#2189 jsbsim ground spin at startup

[Huntley Palmer]

I.ve spent a full week restarting and watching both the dhc2W and J3Cub, jsbsim models, I'm asking for hints where next to look.

For both models at startup, after increasing throttle and releasing parking brake the plane starts rotating ( increasing fast ) about the vertical axis in a ground loop.
Viewing from outside and behind the model the tailwheel should be watched. ( The scaling of the J3Cub tailwheel is wrong at present, best to watch the DHC2 ). The groundloop rotation can be slowed and stopped by using the toe-brakes as expected. However, if the toe-brake is held active sufficiently for the tailwheel to change its caster direction from one side to the other, the direction of yaw starts in the opposite sense, following the tailwheel deflection no matter how small. If the toe-brake is used to halt the plane but not enough for tailwheel angle to cross the fore-aft line then yawing will resume in the previous direction. This action suggests a positive feedback as if the model's yawing is following the tailwheel, not dictating the caster angle.
Since the plane can be ground looped in either direction at will by using the toe-brake to set caster angle it looks as if there's an effect in play that is not related to propwash torque, p-factor or any other aero moment.
I've been running my sixteen channel Pid-o-Scope watching all the moment forces in /fdm/jsbsim/moments but it's difficult to see leader from follower.
At flightgear/src/FDM/JSBSim/models/FGPropulsion.cpp #824 I see:
PropertyManager->Tie("moments/l-prop-lbsft", this, eX, (PMF)&FGPropulsion::GetMoments); PropertyManager->Tie("moments/m-prop-lbsft", this, eY, (PMF)&FGPropulsion::GetMoments); PropertyManager->Tie("moments/n-prop-lbsft", this, eZ, (PMF)&FGPropulsion::GetMoments);
Where other calls to GetMoments call eL, eM, eN but changing those calls had no effect.
One, or both of these models has auto-coordination set active, this obviates any possibility of getting control with the rudder because the rudder will continually be reset by the aileron, paricularly noticeable with joystick twist rudder and low-value noise from the aileron axis. The rudder gets reset under keyboard control with auto-coordination, too. That suggests a case for locking out auto-coordination when WOW is true.
If there's any other place I can look for evidence this is a non-aero bug I'll try and find extra time to do so.

Tks

[Huntley Palmer]

A link to the startup with J3Cub:
[(http://www.bitwisetech.com/fgfs/vokoscreen-2020-02-05_08-52-43.mkv)]

The scope trace shows, starting at the top, all jsbsim moments from properties: fdm/jsbsim/moments with l, m, n components and total moments. The final trace is tailwheel steering angle.
At startup the parking brake is on and moving aileron, elevator and then rudder illustrate that gear reaction opposes elevator and rudder moments as would be expected. Once the brake is released the n-moment from the gear seems to follow tailwheel angle and quickly swamps any aero effect from the rudder.
Is this due to an error in the configuration file or is jsbsim failing to register the fact that a castering tailwheel should not be generating forces so big ?

Tks