Single-isotherm non-identifiability of modified Dubinin–Astakhov parameters and the statement of system-level hydrogen-storage requirements in observable quantities

System-level targets for onboard hydrogen storage are routinely translated into required sorbent properties through the modified Dubinin–Astakhov (DA) isotherm. We quantify how much of that translation a single published isotherm can support. Two independent digitization routes, reconciled under a pre-registered rule with per-point uncertainties, recover eleven AX-21 excess isotherms (30–298K) from the literature; profile likelihoods under an explicit errors-in-pressure measurement model show that one 77K isotherm of ordinary published quality leaves the D–A parameter vector practically non-identifiable: the 95% region spans n max 37–142 mol kg −1 under the frozen parameter bounds, and wider still when they are relaxed, with log 10 ( P 0 / P a ) running from 7.79 to the upper bound of its allowed range at 12. Seven supercritical isotherms fitted jointly pin n max to 61.20–62.63 mol kg −1 , a withinmodel precision whose centre moves by ±13% under model-form and fitset choice. Propagated through a component-resolved system model, the loose single-isotherm region changes predicted usable system capacity by only ±2.4% along the profiled ridge, and by ±3.9% including the exact interval endpoints: the model is sloppy, with loose parameters and tight predictions, so a requirement stated as a limiting uptake at fixed energetic parameters is conditional on a ridge the data do not resolve. The engineering model itself was calibrated on one published HSECoE system and evaluated once against externally pre-registered held-out targets; three of four comparisons passed and the 200-bar gravimetric comparison failed (+41% against a ±20% band), so, per the pre-registered fork rule, no absolute system capacity is claimed and requirement maps are reported in observable axes (deliverable excess uptake between stated full and empty states, and packing density) under two structurally different engineering models. Shapley-effect attribution over the joint posterior and declared engineering uncertainties shows that the variance of predicted capacity is dominated by engineering inputs rather than by material-parameter uncertainty. All data, code, analyses and this manuscript’s numbers regenerate from a single open repository.