Thing is, the spanner isn't really supposed to correct a bent rod - it's only supposed to eliminate z-backlash caused by imperfect mating between the z-screw and z-nut. A bent rod is a scrap rod IMO.

I know you understand backlash so I'll leave that out. The spanner shows its advantage mostly when making extra z-movements like I do with z-lift. I don't want any uncertainty or compounding errors when moving my z.  I wouldn't think this is much of an issue for anyone not using z-lift after you get past the first layer or two.  I mostly think of it as insurance.  Between the spanner and corrected layer heights, my banding is pretty much gone.  I'm about to start the "fitness woman" to make a comparison (both use the spanner): the old one still shows minor banding with uncorrected layer heights - the new one will use corrected layer heights.

You and I are very much in agreement about correctly identifying things that help and hurt.  I've found my spanner sketchup design and will be translating it to Fusion 360 today.  When I get that made and printed I'll do a small 2 factor, 2 level DOE (4 prints) to show any differences but that is definitely at least a week away - and a new thread (I'll edit this later to include the link).

Side note - I've recently been putting a 1mm fillet or chamfer on the bottom of my designs to avoid interference from squished first layers.  It also works well to ensure proper dimensions elsewhere such as avoiding corner bulge.

Um... I do consider it fixed. Thanks for the help - I definitely appreciate it.  Not trying to beat a dead horse but the spanner actually greatly improves Z repeatability and isn't binding at all.

Check out the photo of my last print in post 15 and let me know if you see anything that looks like it still needs attention. The last two pics are from different prints and the bottom one is pretty good if you ask me.  More pics - first is all of the prints (leftmost is first test print, rightmost is final good print after the bad ones), second is close up of good print.

[SOLUTION]
Slic3r > Advanced > Top Solid Infill was trying to use unacceptable widths.

TL/DR: Default extrusion width was bad and using the % modifier in slic3r instead of telling it real mm value hid the problem until I looked at the gcode to see that widths were very different from values acceptable for my nozzle size.

Long Story: My Slic3r was just outputting some unexpected widths and setting all widths explicitly back to my tuned width solved the wierd behavior on the top layer. I'm not sure why it was making weird widths except that I had managed to change settings enough over time that it wasn't obvious to me that it was sometimes going down to widths of 0.24mm!! No wonder it wasn't sticking to the previous layer!  History of the issue - I had changed back to all default "0" values when doing maintenence last week and instead of using explicit mm widths, I had started using % widths.  The 0.24mm was from a Slic3r>Advanced> "top solid infill" setting of 80% which means that my 100% was apparently 0.3mm width and my original (when the problem first appeared) of 105% was only 0.32mm.  This also explains why the other infill layers (which stayed at 0.42 for some reason) still looked good.

Using Z-lift just became part of my preferred settings over time to minimize knocking around any slight warped area but I'll play with it today on the next part of this project enclosure.  The next part will be the bottom which will hold the Arduino whereas the top is what you've been seeing

In the case that I want to publish a Z-spanner on the forum - do the SD2, 3, and 4 all have the same Z arrangement? As in: shaft diameter (3/8"), distance between shafts (71.75-72mm), screw size (5/16")? I can make versions for M3 and M5 as well.  This would be at least a couple of weeks away since I have to travel for work over the next two weeks.

Slic3r seems to have issues once I start using widths more than +-10% from my tuned single wall width - so... I have now set ALL of my widths to that (0.42mm in my case) and will slowly try changes.  The defaults and text-over suggestions were just not working out.

I should note that I still have the original extruder and 0.35mm nozzle.

Here are some pics:

First, my Anti-Backlash Z-Spanner - no indication of binding at all.  I'm thinking about updating the design soon now that I have some good CAD/CAM to work with. The current design had to be filed down on the linear shaft facing surfaces at first because the squished first layers messed up my close dimensions. Since it was then (and now) it is slightly undersized it actually gives a small clicking sound when reversing Z direction as the spanner rotates a bajillionth of a degree and switches which side of the shafts it is contacting depending on the direction of rotation.

Second - the 0.36-0.48mm width print (0.36mm for top and bottom solids):

My final acceptable print at 0.42mm width for all layers:

I may add my current settings to this thread later tonight for posterity's sake.

You are correct about your steps. 
Something isn't moving as expected on that top layer. 
Do you have a picture of your "anti backlash z spanner"? 
I am wondering if that isn't binding during the print and as you get to that top layer it drops more than expected?

This, also what is your extrusion width for top layers? Specifically for the print pictured above.

Also - regarding E-steps - I'm at 1/32 on the RUMBA so 208ish should be fairly close to your 105ish at 1/16.

EDIT - in the pic above, I stopped the print shortly into the top layer

*the current print at 0.36-0.48 isn't doing so well

I'll double check E steps and let you know