Case studies. Not theory. This companion to “Governing AI Without Killing Value” shows what liability looks like when verification is absent by construction (V = 0) and evidence is demanded. It applies the KEV framework and the AI Dependency Cascade to forensic reconstructions across regulated sectors. Each case isolates the decision points that matter: what authority was delegated, what assumptions were imported, what verification was performed and what could not be reconstructed later. Every case ends at the governance boundary – the moment a regulator, auditor, court, investor, ombudsman or customer compels reconstruction and the organisation discovers it cannot provide it. The intent is evidential scaffolding. Boards, executives, audit teams and regulators can use these reconstructions to pressure-test operating models, assurance design and board-level attestations. It is designed to support hostile diligence against current deployments and to specify procurement and assurance requirements that do not confuse prompt discipline with verification. If you can reconstruct the decision under force, you have governance. If you cannot, you have theatre. Case studies included: A bank that loses the ability to explain its own risk under prudential scrutiny A government department that cannot account for its own actions under judicial review An insurer whose reserving trail collapses under audit and supervisory examination A frontier AI lab forced into a >$600m impairment after customers cannot deploy under scrutiny A VC diligence failure where AI validates AI, producing synthetic consensus and regulatory exposure A software scale-up that loses reconstructability of its own codebase, then fails under incident pressure This companion and the wider publication series do not replace independent legal, regulatory or technical advice. They are built to make those discussions harder to wave away and easier to evidence.