What does the first layer look like on something like a 100mm x100mm x 2mm?

If the power supply can be switched from 110V to 220V, then it will behave this way, especially when the bed heater is on. If for some reason the power supply fails, it will behave this way, as well.

Any bad contacts in the bed power circuit? Are the bed power supply wires connected with ferrules?

More details needed.

This is not a simple undertaking right now. I suggest that you start with the mechanical kit version of this:

https://openbuildspartstore.com/openbuilds-workbee-1010-40-x-40/ along with NEMA 23 stepper motors consistent with the Duet2/TMC 2660 current supply.

Then incorporate a Duet2 as the controller. This will be the upper end of low performance, but it will get a job done. As noted above you could wait for the Duet3 and have enough current supply for bigger NEMA 23s, but you'll be limited in voltage necessary to deal with the motor inductance.

Personally, I went this route and selected a Mesa Electronics 5i25 FPGA, Geckodrive G540, and LinuxCNC. I use hardware stepping from the 5i25. You'll need a computer with a 25 pin parallel port for the 5i25. I run the motors with a 48V unregulated power supply. This setup is kind of the lower end of professional performance, IMO.

It may have nothing to do with the loss of connection, but which stepper motors did you get with the WorkBee?

If you bought the 2.8A (Duet2 WiFi upper limit) versions did you change the 2.4A limit to 2.8A.

I went with the Mesa Electronics 5i25, the Geckodrive G540, and LinuxCNC for my WorkBee.

A little more detail: Relative to the nozzle (upgraded to E3D V6), the BLTouch is mounted toward the X tower and toward Z tower. Also, the difference between the ridge and valley on the X tower end is about 0.4 mm. That's not much in absolute size, but more than a usual layer thickness.

So when Z is incremented on the X tower end of the zigzag, the effector on the Z tower side is pulled up. That pull-up fades, so to speak, as the effector moves toward the Y tower.

There are many variables involved. If I follow Occam's razor, then there appears to be a bind between the X and Z tower and/or arms. It looks like the X and Z towers are not parallel or are closer together than X to Y and Y to Z. Furthermore, it could very well be a temperature sensitive issue and inconsistent temperature expansion and contraction of the Rostock melamine plates and/or melamine expansion and contraction with changing humidity. To say nothing of the Rostock frame rigidity.

The brute force solution is either the Duet smart effector or the 713maker.com FSR approach.

I recently completed upgrade of a Rostock Max V2 with the Duet 2 WiFi and incorporated a BLTouch for a Z probe. When running the bed mapping routine, I get strong evidence of effector tilt. I am using the standard Rostock arms with ball cups.
I do not have a screen shot of the resulting bed map, so I will describe it.

The routine begins closer to the Y tower, then decrements X toward the X tower. After completing this line, then Z is incremented toward the Z tower followed by incrementing X toward the Y tower. It zigzags in this fashion until completion. The resulting map shows ridges and valleys. A valley results when X is decremented toward the X tower, and a ridge when X is incremented toward the Y tower. These ridges and valleys decrease near the Y tower to nearly a flat surface. The Z delta between the ridge and valley is largest near the X tower.

I conclude that:

• the effector is tilting

• the effector is rotating about the Z arms.

• tilt appears to be on the order of 0.2°.

By using my cellphone with clinometer app, the bed and towers are within 0.1° of square. The towers are parallel both by clinometer and visual comparison of one tower to another.

Possible causes of tilting include, Z arms are forcing the rotation of the effector along with the cheapskate not being parallel to the bed; there is slop (backlash, loose pulley, belt tension difference, or?) between the X and Y towers.

The Z tower has some influence over the tilting, since when Z is incremented near the X tower, the Z height is higher. The valleys and ridges are smooth. One explanation is that nearer the X tower the Z increment places the effector in a bind, which is relieved by a tilt in the XY tower.