Marlin 2.x: Fixing scaling/calibration issues

I'm rebuilding a Folgertech mini Kossel delta, and I've got it pretty much working, but my calibration cube is the wrong size.

I've got Marlin 2.0.7.2 on it. I've used G33 to fine-tune the delta parameters and I don't have any cupping (so the head stays the same distance above the bed as I traverse in X and Y), I've run UBL to generate a mesh, and I was able to print a calibration cube and it looks fairly decent, but... the cube is about 10% too big in the X and Y direction and about 10% too short in the Z. This doesn't seem to be a problem with the stepper calibration (because strict Z movements, like G0 Z50 and then G0 Z100, move the carriages exactly the right amount). There's also no apparent cupping/bulging of the surfaces of the cube.

It looks like some firmwares have a command (M579) for scaling the XYZ coords prior to moving to a point, but Marlin doesn't seem to support that. I'm presuming that the thinking is that, if your dimensions are off, then something else in your config is messed up (like delta-rod length, delta radius, etc), but I'm loathe to mess with those, since those affect cupping.

Does anyone know what parameters I should be looking at to get the dimensional scaling correct?

Are you sure the problem isn't in your slicer? I once accidentally scaled a print when slicing.
The only things I can think of that would affect Z printing but not Z movement wouldn't make X and Y larger.

Pretty sure it's not that. The plot thickens. I actually have two of these Kossels, and I loaded the same firmware on both, but the rod lengths are a little different between the two (but I've configured them both for their own rod lengths). I tried the same calibration cube on the 2nd one, and it printed almost perfectly.

Is the rod length the only thing you changed between the two printers? Did you use the same gcode file on both printers?

(head slap!) Oh, you said you ran G33! The auto calibration would try to compensate for any errors which would make the bed look flat at the expense of dimensional accuracy if some of the parameters are off.
Based on what you have said so far, I say the rod length is off. I suggest you remeasure the rod length on both printers and compare them with the values in the firmware on each machine. That will make sure you're measuring them correctly.
I'd also make sure you have the right parameters by issuing the appropriate M665 command.

As I recall, the values deduced by G33 were very close to what I started with (like within a mm), but I'll double-check.

I guess the thing that confuses me now is it sounds like you're suggesting the G33 will give you M665 numbers which don't match the physical dimensions of your printer in the interest of getting the effector to maintain a constant height off of the bed, but my understanding was that the effector climbs/dives when moved off-center whenever its internal values do not match the real measurements.

Secondly, if you suggesting that, if I want good dimensional accuracy, then I shouldn't be using G33, then how else am I supposed to be compensating for climb/dive when off-center? UBL leveling? Wouldn't that cause the tops of my prints to be cupped (as the firmware tapered off the bed-mesh compensation after X layers)?

I'm suggesting that G33 will give numbers that don't match the physical dimensions IF it is using incorrect parameters that it doesn't adjust -- such as rod length.

I'm not suggesting you shouldn't use G33, just that it can't fix everything if it is working with incorrect parameters and that might result in what you are seeing. That said, is true that I suspect you may not need g33; the problem is with (most likely) rod length.

As to "how else", you've answered your own question. "whenever its internal values do not match the real measurements". You make sure the values match. To me the value of G33 is in setting the tower angles if they are off. That is hard to measure physically. The height and endstops are a mere convenience. Delta radius should already be known.

BTW it would help if you answer thehankinator's question. I had the same question but assumed the answers were "yes".