Preserved original model output. Check claims against the independent audit →

1. Factual / physics errors

• The draft incorrectly folds three different effects into one per-pixel sum: “distinct clouds along the line of sight, sub-beam structures, and, for a power-spectrum estimate, all the pixels in the mask.” Pixels elsewhere in a mask are not components of $\mathcal P_\nu(\hat n)$. Line-of-sight mixing, beam averaging, and spatial variation across an analysis region produce different statistics and must be modelled separately. The quoted Gaussian-$\sigma_\beta$ decorrelation formula concerns a highly idealized ensemble of spatially varying indices, not general line-of-sight mixing.

• The single-layer polarization formula is presented as exact but is only an approximation. In standard dust radiative-transfer expressions, the grain-orientation term also produces a small correction to $I$; $F$ is not generally a frequency-independent scalar when alignment, composition, and temperature vary within the layer. More seriously, the limitations section later says $\gamma_i$ “matters for the layer weights in intensity.” That contradicts both the displayed model and the physics: inclination principally changes polarized emissivity, not optically thin total thermal intensity.

• The statement that Faraday rotation “constrain[s] the line-of-sight component and hence the inclination $\gamma_i$” is false for the proposed application. Rotation measure gives an electron-density-weighted integral $\int n_e B_\parallel\,ds$, usually without dust-layer separation. A plane-of-sky polarization angle gives the direction of $B_\perp$, not its magnitude, so the two do not determine $\tan\gamma=B_\parallel/B_\perp$. Even the more recent reconstruction gives a line-of-sight-averaged field, not $B_\parallel$ for each extinction layer (Hutschenreuter et al.).

• “The starlight $q_i,u_i$ of a layer multiplied by $R_{P/p}$ gives that layer’s 353-GHz $Q_i,U_i$ directly” is incorrect as written. Optical dichroic polarization is parallel to the projected field, whereas thermal-emission polarization is perpendicular, so the emission complex Stokes vector acquires a minus sign under the $90^\circ$ rotation. More importantly, $R_{P/p}$ was measured for selected, integrated diffuse sightlines with nearly orthogonal optical/emission angles, not demonstrated as a universal per-cloud conversion. The original measurement also found that dust models underpredicted the ratio by about a factor of 2.5, which is hardly support for a precise layer transfer (Planck intermediate XXI).

• The claim that $\cos^2\gamma_i$ and $F_i$ “cancel” in that transfer is only true for an optically thin, uniform layer when emission and extinction weight exactly the same aligned population and field distribution. They weight grain sizes, temperatures, and the stellar pencil beam versus the emission beam differently. Differential stellar Stokes parameters are also only approximately additive at low optical depth.

• The decorrelation formula

is stated without its restrictive assumptions: a fixed morphology, spatially uniform temperature, Gaussian index fluctuations independent of amplitude, and a particular ensemble treatment. Amplitude–index correlations, masking, beam smoothing, noise, and line-of-sight vector cancellation alter it.

• Likewise, $\delta C_\ell^{BB}\simeq4\delta\psi^2C_\ell^{EE}$ describes a small, spatially uniform rotation with negligible $EB$. A frequency-dependent angle generated by spatially varying layer mixtures is not generally a uniform rotation: it couples multipoles and contains terms involving the original $B$, $EB$, and the spatial spectrum of $\delta\psi$. The quoted “1°, 2°, 3° leave 0.2%, 1%, 2%” numbers are illustrative special cases, not general residual forecasts.

• The numerical two-layer examples omit essential assumptions about whether “30%” is a fraction of the summed layer amplitudes or an amplitude ratio, whether layer morphologies are correlated, and what scalar is optimized in the subtraction. The subsequent conversion of a residual fraction to $\delta r$ is therefore not uniquely defined.

• Zelko et al. do not establish that five integrated intensity bands provide accurate $T_i,\beta_i$ for arbitrary individual layers. Their reconstruction proposes parameters for hundreds of millions of voxels through a heavily regularized global inverse problem; the paper reports a two-component Cepheus demonstration at approximately $1\sigma$, not validated per-layer SED templates over the sky (Zelko et al.). Similarly, Edenhofer’s 516 radial bins are samples of a correlated reconstruction, not 516 independent distance-resolution elements (Edenhofer et al.).

• Ghosh et al. and Adak et al. are misrepresented as demonstrations of fitting physical distance-layer templates. They use H I phase/column templates plus a phenomenological magnetic-field model to reproduce statistics at 353 GHz. That does not validate recovery of several distance-resolved polarized SEDs; Adak et al. describe a seven-parameter regional simulation model, not a tomographic component-separation solution (Adak et al.).

• The cross-spectral expression

is valid only if each layer has a spatially constant SED over the region. The proposed model explicitly has $\beta_i(\hat n)$ and $T_i(\hat n)$. In that case the SED factors cannot be pulled outside the harmonic transform; multiplication by spatially varying SED maps creates mode coupling.

