Prioritization of above-the-fold content and hydration work - Real-user monitoring rather than lab-only conclusions ### Part 2: Rendering Ten Million API Records A page receives ten million records from an API and attempts to render them in a table. Explain the failure modes and redesign the data and rendering paths. #### What This Part Should Cover - Network transfer, parsing, heap use, reconciliation, layout, and paint costs - Server-side pagination, filtering, sorting, or aggregation - Row virtualization for the visible subset - Cancellation, incremental delivery, worker usage, and user feedback ### What a Strong Answer Covers - Evidence-driven prioritization tied to user-visible metrics - Work avoidance before micro-optimization - Clear ownership between backend, browser, and rendering layers - Accessibility and correctness under partial or changing data ### Follow-up Questions - What can still be cached in a private session? - Why is pagination alone not enough if the page renders thousands of rows? - When does a Web Worker help, and what can it not fix? - How would you verify that an optimization helped real users rather than one laptop?