When the Regime Breaks_ Identity Failures in Chemistry, Biology, and AI

Abstract This document is a case-study companion to Universal Identity and Persistence: A Forcing Theorem for Identity Under Transformation. It does not extend the theorem, add new axioms, or claim that the examples prove the theorem. Its purpose is diagnostic: to show what happens when a system is treated as persistent while its actual identity-bearing level is undeclared, misdeclared, or governed by the wrong invariants. The paper examines cases from pharmaceutical chemistry, materials science, biology, AI systems, cryptography, and software engineering. In each case, the same failure pattern appears: a system is treated as the same; the identity claim is made at one level; the relevant recurrence occurs at another level; a new attractor or transformation appears; and the declared regime lacks the invariant needed to classify the change as PERSIST, BREAK, REGIME-CHANGE, or UNDEFINED. Part I shows that molecular sameness does not imply solid-state identity, using Ritonavir, tin pest, chocolate tempering, paracetamol, carbamazepine, aspirin, and pharmaceutical polymorph screening. Part II shows that biological identity often resides in maintained corridors rather than isolated molecules or sequences, using prion propagation, viral quasispecies, and microbiome dysbiosis. Part III turns to AI and engineered systems, contrasting exact cryptographic artifact identity with undeclared AI model update identity, capability elicitation, regulatory AI identity, and database schema migration. The contribution is not the discovery of new case facts. It is a cross-domain diagnostic: regime failure occurs when reality changes at a level the identity claim forgot to declare. The cases show why declared identity-bearing units, governed invariants, admissible transformations, drift bounds, and verdict criteria are operationally necessary for decision proof, audit infrastructure, and persistent autonomy systems.