The latest NHC discussion on Chantal says: "The NHC track forecast lies close to the previous one, and the latest HCCA and Google Deep Mind solutions." So they are using it as part of their track forecast guidance, presumably.

@ResidentMetaphysician NCEP (part of NWS) seems to have started running an experimental system utilizing this earlier this year

https://graphcastgfs.readthedocs.io/en/latest/introduction.html

https://epic.noaa.gov/wofscast-a-graphcast-based-emulator-for-the-warn-on-forecast-system/

There is a bit of a precedent with the NHC not using slightly better consensus aids for TCs for example. But this seems a lot harder to ignore. I for one am very excited to play around with it, even if the NWS won't admit that it is doing the same.

There are many ML models already and NWS may well develop their own or use the model ECMWF is developing. So GraphCast may never end up there. Or is this about ML models in general?

@scellus This is specifically about GraphCast. I assume ML is being explored by the NWS, but I specifically picked GraphCast because of the recent high profile Science paper and because seems like it would be easy to monitor for a definitive yes/no resolution.

Would it being mentioned in a forecast discussion from a single forecast office be an example that would lead to a YES resolution?

@kalassak Yes, that'd be sufficient to lead to a YES resolution, but it'd have to be specifically from a NWS forecast office, and to my knowledge the NWS is pretty unified in their policies/procedures and use of models. I think it unlikely a single forecast office would use this without the US government deciding to formally "allow" it to supplement pre-existing models.