Hazard replay is an autonomously triggered forensic recording in which a sensing node escalates its reporting cadence on recognising a hazard, preserving per-second frames that carry the system's own contemporaneous reasoning alongside the raw channel readings.
Detail is expensive to store continuously and invaluable for the minutes around an event. Letting the system decide when to spend that budget - on its own recognition, rather than on a threshold crossing - starts the high-resolution record while the episode is still developing.
Recognition escalates the node from a ten-second baseline to one vector per second for a bounded period. Each frame carries the channels, the genome, the matched signature and similarity, the risk level, the enclosure heading and any camera frame. The platform governs and clears the escalation, not the field hardware.
The frame sequence renders as a narrative of the episode - how it developed, what the system concluded at each stage, when it rotated and what it recommended - forming a portable report for an insurer or an inspector.
Colloquially the black box. The bound sits server-side so a node cannot be left stuck in forensic cadence by a failed field-side timer.
