Validating Residual Correction in Fresh-Food Companion-Animal Diets: A Paired-Arm Reliability Ablation, a Stochastic Reliability Framework, and a Tissue-Kinetic Safety Model

Fresh, home-cooked diets for cats and dogs are biologically appealing but, evaluated against an established nutrient standard, are routinely incomplete. A residual-correction approach — a biologically appropriate fresh base plus a calculated micronutrient correction — can close that gap, but the correction must be shown to hold across the real-world variability of home ingredients, over months of feeding rather than on paper for one idealised bowl, and demonstrably because of the correction rather than because of the base. We describe three complementary methods and report a controlled ablation of the correction itself. (1) A paired-arm reliability ablation. The identical simulation — same diet, same variability model, same pseudo-random seed, same draws — is executed twice, once with the correction applied and once with its contribution reduced to a nutritionally null value. This isolates the correction as the single manipulated variable. Across 100,000 simulated feeding days per arm, the corrected diet clears every nutrient floor on every draw (zero observed failures; Wilson-95 lower bound 99.996%, a value limited by sample size rather than by the diet), while the same balanced fresh base without correction clears every floor on no draw at all (Wilson-95 = 0%). The uncorrected failure is neither marginal nor stochastic: the base misses 11 to 16 nutrient floors simultaneously, several at a literal zero percent of requirement, so the joint criterion cannot be met in any draw at any sample size. (2) A stochastic reliability framework. The Monte-Carlo procedure underlying both arms, which resamples realistic ingredient, bioavailability, potency and fill variability and reports the fraction of simulated daily bowls clearing every floor without approaching a ceiling, with a Wilson confidence bound. (3) A tissue-kinetic safety model. A one-compartment toxicokinetic account of chronic micronutrient accumulation, with homeostatic absorption regulation and first-order clearance parameterised from published literature, showing that a corrected diet remains below safe upper limits over long-run feeding. Together these model results are consistent with adequacy in a fresh diet being a distributional and time-integrated property that ingredient selection alone does not deliver, and they isolate the correction as the manipulated variable within the simulation. Because both arms draw on the same composition inputs, the ablation demonstrates internal dependence on the correction term; it is not a causal effect measured in animals. Formulation proportions and engine internals are excluded by design.