{"id":"51b36a49-547b-484e-9e5c-95163a2186fa","arxiv_id":"2608.06274","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A scientific neural surrogate's posterior is compressed into a few coherent function-deformation modes, conveying uncertainty to any derived quantity via smooth draws.","lead":"The paper builds a low-dimensional uncertainty model for neural network fits: it linearizes the fitting step, compresses the result into a few coherent deformation modes, and draws smooth functions whose derivatives and integrals are meaningful. This lets scientists propagate uncertainty through calculations like forces and densities in under a second instead of running expensive MCMC over thousands of network weights.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Derivative deliverables are under-calibrated by the tangent transport (κ width ratio 1.356 warm, 2.274 cold in §5.6), and the flagship lacks an equivalent derivative width-ratio check, so the claim that Eq. (15) yields derivative-inheriting posteriors remains conditional.","rationale":"The reader's weakest assumption identifies the same tangent-linearization concern, and the paper's own §5.6 numbers are the strongest evidence that it lands. I find no reason to move the verdict: the fixed-branch function-space widths agree at the 10–15% level and the paper is explicit about conditioning and about the Q diagnostic, so a rejection would ignore substantial independent support. I also credit the honest reporting of the [0.47,8.50] near-gauge failure and the availability of warm/cold refit ensembles. The missing piece is a derivative-width ratio at flagship scale; the 24-refit ensemble already exists, so the requested check is cheap. Until it is reported, CONDITIONAL is the correct verdict.","tokens_in":19471,"tokens_out":9894,"duration_ms":104057,"concrete_test":"Reuse the existing 24-refit flagship ensemble of Section 6.4 to compute the ratio of empirical refit width to posterior-sister width for ρ(0), ν(0), and the K_z band, alongside the same ratio for the rotation-curve κ in §5.6. If the 95% bootstrap interval for either flagship ratio excludes unity, or its lower end exceeds 1.1, the derivative-inheritance claim understates flagship uncertainty; if the intervals include 1.0 within Monte Carlo error, the rotation-curve κ deficit is specific to that toy problem and the central claim survives. Report also the fraction of the refit ensemble's centered variance that lies outside the four-mode sister subspace, the flagship analogue of the 13.9% figure in §5.6.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing weak point is the use of the tangent transport of Eq. (3), through the truncated generative sister of Eq. (15), for derived quantities that require derivatives. In the paper's own controlled calibration, the derivative deliverable κ has a transport width of 0.0107, warm-refit width 0.0145 (ratio 1.356, 95% interval [1.021,1.665]), and cold-refit width 0.0243 (ratio 2.274). The authors explicitly rule out rank truncation, metric reweighting, and replacement of H_GN by the exact stationary Hessian, leaving second-order tangent response and branch response as the remaining explanations. Because the central claim explicitly includes derivative-dependent quantities, a 36% understatement on the same optimization branch is a quantitative failure of the 'inherits the posterior' property. The flagship's headline deliverables, ρ(0) and ν(0), sit at the deepest-derivative rung, but Section 6.4 reports only that the three constructions' distributions 'coincide' and gives no width ratio; the four-mode 99% variance is an internal property of the sister ensemble, not a calibration against refits. The construction may still stand as a declared conditional first-order posterior, but the strongest claim is not yet demonstrated at the scale where it is advertised.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper constructs a low-dimensional, function-space posterior for a fitted neural surrogate trained with second-order optimization. The central object is the measurement-to-function transport of Eq. (5), T_X = J_f H^+ J_m^T, obtained by linearizing the converged stationary condition (Eq. (3)) about the fit. Its SVD gives deformation modes, and the truncated \"generative sister\" of Eq. (15) generates coherent function draws with independent standard-normal coefficients. Draws are pushed through nonlinear downstream quantities (mass, epicyclic frequency, density, force), so interval estimates of derived quantities inherit the function-space posterior. Validation on a mock rotation curve compares the sister to an independent spline+MCMC posterior and to 48 warm and 12 cold perturbed-data refits; a flagship P=800 phase-space distribution fit builds a