The reports aren’t based on a detailed measurement of real-world output. Instead, they’re based on measuring the input (how much money was spent), and then using models to project how big the multiplier effect has been. Measuring spending and modeling output means that you can believe the CBO when it says that the stimulus turned out to be more costly than expected, but you should remain wary about any claims made using the “real-world effects” side.
Indeed, CBO director Doug Elmendorf has explicitly made this point, agreeing at a speech earlier this year that that “if the stimulus bill did not do what it was originally forecast to do, then that would not have been detected by the subsequent analysis.” So if in reality no jobs had been created, or only 10 jobs had been created, then the CBO’s reports would not reflect those numbers. It’s using the models that projected the stimulus would create lots of jobs to report that the stimulus did create lots of jobs. Color me unconvinced.