Every system transition in nature and human history leaves remnants of the old system embedded inside the new one. Mitochondria are ancient bacteria still running their original code inside eukaryotic cells. Cancer cells are unicellular organisms running pre-multicellular code inside a multicellular body. Feudal aristocracies became shareholders inside capitalism. New systems fail not because their principles were wrong but because they were forced to compete on infrastructure built entirely for the systems they were trying to replace. This paper proposes the Embedded Substrate Theory: the remnants of previous systems do not disappear at transitions — they embed themselves inside the new substrate and continue running their original code. The conflict between embedded old code and the new substrate above it is the universal mechanism behind disease, aging, economic failure, and civilisational collapse. A prediction methodology is proposed: map the embedded layers of any system, identify the conflict points between each layer and the one above it, and those conflict points constitute the system's failure atlas. A novel epidemiological hypothesis is also proposed: multigenerational co-habitation in East Asian family structures may constitute an embedded substrate transfer vector — elderly cellular dysregulation transmitting to co-habiting infants through microbiome and epigenetic proximity. This paper makes no normative claims about which system is preferable. It examines the structural mechanics of substrate persistence as a descriptive, not prescriptive, framework applicable equally to any system regardless of ideological content. Version 4 adds: observer-dependent cancer progression framing; book analogy for multi-layer reversion; death as complete reversion to monocellular state; Serial Atavism Model extension to simultaneous multi-layer conflict; KAIST December 2024 cancer reversion validation; intervention target redirection to interface suppression mechanism; early EV as embedded substrate infrastructure case study; Wright/Moore's Law efficiency capture analysis; clinical disclaimer in MASC section