Dosezy: A Local-First, Accessibility-Centric Android System for Medication Scheduling and Reliable Reminder Delivery

Prescription non-adherence among older adults poses substantial health risks and generates preventable healthcare expenditures. Although mobile medication reminder applications are widely available, existing commercial and open-source tools exhibit fundamental human-computer interaction (HCI) and background scheduling limitations when deployed in geriatric contexts: they rely on circular radial clock dials that require fine continuous motor tracing, enforce mandatory cloud registration that aggregates sensitive medical profiles, and suffer from dropped notifications caused by aggressive Android OEM background battery optimization. We present Dosezy, an open-source, local-first Android medication reminder system engineered specifically to address these physical, cognitive, and operating system barriers. Dosezy introduces a stationary 12-Hour Grid Time Picker featuring a 12-first layout and oversized >= 56 dp touch targets, enforces architectural network isolation via manifest privilege minimization through the total omission of the android.permission.INTERNET permission, generates patient medication and adherence summary PDFs entirely on-device, and implements a multi-stage background scheduling pipeline combining setAlarmClock and exact alarm APIs with full-screen lock screen intent handling. We report empirical evaluations across systems and human factors dimensions: Multi-device runtime profiling on flagship and entry-level hardware demonstrating cold startups under 780 ms, vector PDF generation in 142 +- 18 ms, and database query latencies under 12 ms, establishing engineering feasibility against Android 60 fps (16.6 ms) frame budgets. A multi-OEM Doze mode alarm benchmark across five physical devices (99 of 100 scheduled alerts met the prespecified <= 5.0 s criterion under tested deep Doze conditions, 99.0%, 95% Wilson CI [94.55%, 99.82%]). An initial counterbalanced within-subjects laboratory usability study (N=16 older adults, aged 61–76, mean age 68.5 +- 4.8 years) demonstrating a 44.4% reduction in total time-setting latency (19.91 s vs. 35.79 s, p < 0.001), an 80.0% reduction in touch input errors (0.44 vs. 2.19, p < 0.001), and a System Usability Scale (SUS) score of 83.44 +- 9.48 (Grade A, 'Excellent') compared to 57.19 +- 12.45 for standard radial dials. This deposition package includes the complete camera-ready manuscript (PDF and DOCX), LaTeX source, BibTeX database, participant-level usability dataset (N=16), physical multi-OEM Doze mode alarm benchmark dataset (100 physical alarms), and the automated Python reproduction suite.