Karl Smith presents a good case for not making too much of the Romer-Bernstein forecast. I like the method of turning it around on another policy and thinking about how we would react.
But I think something he tip-toes around should also be stated much more clearly: the methods by which economists forecast the future and the methods by which economists estimate stimulus impact are completely different, and the latter are much more accurate. People often talk as if the two empirical tasks are one and the same, without realizing that they are entirely different. They're so different, in fact, that when I learned forecasting and when I learned the sorts of methods they use to produce multiplier estimates, they were taught in entirely different econometrics classes. So if somebody says something like "their model predicting the impact of stimulus obviously doesn't work because their forecast was way off", they don't know what they're talking about. They're conflating two completely different things.
You can be (like I am) confident in their multipliers but not all that confident in their forecasts. This shouldn't be that surprising of a position. We trust that climatologists have a better sense of what's happened with climate in the past than they do of what will happen in the future. We trust that evolutionary biologists have an easier time describing the evolutionary tree from 50 million years ago, but we don't think they're particularly good at understanding what the tree will look like 50 million years from now. Only in very mechanical systems, like planetary orbits, do we trust long-term projections. Even in cosmology more generally nobody knows exactly what's going to happen in the far future.