Motor chuffed, and apparently part of nozzle cracked away, seemingly causing asymmetric thrust as the model, a well known very stable design, did the skywriter routine for part of the burn, then straightened out and flew away at about a 40 degree angle, deploying its ‘chute safely.
The first chuff was enough to get the model off the rod, which is not characteristic of these motors. They often chuff in a very small way once or twice before taking off.
Also, delay was short (fortunately in this case) at about 4.3 seconds, per accelerometer data. Flight data are here: https://flightsketch.com/flights/3910/