Did GST 2.0 Reduce Classification Disputes, or Relocate Them?

India's “GST 2.0” reform of 22 September 2025 filtered four-rate tax schedule (5/12/18/28 per cent, plus a compensation cess) into two principal slabs (5 and 18 per cent) and a 40 per cent rate on demerit goods. The government's stated rationale was that fewer rates would reduce classification disputes. We treat that as a falsifiable empirical claim and test it the only way a dispute claim can honestly be tested — against the actual disputes, proxied by published Advance Ruling (AAR) and Appellate Advance Ruling (AAAR) decisions. Our central methodological contribution corrects the binding limitation of an earlier draft of this study: the GST Council's official advance-ruling portal is, in fact, a near-complete and machine-tagged census — roughly 2,940 published AAR rulings, each labelled by its statutory question-type under Section 97(2) of the CGST Act — so the issue-coded census the literature assumed unavailable can be built directly and freely. On the substantive question, the verified record as of late May 2026 supports a disciplined two-part finding. First, the headline dispute-reduction claim cannot yet be confirmed at adequate statistical power on a completed, coded post- reform window, because that window is still maturing under a publication lag; we give the exact sample size at which the test becomes decisive. Second, where the evidence already leans, it leans toward relocation rather than elimination: the portal shows post-reform rulings still litigating the surviving 5%/18% boundary under the new rate notification; the reform demonstrably created or worsened a fresh class of inverted-duty refund disputes; and officialdom contemporaneously stood up a cross-state conflict-resolution forum (the GST Appellate Tribunal acting as interim National Appellate Authority). We supply a verbatim coding rule, a documented portal-sourced ruling subset, the full statistical machinery, and a replication protocol so the decisive test can be re-run reproducibly as 2026–2027 publications populate the post-window.