On the Flat Dilution Curves of Cooked Dishes: Evidence for a Non-Luminous Flavour Component and a Structural Protocol for Its Recovery

Perceived flavour intensity should fall as a dish is diluted. Across 41 cooked dishes carried through a nine-point geometric dilution series spanning a 43-fold range in concentration, it does not: the fitted Stevens exponent is $\beta =0.06$ (95% CI $[-0.01,0.13]$) against a literature value of $0.85$. Five single-ingredient controls run in the same series fall off correctly ($\beta =0.81$), so neither the panel nor the design accounts for the result. We therefore compare two methodologically independent measurements of the flavour held in a pot — luminous flavour $L$, obtained by descriptive panel, and dynamical flavour $M_{\mathrm{dyn}}$, obtained from the dispersion of headspace release rates under a virial closure and never tasted — and find them discrepant by a median factor ${{\rm Y}}_{\star }=M_{\mathrm{dyn}}/L=118$, with a spread of 40 to 913 that we do not attempt to narrow. We read the excess as a non-luminous flavour component: present in the dish, dynamically detectable, and unperceived. Modelling the pot as a weighted digraph of competitive channel occlusion, we show that the perceived flavour vector is the leading eigenvector of the occlusion operator, that the measured spectral gap (${\lambda }_1/{\lambda }_2=6.8$) collapses that vector onto one or two compounds within minutes of the first simmer, and that the dark component is exactly the orthogonal complement — the spectral shadow. The mechanism is structural rather than compositional, and it is reversible. Under a staged-participation schedule that alters only the order in which ingredients are granted thermal contact, with composition, mass, energy input and elapsed time held fixed, $L$ rises 9.4-fold and ${{\rm Y}}_{\star }$ falls to 12.6. Holding that schedule fixed and replacing all twelve ingredients with twelve sharing no item moves $L$ by $0.19$ perceptual units, below the panel’s $0.5$-unit discrimination threshold ($F(1,29)=0.14$, $p=0.71$). Schedule moves perceived flavour 269 times further than composition does. We conclude that $99.15\%$ of the flavour so far prepared has not been tasted, and that the ingredient list is, on this evidence, an inventory rather than a cause.