Its probably poor phrasing in the article. The key is its event based (more or less) and they don't reveal until after all analysis is done. So analysis compl

Its probably poor phrasing in the article. The key is its event based (more or less) and they don't reveal until after all analysis is done. So analysis complete we have X events at a SNR of 3 dB, Y events at a SNR of 2 dB, etc. Then and only then do the blind test dudes reveal "ah well all X events at a SNR of 3 dB were fake, and one of the Y events was fake, and you missed detecting one of the Y level events but the other 42 results were real." We did this a lot in quantitative chemical analysis lab, it kind of spoils the point of the lab session if the TAs tell you the percentage iron in the sample before you start to test it, or while you're evaluating the test run results.