End-to-end (e2e) testing verifies complete user workflows by simulating real interactions through the browser. In Artemis, an open-source platform for interactive learning developed at the Technical University of Munich, e2e tests complement unit and integration tests by covering regressions that span multiple system layers. Maintaining and extending this test infrastructure presents significant challenges: the existing suite is complex to maintain, slow to execute, and difficult to configure, which contributes to limited coverage in several parts of the system.
This thesis investigates these challenges through an in-depth study of the e2e testing process in Atlas, the Artemis submodule responsible for competency-based learning, chosen as a representative case within the broader system. It reveals concrete issues that slow down and undermine the reliability of testing, explores practical ways to stabilize, accelerate, and simplify the process, and examines what barriers prevent contributors from writing and maintaining tests. The contributions include 23 e2e test cases covering the principal Atlas workflows, a two-phase CI pipeline that reduces first-phase developer feedback time from a median of 27.7 minutes to 4.0 minutes, fixes to multiple pipeline bugs that had silently corrupted test results, and a scripted local setup that reduces environment configuration to a single command per operating system. These improvements extend beyond Atlas to the Artemis e2e infrastructure as a whole.