• The layered-fit parameter count is internally inconsistent. If externally supplied $\mathcal P_i$ fixes each layer’s $Q/U$ ratio, each layer has one scalar amplitude. If $Q_i$ and $U_i$ are two free amplitudes, the external angle is not fixing the phase, and different $Q/U$ ratios do not improve spectral identifiability. Adding free $\delta\psi_i$, $\beta_i$, and $T_i$ also makes the claimed “linear problem” nonlinear except after a local linearization.

• The discussion conflates cMILC with fitting externally predicted spatial moment templates. Constrained moment ILC deprojects prescribed spectral moment vectors and pays a variance penalty; it does not use 3D maps to fix the spatial moments, nor does it automatically “break” their degeneracy with $r$ (Remazeilles et al.).

2. Missing physics

• The proposal needs the joint distribution

including correlations between density, radiation field, composition, alignment, and magnetic geometry. Separate maps or priors for their marginal distributions are insufficient because decorrelation is controlled precisely by correlations between layer SED and layer angle.

• “Grain-size distribution” is treated as though it determines the polarized millimetre SED. In the Rayleigh regime, the opacity slope is dominated by dielectric properties, composition, shape, inclusions, and solid-state physics, not size alone. $R(V)$ is a degenerate proxy for the total extinction curve; it is not a measurement of the aligned-grain size distribution. The empirical $R_V$-$\lambda_{\max}$ relation cannot supply a layer-specific alignment function without multi-band stellar polarimetry and a composition model.

• Dust temperature requires a 3D radiation field, shielding, and radiative transfer. Distance from “heating sources” and Local-Bubble geometry alone do not predict $T_i$. The draft supplies no model for the interstellar radiation field or its covariance with dust density and grain properties.

• There is no selection-function model for the stellar tomography: stellar distances, extinction-dependent incompleteness, intrinsic stellar polarization, beam-versus-pencil-beam mismatch, the last-star boundary, and misassignment of extinction jumps to polarization jumps all affect inferred layer Stokes vectors.

• H I morphology is not synonymous with dust morphology or magnetic angle. Molecular, ionized, CO-dark, and H I-poor dust are omitted, while velocity crowding and high-latitude velocity–distance degeneracy remain. In the BK field, the H I template correlates with only roughly 10–65% of the observed $B$-mode map power depending on scale, so it is demonstrably incomplete as a cleaning template (BICEP/Keck XVI).

• No treatment is given for correlated posterior errors in 3D extinction/emission maps. Radial anticorrelations can manufacture or erase apparent layers. Template noise and errors in regressors bias fitted coefficients unless marginalized jointly.

• The proposal lacks a scale-dependent treatment. Degree-scale stellar angles may help LiteBIRD at low $\ell$, but they cannot specify the arcminute field and SED structure relevant to ground-based maps. Beam averaging itself can produce decorrelation even if the distance layers were recovered perfectly.

• The foreground likelihood is incomplete. Synchrotron curvature, spatially varying dust–synchrotron correlation, calibration, bandpass integration, beam uncertainty, polarization efficiency, angle calibration, filtering, $E/B$ purification, lensing $B$-modes, and chance CMB–template correlations must enter the same end-to-end inference, not be appended as caveats.

• The draft never establishes whether the extra physical priors reduce bias more than they increase variance or introduce prior bias. That is the central dedusting question. Existing studies show that apparently adequate two-parameter fits can become severely biased in special angle–SED configurations (McBride et al.), while free moments can recover $r$ only with a substantial uncertainty penalty (Vacher et al.).

3. Overclaims

• “What is needed is external information that fixes the structure of the superposition. That is what the three ingredients provide.” None currently fixes it. Extinction maps constrain column placement; $R(V)$ weakly constrains total-grain extinction properties; projected-field tracers constrain selected angles. Polarized emissivity per layer remains largely latent.

• “The 3D emission fit gives $T_i$ and $\beta_i^I$ per layer” upgrades a regularized, underdetermined reconstruction into a measured data product. The cited work supports feasibility studies, not the precision required for $r\sim10^{-3}$.

• “Current band counts can handle $N\le3$” is unsupported parameter counting. Rank is not sensitivity: dust-layer MBB spectra are nearly collinear over CMB bands, and the result depends on noise, bandpasses, angular smoothing, angle-prior precision, synchrotron complexity, and calibration.

• Priors “of a few tens of per cent” on amplitudes and “small” angle corrections are simply asserted. No data establish these widths or their spatial covariance, and tens-of-percent amplitude errors are not obviously compatible with a $10^{-3}$ $r$-budget.

