The portal publishes seven experiments for the Northwest Atlantic. Two are readable, and the other five are refused here rather than being attempted — each for a reason that has nothing to do with the caller and would otherwise surface as an obscure failure a long way from its cause.
Why the seasonal forecast is not among them
Its files carry a 64-bit integer coordinate, which DAP2 has no type for, so
the PSL server answers NcDDS Variable data type = long with an HTTP 500 to
every classic OPeNDAP request for one. The DAP4 endpoint does describe them —
and then reads them wrongly: ncdf4 transposes the start and count vectors
against the dimension order it reported, and the read segfaults R rather
than returning anything. A reader that crashes the session is worse than no
reader, so this refuses instead.
The files are downloadable whole from the THREDDS fileServer endpoint and
open correctly once local, at roughly 260 MB per variable per
initialisation. That is the route if you need them.
Why the projections are not among them
long_term_projection and multi_decadal_outlook have directories on the
server and no files in them. The portal announces an experiment before its
output is posted, so this may simply be early; cefi_archive() will read
them the moment they appear.