It's not just garbage-in, garbage out; it's also sample size.
The IHME works great in New York, where you have a huge sample size, high density and lots of hospitalizations. But the smaller the populace, the more that statistical noise gets baked in. You have less density, less person-to-person contact, smaller social circles and thus a smaller R0.
Since North Dakota isn't tracking recoveries, I'll use South Dakota as an example. South Dakota has 66 counties. A majority of them—I counted 44, but I might have missed or double-counted while scrolling—either no longer have any active cases or never have had any to begin with. But that Sioux Falls meat-packing plant outbreak skews the results for the whole state. Those 44 counties have no reason to be locked down.
I will also note that the IHME also projects no shortage in hospital or ICU beds for either North Dakota or South Dakota, so in those cases there is arguably no need for lockdown there.