A hazard genome is a twelve-element, dimensionless feature vector computed from a rolling window of sensor readings, describing exclusively how those readings are moving rather than what they currently are.
Absolute magnitude is the thing low-cost sensing elements are worst at, and the thing that drifts as an element ages. A rate of change is neither: a slow baseline drift moves a reading without moving its slope. Building the feature vector from derivatives alone is what lets a classification survive both a cheap sensor and a long service life.
Twenty-four recent sample sets are reduced to twelve numbers: least-squares velocities for the gas, temperature, surface-temperature and humidity channels; a flame-persistence fraction; gas and temperature accelerations taken as the difference between recent-half and earlier-half window slopes; an acoustic rise ratio; a vibration-activity measure; and a saturating duration term. Each channel carries its own normalisation scale, so the dimensions are commensurable.
A gas leak strongly activates the gas velocity and gas acceleration features while the thermal features stay flat. A thermal runaway drives surface-temperature velocity and acceleration hard while ambient temperature lags. Same twelve dimensions, visibly different shapes.
Computed identically on the sensing controller and on the platform, using the same feature definitions in the same order with the same scales, so a local display and a remote dashboard show the same assessment without needing to synchronise.