• The claim that grain physics can provide a prior of width $0.1$ on $\beta^P-\beta^I$ has no demonstrated inference behind it. Competing composition models and Planck analyses are cited precisely because they disagree.

• “Real 3D structure replaces ad-hoc layer statistics” is false for the listed simulation models. PySM d12 and related constructions are plausible synthetic skies calibrated to projected statistics; they are not observations of the actual layer structure in a target CMB field.

• “The forward model predicts” spin-moment maps and the “full multi-frequency dust covariance” is too strong. It predicts them conditional on uncertain and mostly unmeasured polarized layer emissivities. With spatially varying SEDs, the simple covariance expression is itself inadequate.

• The Guillet et al. result cannot support the statement that temperature variation “may be the dominant source.” SRoll2 was compatible with temperature-only variation, while PR4 was compatible with neither temperature-only nor index-only variation; the paper explicitly exposes data-processing/systematics dependence (Guillet et al.).

• The final “residual below $10^{-4}$ in $r$” sanity check assumes the very quantities the method cannot currently know: polarized rather than extinction column fractions, complete layer inventory, exact layer angles, a fixed $\Delta\beta$, and no sub-layer structure or instrumental error. It is not a validation or forecast.

4. Weak links in the pipeline

5. Referee demands

• Define separately LOS decorrelation, beam decorrelation, and mask-level spatial SED variation, and state which effect each external tracer is intended to constrain.

• Replace parameter counting with Fisher or posterior-conditioning calculations using realistic bandpasses, noise, beams, calibration parameters, synchrotron, and external-prior covariance. Report condition numbers and failure modes as layer SEDs and angles become similar.

• Construct an explicit hierarchical likelihood in which extinction, emission, stellar polarization, H I, and microwave polarization are noisy tracers of latent layers. All radial, angular, and cross-tracer covariance must be propagated.

• Demonstrate empirically that recovered extinction or H I layers correspond to distinct polarized-emission layers. This requires held-out multifrequency polarization tests, not agreement with the same Planck intensity data used to build the templates.

• Calibrate $R_{P/p}$ per environment and layer class, including the optical/emission $90^\circ$ sign, beam mismatch, stellar-distance uncertainty, and possible frequency dependence. Show its scatter, rather than adopting a universal scalar.

• Provide an end-to-end $r$ analysis comparing at least: single MBB, spatially varying MBB, free moments, cMILC, H I templates, and the proposed layered prior. Report $r$ bias, variance, coverage, and Bayesian evidence or goodness of fit.

• Test robustness to missing distant dust, incorrect layer number, merged/split clouds, molecular and ionized components, radial map covariance, unresolved within-layer tangling, and incorrect grain-composition priors.

• Replace the universal $1-R^2\rightarrow\delta r$ conversion with the actual residual $BB$ spectrum passed through the experiment’s mask, transfer function, delensing, likelihood, and foreground marginalization.

• Validate on independent sky regions and data splits. A model tuned on Planck/IRAS and then “validated” against Planck residuals is partly circular.

• State a quantitative requirement on external angle, amplitude, and SED accuracy needed to reach $|\Delta r|<10^{-3}$, then compare that requirement with present PASIPHAE, H I, extinction, and emission-map performance.

6. Severity list

• blocker The draft never demonstrates identifiability or acceptable $r$ bias/variance for the proposed multi-layer fit.

• blocker It treats extinction/intensity tomography as though it supplied polarized layer SEDs and amplitudes.

• blocker The direct per-layer conversion $q_i,u_i\rightarrow Q_i,U_i$ via universal $R_{P/p}$ is physically and observationally unjustified.

• blocker External-map uncertainties and tracer-selection functions are not propagated into the CMB likelihood.

• major The parameter count and claim of linearity are internally inconsistent.

• major Faraday rotation does not determine dust-layer inclination.

• major Grain size and $R(V)$ are overinterpreted as constraints on millimetre polarized emissivity and $\beta^P-\beta^I$.

• major The cross-spectral covariance equation is invalid for the spatially varying layer SEDs the proposal assumes.

• major LOS, beam, and sky-mask decorrelation are conflated.

• major The rotation-to-$BB$ and decorrelation-to-$\delta r$ formulae are used outside their restricted assumptions.

• major HI-based phenomenological simulations are mis-cited as validation of distance-resolved polarized component separation.

• major Claims that predicted moments break the $r$ degeneracy and that $N\le3$ is feasible lack forecasts.

• major The $r<10^{-4}$ residual example assumes unavailable polarized layer information and perfect modelling.

• minor The optical-to-emission Stokes conversion omits the $90^\circ$ sign reversal.

• minor The simplified single-layer $I,Q,U$ equations should be labelled approximations.

• minor “BICEP2’s detection turned out to be dust” is historically too categorical; the excess became consistent with dust plus lensing, with no significant primordial detection.