A Second Life Under Constraints: An Experimental Study of Repurposing an Android Smartphone as Computing Infrastructure

Prior work on smartphone reuse often replaces Android or uses boards built around mobile systems-on-chip. This study follows one rooted OnePlus 9 kept on stock Android 14 and its vendor kernel while hosting personal infrastructure. A twelve-week record, including a pre-study on a second phone, contains 28 completed experiments, 6 refused or interrupted runs, 8 incidents and 10 offline analyses. The observed limits concerned kernel ABI compatibility, process isolation, Android network policy, supervision overhead, power delivery and GPU precision. Disabling F16 matrix paths brought embeddings within fixed tolerances, but a cold 36-text GPU job remained slower than the CPU (3.10 s versus 2.04 s; one run per configuration). Offline tests reproduce a thread-starvation mechanism behind an apparent GPU advantage, while the on-device cause remains unknown. This single-device case study does not establish population-level performance, energy savings or long-term availability. Shailesh Rawat, Independent Researcher, self-funded this work. No external organisation funded or endorsed it. The author built and operates the evaluated system. AI assistance and limitations are disclosed in the manuscript. Version 1.2 contains editorial changes and adds no device experiments.