Validated but unconstrained: why total-clearance data cannot constrain minor elimination pathways, and how a thirtyfold parameter revision went undetected in ocular pharmacokinetics

Mechanistic models in drug development are validated against quantities that can be measured and then used to predict quantities that cannot. This work shows that the licence this confers is computable, often close to empty, and gives a documented case in which a thirtyfold parameter error survived validation for years and propagated into models used for clinical translation. When the validated observable is a total clearance rate k = k_a + k_p and the prediction of interest is the share f = k_p/k carried by one parallel pathway, the normalised sensitivities with respect to any parameter entering only k_p satisfy S_obs/S_pred = f/(1-f), exactly. Validation informativeness degrades in proportion to how minor the pathway is, and minor pathways are what such models exist to quantify. The identity holds for every parameter entering only the minor route, including geometric terms; the single exception is the parameter the observable actually measures. There is a matching detection threshold. With proportional assay error the Fisher information for the minor rate constant reduces to N times the sum of squared sampling times divided by the squared CV, so the smallest detectable value is k_p,min = 2.49 CV / sqrt(N sum t^2), depending only on sampling times, assay precision and group size. For a standard rabbit intravitreal design this is 0.0077 per day. The case study is intravitreal pharmacokinetics, where the prediction is how much drug reaches the retina and the observable is the vitreal half-life. A published first-passage model (Lamirande et al., IOVS 2024;65:21) is independently reimplemented in open-source code, reproducing its retinal areas to within 0.2-1.0% and its mean first passage times to within 1-4%. The two barrier permeabilities on which its posterior boundary rests are shown not to be determined by the data from which they came: bootstrap resampling gives P(p_RPE > p_ILM) = 0.59 for Fab and 0.43 for IgG, and the source paper's own confidence interval on that ratio spans one. A 2026 re-estimation using new retinal measurements has since revised the retinal pigment epithelium permeability roughly thirtyfold downward, to 5.9e-9 cm/s, inverting the assumed ordering. Published estimates of the barrier ratio now span forty-fold. At the revised value the posterior route lies a factor of five below the detection threshold above: it could never have been distinguished from zero by vitreal sampling at any assay precision. Meanwhile the superseded value propagated into whole-body models for rabbit, monkey and human, whose reported validation covers plasma, aqueous and vitreous humor - the compartments least able to constrain it. The work also computes what would have settled the question in advance. With all parameters free, the relative standard error on the effective posterior permeability from vitreous-only sampling runs into the hundreds of percent and is not improved by more animals, more time points or longer follow-up. Adding retinal tissue concentration at ordinary homogenate precision improves it by about two orders of magnitude. Retinal tissue alone carries essentially all of that information; the compartments ocular studies are built around, and the only ones accessible in humans, carry almost none. A second structural check evaluates the model's global feasible set. The cynomolgus monkey half-life is reachable only by forcing a posterior elimination fraction of 56-66%, against the 10-30% the same model reports. Modelling the shortfall as a uniform volumetric clearance yields a single flow of about 0.21 uL/min accounting for both a 50 kDa Fab and a 150 kDa IgG: a falsifiable signature of convective bulk flow. No new experiments are reported. All source data are published. All analysis code is archived at doi:10.5281/zenodo.21887689 and a single command reproduces every number and figure. Version 3 restructures the paper around the 2026 parameter revision and its propagation, adds the detection threshold and a measurement-design analysis, replaces an earlier debris argument with a hindered-diffusion treatment of the inner limiting membrane, corrects the Peclet analysis of the convective inference, traces every permeability value to its primary source, and removes a comparison that was circular. Earlier versions presented the boundary reduction as the principal finding; this one does not. Detailed derivations, the falsification enumeration and the feasible-set analysis are now in Supporting Information. Preprint. Not peer reviewed.