vertical-force and density tower, with four modes reported to capture 99% of the force posterior variance.","tokens_in":19753,"tokens_out":9394,"duration_ms":88660,"significance":"If the construction were fully calibrated, it would be a useful contribution: it gives a practical route to coherent derivative- and integral-aware uncertainty propagation for high-dimensional neural surrogates without MCMC, it separates repeated-experiment from posterior covariance, and it packages the uncertainty as a low-rank generative model. The paper's controlled validation is a genuine strength: the rotation-curve mass posterior agrees with an independent spline+MCMC reference to a few percent, the comparison to full nonlinear refits is an unusual and welcome calibration check, and the four-mode stability under force-basis refinement is a real empirical finding. The public code and explicit enumeration of conditioning choices further support reproducibility. The main weakness is that the derivative-dependent part of the claim, which is central to the abstract and to the flagship deliverables, is under-calibrated in the controlled test and not quantitatively checked in the flagship.","major_comments":[{"comment":"The controlled calibration shows the tangent transport under-estimates the derivative deliverable width: for κ, the transport width is 0.0107, the warm-refit width is 0.0145 (ratio 1.356, 95% interval [1.021,1.665]), and the cold-refit width is 0.0243 (ratio 2.274, interval [1.036,3.010]). Both 95% intervals exclude unity. Since the central claim—abstract and §1—is that \"any derived quantity, including those requiring derivatives or integrals... inherits the posterior,\" and since §5.3 explicitly differentiates every draw for κ, A, and B, this is a quantitative failure of the advertised property for derivative deliverables. The manuscript rules out rank truncation, metric reweighting, and replacing H_GN by H_stat, and then leaves \"cutoff response, tangent nonlinearity, and optimization-branch response\" as candidates; no correction or calibration factor is supplied. The claim should be weakened to a first-order tangent posterior with a known derivative under-coverage, or the interval construction should be corrected and recalibrated.","section":"§5.6"},{"comment":"The sentence \"the exact-ρ(0) distributions from the three constructions coincide\" is not a quantitative calibration. The headline deliverables ρ(0) and ν(0) require the deepest derivative rung of the fitted surface (§6.1–6.2), exactly the rung where §5.6 found the transport under-calibrated, yet no width ratio, coverage interval, or bootstrap error is reported. The four-mode 99% variance of Fig. 3e is an internal property of the empirical re-solved force draw ensemble, not a comparison to refits. Without an analogue of the §5.6 ratio for ρ(0) and ν(0), the flagship does not demonstrate derivative inheritance at the scale advertised in the abstract. Please report refit-to-transport width ratios for these deliverables, ideally with bootstrap intervals, and state explicitly whether the κ under-coverage persists.","section":"§6.4"}],"minor_comments":[{"comment":"The \"four uncertainty coordinates\" of the flagship are empirical re-solved force modes whose coefficient law is not the independent-Gaussian law of Eq. (15); make this distinction at the first occurrence so the abstract is not read as saying Eq. (15) generates the four-mode force posterior.","section":"Abstract/§6.3/§8"},{"comment":"The caption labels the panel \"(M, κ) joint cloud\" but the panel title printed in the figure is \"(c) (M, ) joint cloud\"; the missing \"κ\" should be restored.","section":"Figure 2c"},{"comment":"References to R. A. Ibata et al. 2026a and 2026b are to unpublished/in-prep papers; since §1 says the optimizer and SPLA machinery come from these papers, please confirm availability or summarize the needed properties in an appendix.","section":"References"},{"comment":"The empirical-Bayes grid uses the same data to select α and to produce the flagship figures; the paper states this conditioning, but the abstract's unqualified \"posterior\" should say more visibly that the reported object is conditional on the evidence-selected prior strength.","section":"§6.5"}],"recommendation":"major_revision","confidential_remarks":"To the editor: the paper is close to publishable, and I do not see grounds for rejection. The central construction is a useful conditional first-order transport, the controlled validation is unusually careful, and the code is public. The required revision is focused: either add the missing derivative calibration in the flagship and correct or weaken the abstract's \"any derived quantity... inherits the posterior\" claim, or explicitly re-scope the paper to a conditional tangent posterior with a stated derivative under-coverage. I would also encourage the authors to make the distinction between Eq. (15) value-space modes and the empirical re-solved force modes more prominent in the abstract."