Before 7700: Detecting Persistent Trajectory Precursors in OpenSky Data

Squawk code 7700 declares a general emergency, but it does not show when unusual flight behaviour first became visible. We ask whether surveillance trajectories can provide useful warning before the first sustained 7700 transmission. Using OpenSky Network data, we build a retrospective pilot cohort of 42 emergency flights and 162 matched non-emergency controls; an exact-route sensitivity analysis contains 29 emergencies and 116 controls. Causal kinematic deviations are measured relative to matched controls, converted to empirical tail-surprise scores, and accumulated with a fixed one-sided cumulative-sum detector. Evaluation uses 20 repetitions of grouped five-fold cross-validation, keeping each emergency and its controls in the same fold. Thresholds are calibrated only from inner cross-fitted control-flight scores. At a 5% target, the primary cohort achieves a 5.25% mean held-out false-alarm rate and detects17.62% of emergencies, with a median repeat-level warning time of 86 seconds. The exact-route sensitivity achieves a 5.09% false-alarm rate and an 18.10% detection rate, with a median warning time of 130 seconds. A 60-second exponentially weighted moving-average baseline detects about 10% at the same operating point. Persistent pre-declaration trajectory precursors are therefore visible for a meaningful minority of emergencies, although most events remain indistinguishable from normal operations using surveillance kinematics alone.