Closing the Last Mile: Field Evaluation of CAP-Driven Multi-Hazard Alerts

Last-mile disaster warnings in India increasingly depend on standardized public alerting, yet device-centric pilots often fail to convert forecasting advances into timely, actionable protection for outdoor and low-connectivity populations. This paper addresses the documented limitations and deployment gaps associated with the Bihar NITISH pendant by repositioning it within an SDG-aligned, data-driven early-warning and action ecosystem. A multi-layer framework is presented that (i) ingests real-time meteorological and hazard feeds, (ii) mitigates false alarms and missed alerts through an interpretable hybrid rules + ML risk tiering layer, (iii) operationalizes multi-channel delivery via a unified alert schema, and (iv) closes the last-mile usability gap through complementary endpoints, an app-based interface and an administrative dashboard for situational awareness, escalation, and compliance tracking. To strengthen evidence and accountability, a reproducible evaluation protocol is defined using CAP-anchored timestamps and device/app logs, reporting Probability of Detection (POD), False Alarm Ratio (FAR), missed-alert rate, lead-time distributions, and latency decomposition (platform, last-mile, and human acknowledgement) with percentile statistics. Inclusivity and governance safeguards multilingual prompts, audit trails, and equity checks, ensure that warnings remain actionable for diverse user groups and operational contexts. The proposed approach converts a single-endpoint pilot into a measurable, scalable, and policy - ready early-warning system capable of reliable operation under variable connectivity and real user routines.