The paper uses two definitions of Academic Merit, one of them being "does the student end up being admitted" which is oddly circular. And they also use another one related to test scores.

All of the ones related to test scores DO show statistically significant improvement with race-blind admissions. See page 8 of the pdf.

Clearly they didn't get the result they wanted so they decided to use some unusual definitions of "Academic Merit" so they could get the headline and abstract they want. "Trust the science" clowns.