Strange motion offsets in one direction on one axis

Oh, it is a little baby machine by many measures- 1 kW spindle, 12x6x10" XYZ travel or so, etc. Weighs, I dunno, maybe 500 lbs? I built it with a welded steel frame filled with epoxy granite, with proper linear bearings and ballscrews (to be fair, I got nice THK linear bearings, cheaped out on the ballscrews), so hobby bench mill sized, but with way more rigidity. The BT30 spindle is probably overkill, but it was available, and there's a lot of tooling options. TTS didn't seem like a great option, and ISO20 seemed too small.

Aside from this issue popping up, it's been a great experience coming from the gantry-style machine I'd been tweaking for years. The rigidity, Z height, and toolholders are the huge upgrades, but the MPG, touch probe, covered ways, coolant control, spindle/axis stall-detection, and servos are very nice as well. Having a spindle with enough torque at reasonable RPMs to actually handle (small) drills is still blowing my mind.

Sorry for the crappy picture- literally just walked over to snap it, so you can see it's a bit disassembled for debugging (and with random scrap taped over the t-slots so I didn't have to clean them out). That's a Kurt 4" DX4 vise for scale.

Good to hear the wiring, solder joints and crimps are all solid. Nasty checks... but had to be done. 

Trying to reproduce would be greatly helpful. Cutting material or air, up to you. If air cutting just ensure you record a reference before for when you pause the machine where it crapped out last time, and meas offsets.

One important asset you forgot to post is your gcode. That also needs to be ruled out. Not doubting your skills here, but I'd be happy to spare a few minutes to validate it inside an industrial ISV/sim running a proven Lcnc postprocessor for you. That'll reveal if or not.

Nice machine-tool. You'd obviously know from half a mile away if indeed hardware related already (you built-it, it's in the blood line).

Hi dbtayl,
I came across this thread and wanted to offer a suggestion that may help with further troubleshooting.
When troubleshooting things like mis-positioning, I recommend verifying the position in each of the following places:               
               - In the original command code (this would be your G-Code passed into the controller)
                -The controllers account of how many steps/quad cnts have been sent out (I'm not 100% certain of where you would find this in LinuxCNC)
                -The motors actual position
                -The motors commanded position (add/subtract the motor's tracking error to the measured position)
                -The actual position of the mechanics (This is where your cut comes into play)

This process will let you immediately narrow down whether the problem is hardware, electrical, G-Code, or a controller issue.

For verifying the motor's position, Teknic's ClearPath SD and MC motors come with a motor setup and configuration software called MSP. The MSP software includes many diagnostics tools that can be helpful when troubleshooting a machine’s/system’s performance. For example, MSP includes a digital oscilloscope which can be used to monitor different application variables in real time.With MSP you can observe the motor's measured position, tracking error, torque, and velocity command in real-time. 
Here is a short video on how to use MSP's scope tool:

I hope this is helpful. Good luck with your continued troubleshooting!
-Brendan F. Teknic Servo Systems Engineer

Assuming the mechanics are good just because I built it myself is possibly not warranted That said, aside from the ballscrew I already ordered a replacement for (will probably take weeks to get here, though), I do believe the mechanics are in good operating shape.

From an electrical perspective, I'm getting an order together for ferrites and some PSU filters to cut down on noise- a lot of the noise seems to be coming from (or through) the power supply. Catching it at the source is better than trying to band-aid further down the line.

Those are the two things I know are suboptimal- fixing them may or may not be related to this problem, but definitely aren't going to hurt.

I'll hopefully get a chance to re-run the code tonight/cut air on the existing workpiece, and will report back when I get to that. I'm thinking I'll put masking tape over the -Y pocket surfaces. If the machine behaves, those shouldn't get touched. If cutting air looks OK, I can repeat in a fresh block of material. I also intend to remember to grab a screenshot of where LinuxCNC thinks it was cutting- forgot to do that last time.

I'll dig up a Windows PC to monitor the servo while I do the above and just jog the axis around. Thanks @Teknic_Servo for reminding me that's an option.

Gcode attached. I cut a couple operations from the end that I never ran, so what you see is everything that was run on this pocket. I've looked at it 5 ways (verifying coordinates in the code, viewed in LinuxCNC, FreeCAD gcode viewer, FreeCAD cut simulator, and some online viewer) and not found any issues, but would love for that to be the problem.

So... I go to reattach the Y servo so I can run the tests mentioned above, and... the bloody nut retaining the screw is loose?!?!? Apparently I somehow missed that despite specifically checking it earlier AND pulling on the axis and not seeing motion on an indicator. Not seeing a beating-head-against-wall emoji, so you'll have to imagine one.

This might be an easy fix... well, the machine at least. The operator/debugger might be a lost cause.

Still going to try to mitigate the electrical noise and replace the ballscrew either way, still going to run tests.

(also definitely owe the forum a writeup on my way cover design as a thank you for the help- I've been pleased with them. Way better than the accordion/leather/rubber suggestions you typically see, much more affordable than the sliding covers you see on industrial machines, all while being nice and low-profile for machines without travel to spare)

ETA: Writeup here, may be clear as mud: forum.linuxcnc.org/show-your-stuff/54503-metal-way-covers