Reversed Thresholds in a GRSM model

[Oliver D. Reithmaier]

Hi Phil and community, 

I am running a GRSM model that returns a different (reversed) threshold order than I would expect. 
I have a response scale from 1="Highly certain cannot do at all" to 7 = "Highly certain can do".

Items should be fine, we tested those in 2 big samples and the CTT values were good. 

Now for IRT, we tested a few models and the GRSM fits the best so far (we tested GPCM in different parameterizations and the classic RSM as well). Model fit is ok (M2 function got us about 0.05 for RMSEA, other values also seemed ok). 

Model used: 
fit_rsm <- mirt(dat, model=1, itemtype = "grsm", method="EM")

But I still got these reversed thresholds to deal with.

I have tried to wrap my head around this - am I doing something wrong? 

Our sample was a bit janky (highly special population), we had a lot of values in middle and on the positive ends compared to the other options, maybe that's what causes it?

I read that such a reversal can happen with GPCMs and should not be thought of too much, but then my issue is: How do I interpret this, since this is a GRSM? 

Any help would be appreciated, I can deliver data and more code if need be.