So, looking at your relative IR intensities (which is basically what the REF photodiode sees), I see some instability at 25h and 70h, which correspond to your jumps.
As explained in this other post, the displayed OD is actually a ratio of [90deg signal] / [REF signal]. So if the REF changes significantly independently, it will effect the reported OD.
What I suspect happened was that the REF was slightly moved during the sampling, and this caused the shift. (In our next version of the Pioreactor, we fix this. For now, you can try to fix the photodiode in place with glue or sticky tac).