Strange spike in actuator outputs during PX4 SITL flight

Can anyone explain why this loiter operation has a sudden glitch in actuator outputs causing the drone to drop at around 1 minute 30 seconds into the flight?

See flight log: https://review.px4.io/plot_app?log=d5fbb802-3d51-42b1-999c-34e5a1a72a58

This is latest PX4 master bits running in Ubuntu with make px4_sitl_default none_iris, connected to AirSim bits from fix broken links and add documentation for PX4 WSL2 support. by lovettchris · Pull Request #3546 · microsoft/AirSim · GitHub.

Also I’d love to know why during TCP based lockstep the PX4 is reporting those warnings in logged messages…(messages are not supposed to go missing during lockstep - that’s the whole point of lockstep?)

This is a picture of my log file containing all incoming and outgoing mavlink messages zoomed in on the the spike, the bottom yellow line is hil_actuatorcontrols[0]. The white line is just a marker and I think it helps you see that the first mover is the actuator controls dipping, it then over reacts with a spike, then it ramps down and back up in a V-shape. Perhaps some sort of out of control over reaction. But I have no idea why it dipped in the first place. The top 3 lines in the chart are HIL_GPS and HIL_SENSOR inputs which are smooth and flat within their generated noise limits. I don’t know what altitude_local is from PX4, perhaps an estimation of some sort but it seems to react a lot faster than the HIL_GPS messages going back from AirSim.