Section 39 of the Bharatiya Sakshya Adhiniyam 2023 carries forward, almost unchanged, the nineteenth-century architecture of opinion evidence: a person “especially skilled” in a field furnishes the court with reasoned criteria it can independently test. Section 330 of the Bharatiya Nagarik Suraksha Sanhita 2023 compounds the difficulty by allowing an expert’s report to be read into evidence, and the expert to go unexamined, unless a party affirmatively disputes it. Both provisions presuppose a human opinion-holder, someone who studied a subject, formed a view, and can defend that view under cross-examination. That presupposition fails when the witness produced in court is a police technician who ran a proprietary facial-recognition or predictive-risk tool without access to its training data, its matching logic, or its validated error rates. This article argues that such testimony is not opinion evidence within the meaning of section 39 at all, but a species of proof Indian evidence law has not yet named. It develops an analytical taxonomy of the opacity such testimony presents, and reads section 39 and section 330 against three comparative bodies of law: the American Daubert–Frye reliability tradition and its confrontation-clause jurisprudence on machine-generated forensic reports; the English common-law presumption that computers operate correctly and its catastrophic failure in the Post Office Horizon scandal; and the European Union’s data-protection and artificial-intelligence law, which increasingly regulates algorithmic opacity before a tool is ever deployed rather than after it produces evidence. The comparative material shows both that mature legal systems have not settled on a common answer, and that each has learned something India can still learn pre-emptively. The article closes with a doctrinal response anchored in the certification model Indian law already applies to electronic records.