While discussing with Anamaria about what might be going ON, I drew up two tables to help me better elucidate the problem. I include them here, 'cause they might be useful for broader discussion.

If we take the magnet polarity from E1000617 as cannon, then I would expect the M0 face sensors to have the following signal-sign-flip chain (assuming a positive coil current repels a N, and attracts a S magnet)

Test OUT	E2O Gain	COILOUTF Gain	Current Out	Magnet Pol.	Resulting Force		Resulting Motion

         { F1	0		+1		0		N               none  }
+ Long---{ F2	+		-1		-		S               Push  }	-- 		+ Long
         { F3	+		+1		+		N               Push  }

         { F1	-		+1		-		N               Pull  }
+ Pitch--{ F2	+		-1		-		S		Push  } -- 		+ Pitch
         { F3	+		+1		+		N               Push  }

         { F3	0		+1		0		N              	none  }
+ Yaw----{ F2	+		-1		-		S              	Push  }	-- 		+ Yaw
         { F3	-		+1		-		N               Pull  }
This is how I had set up the COILOUTF gains before today (though I was off by an overall minus sign). 

However, since Jeff G.'s has re-installed COILOUTF gains as he has shown above, and the damping loops close and are stable, then that implies the following magnet polarity 

Test OUT	E2O Gain	COILOUTF Gain	Current Out	Magnet Pol.	Resulting Force		Resulting Motion

         { F1	0		+1		0		N               none  }
+ Long---{ F2	+		+1		+		N               Push  }	-- 		+ Long
         { F3	+		-1		-		S               Push  }

         { F1	-		+1		-		N               Pull  }
+ Pitch--{ F2	+		+1		+		N		Push  }-- 		+ Pitch
         { F3	+		-1		-		S               Push  }

         { F1	0		+1		0		N              	none  }
+ Yaw----{ F2	+		+1		+		N              	Push  }	-- 		+ Yaw
         { F3	-		-1		+		S               Pull  }

In other words, though the magnet polarities are properly compensated for, it doesn't match E1000617. That means if any other suspensions are assembled to the E1000169 specification, we can't necessarily use these COILOUTF gains.

Other thoughts:
- Because both of the above arrangements should result in the same motion, I'm lowering my EURIKA! flag a little on the guess that this is the source of the cross-coupling.
- It's strange that the cross coupling does not appear in both degrees of from, i.e. Yaw resonances don't show up in the L2L TF, but L resonances show up in Y2Y. I think I may have thought this before which let me down the "it's air currents!" path.

Anyways, my path forward outlined above still holds... we need a few more measurements before we start changing hardware -- or close the chamber and "aLIGO accept" the suspension.