I might go with a strategy to adjust colder analogs warmer by grouping all potential matches together, and then creating a second group that leaves out the older/colder cohort.
I did this a few pages back for MEI lower than ONI by 0.5, and the forcing and precipitation patterns still turned out to be similar, except for temperature and the polar domain.
That way we can tease out what other contributing variables drive the ao/nao and temps. Don’t know if this will work in a seasonal forecast, but I’ll test this strategy this year.