We've had a couple of cases of anti-NMDA encephalitis through our ICU, failure of disappearance of peripheral blood auto-antibodies with treatment tends to indicate a poor prognosis - if treatment is successful blood levels should drop, however this can take some time (recent patient in ICU for >1 year).  We didn't tend to repeat CSF auto-antibodies, and relied on blood.

As far as I remember, Josep Dalmau mentioned in an extensive speech that the antibodies are more difficult to detect in the CSF, and disappear with treatment. This would imply that (1) they disappear after successful treatment, (2) if the encephalopathy has "burned out" after some time, they shouldn't be detectable as well. I think Prof Dalmau is open to any question you pose, and certainly the one who knows most of the details.

Several aspects have to be kept in mind when discussing the persistence of auto-antibodies in the CSF. In the case of paraneoplastic syndromes the source of autoantibodies are presumably plasma cells in secondary or tertiary lymphoid structures outside the CNS. Therefore, levels of antibodies in the CSF should be linked to serum levels. It has to be noted that paraneoplastic antibodies can persist for years, even when the tumor has been removed (Alexopoulos H  et al. Paraneoplastic anti-NMDAR encephalitis: long term follow-up reveals persistent serum antibodies. J Neurol. 2011 Aug;258(8):1568-70).