Published Verification Reports
The IVR series takes claims that are widely repeated in the AI market and checks each one against the primary source — the statute, the regulator, the official audit record. Every verdict is sourced and stated plainly: VERIFIED, PARTIALLY VERIFIED, or NOT VERIFIED.
Two disciplines govern every report. We verify claims, not claimants — no company is named, in praise or in criticism. And precision beats posture: the goal is the exact position, nuance included, because an overcorrection is just a new error. Corrections to any finding are welcome and are published.
Verdict scale · VERIFIED / PARTIALLY VERIFIED / NOT VERIFIED
Three widely-repeated claims about automated and algorithmic coverage denials in Medicare Advantage, checked against the primary HHS-OIG audit record. Where a figure reflects a specific audit period, the report says so.
Read IVR-003 →Three circulating claims — the federal state-AI moratorium, Texas TRAIGA, and California's SB 1120 — checked against the record. One false, one true, one garbled.
Read IVR-002 →The first report in the series: the EU AI Act high-risk timeline, Colorado's repealed SB 205, and NAIC model bulletin adoption — three "facts" that had all moved within roughly six months.
Read IVR-001 →What every report holds to
- Primary sources only. The statute, the regulator, the official record — not press coverage of them.
- Claims, not claimants. No company is named. Statements are graded, not people or products.
- Corrections published. Any finding shown to be wrong is corrected publicly. That is the point.
MarginSignal OS — independent verification of what deployed AI, and the claims around it, actually do.
These public reports are issued at no charge and are not legal advice. For the paid engagement built on the same discipline, see the sample First-Verdict report.