The same medical image can be annotated with abnormal regions by two people; the same legal text can be classified by two individuals. The submitted labels may be neatly organized and easily ingested during model training. But if the task requires professional judgment, the annotator’s qualifications shouldn’t be determined solely by a single line in their profile saying, “I am an expert.” @primus_labs proposes using zkTLS data proofs to verify the credentials of professional annotators in fields like healthcare and law—for example, confirming whether they hold the requisite professional credentials. This approach targets one of the earliest steps in AI training: before evaluating how well the model learns, we must first ensure that the people providing the judgments are qualified for the task. Credential verification, of course, doesn’t guarantee that every annotation is correct—ongoing sampling, review, and correction are still necessary. But if qualifications rely only on self-declaration, quality control lacks a foundational basis from the start. This is currently still described as an official use case. I hope to see more concrete implementation details in the future: how credentials are verified, how annotators’ personal information remains protected without full disclosure, and how erroneous labels can be traced back once identified. If AI is to learn judgment, those training it must also be verifiable. #PrimusLabs #AIData @KaitoAI