This is a brand new project Cognizant has with a universally-known client. Since there is no objective way of measuring your performance, a "quality analyst" in the company simply looks at a small portion of your work and makes the same judgment that you did. If they disagree (theoretically, because you acted incorrectly based on the policy), you get it "wrong."
The problem arises because a small percentage of your work will involve judgment in answering questions that are ambiguous by policy, because no fixed policy can address all of infinite ways words (let alone typographical errors, sounds, emojis, etc.) can be arranged by a user. Depending on which individual analyst checks your work, your work may be considered right or "wrong." So, you'll have a 50% (for example) chance of getting that "wrong" regardless of which decision you make.
Since your measured "performance" an illusion of objectivity behind it, you will eventually be terminated because your numbers fall short of a specific threshold. Even if you execute every decision as flawlessly as possible, it's only possible to meet this threshold as long as you stay lucky enough to not be assigned "judgment cases."
