I can't speak to which data they used for this particular evaluation, but interviewing.io also hosts many companies' first phone screens -- it's not just that "

I can't speak to which data they used for this particular evaluation, but interviewing.io also hosts many companies' first phone screens -- it's not just that "some interviewers may funnel you in to their real hiring pipeline", for some companies interviewing.io is the start of their real hiring pipeline. From what I understand, they do a lot of calibration to make sure the "advance to next round" flag correlates with their clients' version of that flag. All that said, the interviewers they work with are all also interviewers are FANGs and similar -- the likelihood of working with someone you interview at a company with 10000s of employees is...about as low as for a mock interview. You've definitely identified a possible confound, but I think the i.io team is aware of it and does a lot of mitigation (for many reasons, not just for this one report).