6DOF Rotational Constraint Softness

[Twingy]

I have modeled a pendulum affixed to a large box using a 6DOF constraint with rotation about the X axis. Both objects have the same density. The center of mass for each object was calculated and is physically correct. A Torque is applied to the pendulum with a regular 200ms pulse every 1000ms. With a relatively low force the pendulum swings back and forth as one might expect. However, as the force is increased the axis of rotation for the pendulum shifts. I read a considerable number of posts regarding soft constraints posted by other users as well as the 51 page Erin Catto presentation on CFM and ERP. It seems as though no matter what I set the m_maxLimitForce, m_stopERP, m_normalCFM, m_damping, m_limitSoftness, m_stopCFM, or m_bounce to for all 3 axis there is no change.... My simulation tim step is 10ms (100 Hz simulation) with 10 substeps for each step. The white lines mark the constraint axis of rotation, as you can see they should colinear with one another, but separate under a large enough force. I do not mind adjusting parameters such that the computational expense is very high... Accuracy is more important than computational efficiency right now.

pendulum_constraint->getRotationalLimitMotor(i)->... = ...

I am attaching some screen shots of what is happening over time.

Is there a better explanation of how all parameters "stiffen" or "soften" a rotational constraint? For example, what is m_maxLimitForce and what is the difference between m_stopERP and m_stopCFM versus m_normalCFM. I understand CFM and ERP can be translated into dampening coefficient "c" and spring constant "k", but I do not have an intuitive feel for those yet.