High-stakes environments like semiconductor cleanrooms traditionally rely on synchronous "master-apprentice" training, which is difficult to scale and resource-intensive. We present a Mixed Reality (MR)-assisted learning system equipped with a web-based authoring tool that allows instructors to create spatialized training modules and practical learning sessions. Through a long-term study (N = 15) spanning two semesters, we implemented MR usage as an assisting tool during the training. The findings reveal that this asynchronous scaffolding not only lowers the threshold to content creation but also provides an alternative way of learning and training in restricted environments. Video analysis also revealed a shift in social interaction: while training inside the cleanroom was previously dominated by instructors, MR promotes individual learning rather than imitating the instructors, although this shift to autonomy resulted in a state of situated isolation. Ultimately, this work validates a scalable, instructor-authored MR system that successfully transitions novices to self-regulated autonomy in high-stakes environments. CCS Concepts • Human-centered computing → Collaborative and social computing systems and tools; Mixed / augmented reality; • Applied computing → Interactive learning environments.