Random relative error for identical compute run

I have just spent... don't ask.... a long time doing dozens of runs of the attached 2 plate capacitor driven in EC by a CIR current source single pulse.

* CONVERGES fine. Identical every time.

BUT
I get a Relative error in, say 10 repeats of the exact same mesh, exact same everything.
Erel = 7.9, 11.0, 8.6, 7, 3.8, 3.2, 1.9, 5.3, 7.1 ...

I have tried multiple runs where I "Delete solutions" each time.
I have tried multiple runs where I "Delete history" each time.
I have tried multiple runs where I "reload a whole new save, solution-emptied .MPH" each time.
I have tried multiple runs where I "reload a whole new save, solution-emptied .MPH + restarting COMSOL each time" each time.

The data it produces is the same. Except once, where a DELETE SOLUTION radically altered the solution from one solution to another that was then consistent.

Increasingly fine meshes do nothing. It just takes longer. Everything else stays the same, except the solution (max voltage in a potential slice) changes slightly.

Q1. Why? What do I have to do to get a consistent behaviour out of it?
Q2. I am told that these values for Erel are awful and I should not trust the data. Yet all I get is a 'warning' and no clue what to do about it. All the graphs look OK.

I am in need of some wisdom.
Please.... anyone out there got any pointers?
Offending .MPH is attached. It's small and takes 3 mins ish to run with mesh 2.

regards

Colin