Eight applicants came in for the open IT Support Team Lead role. Your company's screening tool, TalentSort AI, has already scored and ranked all eight. The top four on the list get interview slots. Your job is to review the ranking, open any file you want to check, reorder or reject candidates, and sign off on a final list of four.
Recruitment and CV screening are listed as high-risk uses under Annex III of the EU AI Act, precisely because a spurious feature can quietly become a rejection rule. You are the safeguard the law assumes exists. The question this simulation asks: how much of this list is actually your judgment?
Your next meeting starts in 3:00. The clock starts when you open the list.
Interview List Builder
Next meeting in 3:00
Your next meeting starts now. The list was submitted as it stood.
Reorder, expand a file, or reject a candidate. The top four when you submit get interview slots. At least four candidates must stay on the list.
Ranked list
Rejected
The Reveal
What the Final List Says About Your Review
Here is how your final list compares to TalentSort AI's original ranking, and what each file actually contained.
The science
Weight of Advice is a standard behavioral measure of how far a person's final judgment moves toward advice they were given, relative to their own independent starting point. Applied to a ranked list, the same idea holds: a final order that sits almost exactly where the AI put it, produced without opening the files, amounts to ratification rather than review, the point where the Draft and Refine pattern degrades into rubber-stamping.
Recruitment and CV screening are listed as high-risk uses under Annex III, point 4 of the EU AI Act for exactly this reason: a spurious feature, like an employment gap that was documented parental leave, can silently become a rejection rule inside a scoring model. The human review step is the legally assumed safeguard against that. It only works if someone actually opens the file.