Had a similar situation to xetroV a few days ago....
IRS L suddenly developed a huge error - approximately 12nm.
The warning was the CDU message in the scratchpad.
Radio position was good, ANP below RNP and IRS R sensible.
I expected some map shift...but none to my knowledge.
I wondered whether if the IRS output is discontinuous, the FMS dumps that input into the system position computation, just like it does if the radio position from a suspect navaid falls outside the Circular Error of Probability (CEP).
Anyone?