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know about this paper is that it is a serious, carefully built method paper, not a repackaging of linearized Laplace. The algebra is indeed classical — they admit it — but the end-to-end construction is new: differentiating the full stationarity system including priors and calibration parameters, separating the repeated-experiment transport from the Laplace posterior transport, building a function-space Karhunen–Loève basis with derivative coherence, and adding a Q diagnostic that flags tangent-inadequate directions before expensive refits. The controlled validation on a mock rotation curve is the best part: the sister posterior agrees with an independent spline+MCMC reference to a few percent, and stationary warm refits match the function-band width at the 10–15% level.\n\nThe soft spot is exactly where the stress-test note points. The paper's headline claim is that derived quantities, including derivatives, inherit the posterior. But in its own calibration, the κ width from the transport is 0.0107, warm refits give 0.0145 (ratio 1.356, 95% CI [1.021,1.665]), and cold refits give 0.0243 (ratio 2.274). That is a 36% understatement on the same branch. The authors rule out truncation and metric reweighting, leaving second-order response or branch response as the cause. In the flagship, ρ(0) and ν(0) sit at the deepest derivative rung, and the paper only reports that the distributions 'coincide' — no width ratio against refits. The four-mode 99% variance is a property of the sister ensemble, not a calibration result. The central claim that Eq. (15) yields derivative-inheriting posteriors is therefore not yet demonstrated at the scale where it is advertised.\n\nThat said, the paper is honest about its limits. The Q diagnostic, the convergence certificate, the cold-start ensembles, and the explicit statement that the construction is first-order on one branch are all there. The evidence-selected α is a mild empirical-Bayes circularity, and the code/data are not released with a commit hash, but those are minor.\n\nWho is this for? Astrostatistics and scientific ML. It deserves a serious referee. I would send it to review, and ask the authors to do a derivative-width ratio check on the flagship and to soften the 'inherits the posterior' phrasing to 'inherits to first order, conditional on the declared branch.' If they do that, the paper is solid.\n\nRecommendation: engage with it — send to peer review with those requests.","headline":"A well-built first-order function-space posterior for neural surrogates, with honest validation, but derivative deliverable widths are understated by ~36% on a warm branch and the flagship never checks that ratio.","tokens_in":20318,"tokens_out":2717,"would_cite":true,"duration_ms":26954,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper shows that the uncertainty of a fitted neural surrogate lives in a low-dimensional space of coherent function deformations, with four modes carrying 99% of the force posterior variance in its 800-parameter flagship fit.","keywords":["function-space posterior","neural surrogate uncertainty","measurement-to-function transport","deformation modes","Karhunen-Loève expansion","Laplace approximation","uncertainty propagation","phase-space distribution function"],"falsifier":"Perform a warm perturbed-data refit ensemble at reduced noise ($\\gamma=1/8$) on the rotation-curve benchmark and check whether the ratio of refit to transport function-band width approaches 1; the paper's own data show a persistent residual-to-signal RMS of 0.30-0.38 even at small noise, so a clean falsifier is whether the residual vanishes or is absorbed by branch scatter as $\\gamma$ is further reduced.","tokens_in":19246,"feed_emoji":"📊","tokens_out":6567,"duration_ms":54515,"temperature":0.7,"pith_summary":"This paper argues that the scientifically useful uncertainty of a neural-network fit is not a cloud of thousands of correlated weights but a small set of coherent ways the fitted function itself can deform. Linearizing the converged fitting procedure with respect to the measurement noise defines a transport operator whose leading singular functions become deformation modes, and independent Gaussian coefficients on those modes generate whole smooth function draws. Because each draw is a function, derivatives and integrals of the draw are meaningful, so derived quantities such as force, density, surface density, and vertical frequency inherit a joint posterior automatically. In the