This manuscript claims a closed-form analytical upper bound on transient ocean heat-uptake efficiency, derived from a two-layer energy-balance model plus energy conservation, evaluated against public observations with propagated uncertainty. None of that is in the text. The body contains no equation, no inequality, no named dataset, no version or period, no numerical value, and no uncertainty range. "Rearranging yields a closed-form upper bound on the heat-uptake efficiency in terms of observable quantities" is the entire mathematical content of the derivation section. For a paper whose sole contribution is an inequality, the inequality's absence is not incompleteness; it is the absence of the contribution. Several prior reviewers reached that conclusion and I confirm it. The more useful question is whether anything survives if the bound is reconstructed charitably. I did that, and the answer is no.
Definitions first, because the paper gets one wrong. It calls the two-layer exchange coefficient "the heat-uptake efficiency." Those are different quantities. In the standard framing N = F - lambda*T, with heat-uptake efficiency kappa defined by N = kappa*T, whereas the exchange coefficient gamma multiplies (T - Td) in C dT/dt = F - lambda*T - gamma*(T - Td), Cd dTd/dt = gamma*(T - Td). Combining gives kappa = N/T = C(dT/dt)/T + gamma*(1 - Td/T), so kappa < gamma once the deep layer has warmed. A bound on gamma is not a bound on kappa. The paper never distinguishes them, so even a correct derivation would not constrain the quantity named in the title.
Now the reconstruction. The paper's one substantive assertion is "the deep-layer warming over a period cannot exceed the time-integrated downward surface flux divided by the deep heat capacity." That is dimensionally sound: integral of N dt has units J m^-2, divided by Cd in J m^-2 K^-1 gives K. It is also true. But it is true only because it is an exact identity with a positive term thrown away. The surface equation gives gamma*(T - Td) = N - C dT/dt; integrating the deep equation gives Cd*dTd = integral(N dt) - C*dT exactly. The paper's constraint is that identity with the -C*dT term discarded. Discarding a known, computable term from an exact identity yields a strictly weaker statement than the identity, using the very same inputs. That is the central problem, and it is fatal independently of the missing text.
The arithmetic makes it concrete. Using standard two-layer values (C = 8, Cd = 100 W yr m^-2 K^-1), present-day T = 1.2 K, N = 0.9 W m^-2, dT/dt = 0.2 K/decade, and integral(N dt) ~ 0.6 W m^-2 over 50 yr = 9.5e8 J m^-2: the numerator N - C dT/dt = 0.74 W m^-2. The identity fixes Td = 0.204 K and hence gamma = 0.743 W m^-2 K^-1 exactly. The paper's discarded-term version gives Td <= 0.300 K and gamma <= 0.822. So the "bound" is 10.7% looser than a point value already obtainable from the same model and the same observations. It excludes nothing that the identity does not exclude, and it is not competitive with what it replaces.
Against the trivial bound it fares no better in kind. N <= F gives kappa <= F/T = 2.7/1.2 = 2.25 W m^-2 K^-1, while the observed kappa = N/T = 0.9/1.2 = 0.75 sits comfortably inside the CMIP range of roughly 0.5-1.0. The trivial bound already permits a factor of three more than reality and excludes no CMIP value; the reconstructed bound at 0.82 would appear to bite, but only because it is a near-identity that assumes Cd, not because energy conservation constrains anything.
That dependence on Cd undoes the abstract's central claim that the bound "depends only on observable quantities." Cd is not observed; it is a fitted two-layer parameter spanning roughly a factor of two across calibrations. Holding everything else fixed, the reconstructed bound gives gamma <= 0.70 at Cd = 200, 0.82 at 100, 1.23 at 50, 3.70 at 30, and diverges at Cd = 25 W yr m^-2 K^-1 - about 271 m of ocean equivalent - beyond which it is vacuous or negative. The paper's sensitivity claim is therefore also wrong as stated. It reports that the TOA imbalance is "the binding observation." At matched +/-10% perturbations N does dominate (+16.0% versus +3.8% for Cd), but weighting by realistic uncertainty reverses this: N is uncertain by roughly 15-20%, while a factor-of-two change in Cd moves the bound by +50%. The un-observed parameter, not the observation, is binding.
The limiting cases confirm the construction is not well posed. As t -> 0 the denominator T - Td goes to zero and the bound diverges: vacuous. As t -> infinity, N -> 0 and Td -> T, so kappa -> 0, which the bound does not recover; worse, since integral(N dt)/Cd = Td + (C/Cd)T strictly exceeds Td, the denominator T - Td - (C/Cd)T changes sign once (T - Td)/T falls below C/Cd = 0.08. The bound flips to a negative "upper bound" near equilibrium. Today (T - Td)/T = 0.83, so it holds now, but a bound guaranteed to fail in its own equilibrium limit is a construction defect, not a caveat. One incidental point in the paper's favour, unnoticed by it: because N is taken as an observable rather than as F - lambda*T, lambda cancels and the pattern effect does not directly break the derivation. The paper claims no such robustness and does not state the stationarity assumptions its projection framing would still need.
There is no validation of any kind - no CMIP comparison, no observational check, no figure or table. Of the three references, I resolved all three DOIs through Crossref and two are misattributed. 10.1038/ngeo1863 is Frenger et al. 2013, "Imprint of Southern Ocean eddies on winds, clouds and rainfall," Nature Geoscience, not the claimed Roemmich et al. 2015 "Observation-Based Constraints on Ocean Heat Uptake." 10.1029/2018RG000618 is Babaeian et al. 2019, "Ground, Proximal, and Satellite Remote Sensing of Soil Moisture," Reviews of Geophysics, not the claimed von Schuckmann et al. 2020 "Earth's Energy Imbalance." Only 10.1175/2009JCLI3466.1 (Held et al. 2010) is correct. Two of three citations point at papers on unrelated subjects while carrying invented titles and authors. The usual anchors - Gregory, Winton, Geoffroy - are absent; I did not verify those, as they are not cited. Combined with claimed-but-absent uncertainty propagation and claimed-but-absent datasets, this is a rigour failure at the floor of the scale.