A Performance Engineering Framework for Large-Scale Customer-Facing Web Applications: Optimizing Core Web Vitals in Micro-Frontend Architectures

Abstract

Large customer-facing web applications get slow through accumulation rather than through a single bad decision, and micro-frontend architecture makes that accumulation structural rather than incidental. The architecture's defining virtue is that independent teams ship without coordinating, and its unavoidable consequence is that page performance stops having an owner: each team measures its own fragment, every fragment looks acceptable, and nobody measures the composite the customer actually loads. Duplicate dependencies arrive because no team's dependency graph contains any other's; independent data fetches serialise because sequencing emerges from mount order rather than from a decision; components initialise off-screen because their owning team cannot see what is above the fold in the assembled page. Every decision is locally defensible and the sum is not. This paper proposes a framework whose contribution is governance rather than technique. A page-level interaction budget is fixed from the business requirement and then divided among fragment owners as an allocation they draw against, so that one team needing more must obtain it from another and the trade becomes negotiated rather than invisible. Enforcement runs against the assembled page in continuous integration rather than against fragments in isolation. On a production billing page in a large consumer telecom application, applying these techniques reduced full interactive load from approximately twenty-one seconds to approximately eight, a reduction of roughly twelve to thirteen seconds. That figure is measured and shipped. The paper is explicit about which of its other figures are projections rather than results.