flagship example, four modes carry 99% of the posterior variance of the vertical force from an 800-parameter phase-space fit, stable when the force basis is refined, and 4000 coherent draws propagate through the dynamical tower in under a second. The construction is conditional on the declared model choices, and its linearization is tested against full perturbed-data refits, which reveal that cold-restart branch scatter exceeds the local tangent prediction.","feed_headline":"Four modes capture 99% of an 800-parameter fit's uncertainty","feed_subtitle":"Neural-fit noise becomes coherent curve deformations, so forces and densities get error bars without MCMC.","key_machinery":"The load-bearing object is the measurement-to-function transport $T_X := J_{f,X} H_{\\mathrm{GN}}^+ J_m^{\\top}$ (Eq. 5), the linearized derivative of the whole fitting map: it tells how a small perturbation of whitened measurements moves the converged fitted function. Its weighted singular value decomposition yields orthonormal deformation modes $\\psi_k$ and singular values $s_k$; the generative sister model $F_\\zeta(x)=\\hat{f}(x)+\\sum_{k=1}^{K} s_k \\zeta_k \\psi_k(x)$ (Eq. 15) turns those into smooth function draws with independent standard-normal coefficients. The same machinery gives scalar intervals at one curvature solve, e.g. $\\sigma_M^2 = m^{\\top} H^{-1} m$, and propagates correlated deliverables jointly, using the Jacobians and Gauss-Newton curvature the NestyNet second-order optimizer already computes at convergence plus the evidence-prior curvature that makes the 800-parameter posterior proper.","core_discovery":"On its own terms, the central claim is that the generative sister model $F_\\zeta(x)=\\hat{f}(x)+\\sum_{k=1}^{K} s_k \\zeta_k \\psi_k(x)$ with $\\zeta\\sim\\mathcal{N}(0,I_K)$ is a valid low-dimensional posterior over the fitted function, so that any derived quantity, including derivative- or integral-dependent ones, inherits the posterior coherently. The $\\psi_k$ are the leading singular functions of the measurement-to-function transport $T_X = J_{f,X} H^+ J_m^{\\top}$, the derivative of the data-to-fit stationary branch, and $s_k$ their singular values; truncation after $K$ minimizes expected function-space error by the Eckart-Young property. The paper distinguishes the repeated-experiment covariance $T_{\\mathrm{rep}}$ from the local Gauss-Newton/Laplace posterior $T_{\\mathrm{post}}$ and shows empirically in the flagship example that they differ, with the posterior band wider by the prior term. It validates the tangent approximation against an independent spline-MCMC posterior on a mock rotation curve and against warm and cold refits of perturbed data; fixed-branch function widths agree at the 10-15% level, while cold starts reveal extra branch scatter.","pith_inferences":["A reader might infer that the same construction transfers to other differentiable surrogate classes beyond the segmented NestyNet model, as long as the optimizer exposes converged Jacobians and curvature and the fit is deterministic; the transport identity itself is generic response analysis.","The four-mode compression suggests that for derivative-dependent science, the relevant posterior lives near a low-dimensional manifold of function space, so a natural test is whether non-Gaussian coefficient laws over those same $K$ coordinates (which the paper says would leave the basis unchanged) resolve the persistent midplane-density derivative-gap systematic.","The cold-start excess (kappa-width ratio 2.274 versus 1.356 warm) implies that for problems with multiple basins, the local linearized posterior is only part of the uncertainty budget; a complete scientific error report would need to combine the sister with an explicit basin-averaging or model-averaging step.","Extending the method to time-dependent or field-valued surrogates would give each draw a coherent spatiotemporal field, allowing uncertainty-aware simulations whose initial conditions and forcing inherit calibrated correlated perturbations."],"forward_implications":["For any scientific deliverable computed from a converged surrogate fit, the posterior of the deliverable can be obtained from coherent function draws, including deliverables that require derivatives or integrals, at the cost of a curvature solve plus a few Jacobian-vector products.","The effective dimension of the posterior is set by the data, prior, model, and chosen deliverable, not by the number of network parameters: an $P=800$ fit collapses to $K_{1\\%}=4$ for the vertical-force posterior, and the count is stable when the force basis is refined from 8 to 32 splines.","The construction separates repeated-experiment estimator scatter from the Bayesian posterior; on the flagship fit the refit ensemble matches