Hmm... I think I can have partial lead, albeit I would not rule out some other fundamental workings from the equation.
First of all,
JonF, what was that Fe phase (edit: yeah I get it - its steel, but metallic Fe tends to form some Fe minerals where some are magnetic) and how big it was?
I think the moral the story below will be "standards should be big, but not too big, especially magnetic standards".
So this year we got to do analysis of metallic meteorite sample as polished slab (not thin-section) something like 2.5cm x 4cm with about 1.5cm thickness.
My colleague asked me to take a look, as totals of analyses was "quite too far" from the expected analytical 100% (that was on SX100 but I don't think this matter). Then I noticed that optical image is extremely shifted from what we see on electronic image. The problem was that this Fe in meteorite formed some magnetic minerals, which because of huge thickness was producing strong enough magnetic field to significantly shift the beam up to and more than 100 µm!!! We looked through the sample and there had to be different crystals with different magnetic orientation (image shift relative to optical had different directions depending from place on the thin section), we looked for least "image-shifted" (or rather magnetically shifted) spot, and used microprobes built-in beam shift capability to compensate that shift further more and then we got analytical totals close to 100%.
WDS is very sensitive to geometry (that is why I am absolutely skeptical about replacing wire-based proportional counter with SDD based counter. Unless such SDD could be made as thin 100µm wide long (>2cm) wire?). That is advantage (better spectral resolution) but also disadvantage (if e-beam position is not guaranteed).
Just think about it. It is enough to move stage in Z direction out from Rowland circle few µm to get systematic lower totals. Thus shifting beam for more than 10µm in some directions can also move the geometry from perfect Rowland's circle significantly.
This also explains, why often verification of spectrometer "is our best Friend" - beam can drift during day in some semi-circle due to natural daily magnetic field fluctuations. Verification of spectrometers is important even if no crystal flip was done and from my experience in most of low analytical total situation it can fix that.
BTW special caveat is needed for labs which has huge solar panels installed (going gr
eeeeeeeeeeeeen, huh!) on roof or is planned to be installed - Your probe is at a risk of being influenced by a huge DC field during the day, which can't be compensated anyhow (only nights are warrantied to be safe). In such unfortunate case the X-ray registration is not only susceptible to atmospheric front (low pressure GFPC sensitivity to humidity and atmospheric pressure changes), but also sensitive to clouds passing-by. Way to go
.
However, this does not explain observations of probeman, as shift in his case was very minimal. I think there are some fundamental areas still not well understood or overlooked:
1. not efficient or not existent self-absorption correction
2. diverging probabilities of beam electron collision with different shell electrons depending from valence electron state of metal atom
3. the lax approach with dead time correction when historically collecting the database of MAC's - biased matrix correction.
...something else...