the repeated-experiment transport within sampling scatter while the posterior sister band is wider by the prior term.","Convergence of every fit is certifiable by a stationarity check ($\\eta_N, \\eta_f \\leq 10^{-3}$), and tangent-inadequate directions are flagged in advance by the $Q$ diagnostic, so the method carries its own validity conditions.","Because draws are functions, downstream nonlinear compressions such as $M(r_0)=r_0 v^2/G$, epicyclic frequency, and the collisionless-Boltzmann tower can be pushed exactly through each draw, preserving correlations among deliverables."],"supporting_citations":[{"why":"Supplies the linearized Laplace / Gauss-Newton reading of a converged fit as a locally Gaussian posterior.","marker":"D. J. C. MacKay 1992"},{"why":"Provides the optimality of the truncated singular basis, justifying the K-mode truncation via the Eckart-Young-Mirsky theorem.","marker":"C. Eckart & G. Young 1936"},{"why":"Paper I supplies the NestyNet second-order optimizer, segmented model, and evidence-based prior that the transport is built from.","marker":"R. A. Ibata et al. 2026a"},{"why":"Supplies the affine-invariant ensemble sampler used as an independent MCMC calibration of the rotation-curve posterior.","marker":"J. Goodman & J. Weare 2010"},{"why":"Provides the practical emcee MCMC implementation used for the independent spline posterior.","marker":"D. Foreman-Mackey et al. 2013"},{"why":"Defines the razor-thin exponential-disk rotation curve that generates the mock data in the validation suite.","marker":"K. C. Freeman 1970"},{"why":"Supplies the dynamical formulas (epicyclic frequency, Oort constants, Poisson and Gauss relations) used for the deliverable pushforwards.","marker":"J. Binney & S. Tremaine 2008"}],"fun_headline_variants":["Four modes beat MCMC for 800-parameter neural fit errors","Low-dimensional coherent errors: 4 modes, 0.8s for 4000 draws","Function-space posterior: 4 modes replace 800-parameter MCMC","Avoid MCMC: 4 function modes carry 99% of fit uncertainty"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The construction rests on the assumption that if the measurements were wiggled by the size of their quoted noise, the refitted curve would move proportionally, with no significant curvature and no jump to a different solution of the fit; where that fails, the reported bands understate the refit scatter.","fun_headline_variants_meta":{"raw":{"variants":["Four modes beat MCMC for 800-parameter neural fit errors","Low-dimensional coherent errors: 4 modes, 0.8s for 4000 draws","Function-space posterior: 4 modes replace 800-parameter MCMC","Avoid MCMC: 4 function modes carry 99% of fit uncertainty"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000877,"raw_usage":{"total_tokens":3873,"prompt_tokens":1107,"completion_tokens":2766,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":723,"completion_tokens_details":{"reasoning_tokens":2682}},"tokens_in":723,"tokens_out":2766,"duration_ms":16767,"temperature":1.0,"reasoning_tokens":2682,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T10:47:52.542618+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Perform a warm perturbed-data refit ensemble at reduced noise ($\\gamma=1/8$) on the rotation-curve benchmark and check whether the ratio of refit to transport function-band width approaches 1; the paper's own data show a persistent residual-to-signal RMS of 0.30-0.38 even at small noise, so a clean falsifier is whether the residual vanishes or is absorbed by branch scatter as $\\gamma$ is further reduced.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the linearized Laplace / Gauss-Newton reading of a converged fit as a locally Gaussian posterior."},{"cited_title":"Psychometrika , volume =","cited_arxiv_id":null,"evidence_quote":"Provides the optimality of the truncated singular basis, justifying the K-mode truncation via the Eckart-Young-Mirsky theorem."},{"cited_title":"Communications in Applied Mathematics and Computational Science , volume =","cited_arxiv_id":null,"evidence_quote":"Supplies the affine-invariant ensemble sampler used as an independent MCMC calibration of the rotation-curve posterior."},{"cited_title":"and Lang, Dustin and Goodman, Jonathan , title =","cited_arxiv_id":null,"evidence_quote":"Provides the practical emcee MCMC implementation used for the independent spline posterior."},{"cited_title":", title =","cited_arxiv_id":null,"evidence_quote":"Defines the razor-thin exponential-disk rotation curve that generates the mock data in the validation suite."},{"cited_title":"2008 , isbn =","cited_arxiv_id":null,"evidence_quote":"Supplies the dynamical formulas (epicyclic frequency, Oort constants, Poisson and Gauss relations) used for the deliverable pushforwards."}],"review_version":1}