{"id":"88198793-0cd0-47f3-9485-2fdf93fdf80d","arxiv_id":"2608.07377","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":10,"one_line_summary":"Combining clipped and standard shear correlation functions from KiDS-1000 tightens S8 by 16% and w0 by 24%, yielding S8=0.724±0.027 and w0=-1.24+0.26-0.28, consistent with ΛCDM.","lead":"A new KiDS-1000 analysis applies a 'clipping' filter that removes the densest regions of cosmic shear maps, then uses these clipped statistics together with the standard ones to sharpen cosmological constraints. The combination tightens the matter clustering parameter S8 by 16% and the dark energy parameter w0 by 24%, with results consistent with earlier KiDS measurements.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The IA bias model assumes separability from cosmology and masking that is untested and potentially violated; it is the main load-bearing vulnerability in the combined-probe claim.","rationale":"The paper is careful and the overall consistency with A21 for Ωm, S8, and w0 is reassuring. I do not see a reason to reject the central claim; the mock tests, the emulator-error accounting, and the independent A21 comparison are real supporting evidence. The vulnerability is that the IA bias, which is one of the dominant systematics, is imported from a single-cosmology, unmasked simulation suite. The paper explicitly acknowledges this assumption in Sec. 2.2(iii), but it is not tested by the mock validations because those mocks use the same model. The 3.2σ AIA offset relative to A21 is a concrete hint that the IA model is not fully correct, and since clipping is specifically designed to add IA-sensitive information, this is the most load-bearing point. The proposed check—recomputing B_IA with the footprint—would settle whether masking is the culprit. If the offset persists with the footprint included, the cosmology-dependence of the IA bias shape would be the next suspect. The reader's weakest_assumption identified the same general area; I partially agree because the paper's separability assumption has two parts (cosmology and masking), and the masking part is the sharper one for the clipped statistic. The verdict remains CONDITIONAL pending a direct test of this assumption.","tokens_in":32655,"tokens_out":12336,"duration_ms":120019,"concrete_test":"Take a subset of 20 realisations from the IA Set and rerun the full 18-tile KiDS-1000 footprint construction (mask, galaxy positions, per-object weights) for AIA=2 and AIA=0, then recompute B_IA for the clipped and unclipped ξ±. If the footprint-infused B_IA differs from the current unmasked B_IA by more than the realisation noise on B_IA, or if the AIA posterior shifts by more than ~0.1 when the masked B_IA is used, the separability assumption fails and the combined-probe constraints must be re-derived.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that adding the clipped ξ± to the unclipped ξ± yields 16% (S8) and 24% (w0) precision gains—rests on the IA bias model of Secs. 2.2(iii) and 3.3. The IA Set is generated at the fiducial cosmology only, without the KiDS-1000 footprint. Equation (11) defines B_IA as ξ(AIA≠0)−ξ(AIA=0) from these full-lightcone mocks, and this single bias shape is then scaled by AIA and added to the cosmology emulator at every point of the parameter space. This assumes (i) the IA term is additively separable from the cosmological term, including the GI cross-term, whose amplitude depends on cosmology; and (ii) the survey mask does not change the IA response of the clipped statistic. Both are load-bearing because clipping is a non-linear, mask-dependent operation applied to the reconstructed convergence field. The mock validations in Appendix C1 do not test either assumption: the systematics-contaminated mocks are generated with the same emulator model, so they can only confirm internal consistency. The 3.2σ disagreement between AIA here and in A21 is exactly the kind of signature a biased IA bias would produce. Since the headline gain includes a 27% tighter IA constraint and the clipped probe is informative precisely because it responds differently to IA, an undiagnosed bias in B_IA would corrupt the combined likelihood and the claimed improvements.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents the first systematics-controlled application of density-field clipping to cosmological inference from real weak lensing data. Clipping replaces convergence-map pixels above a fixed threshold with the threshold value and recovers a clipped shear catalogue; the authors model the clipped and unclipped shear correlation functions jointly. The model is built from the cosmoSLICS and SLICS N-body suites, with the KiDS-1000 footprint, shape noise, multiplicative and additive shear calibrations, and photometric redshift distributions injected into mock catalogues. Gaussian-process emulators trained on 26 cosmologies predict the cosmological dependence; bias emulators and linear models describe intrinsic alignments, photo-z shifts, baryonic feedback, and source-lens clustering. The combined clipped and unclipped data vector yields Omega_m=0.263+0.035-0.038, S8=0.724+/-0.027, w0=-1.24+0.26-0.28, consistent with A21, with reported precision gains of 16% for S8 and 24% for w0 over the unclipped statistic alone, a 27% improvement on the IA amplitude, and an upper limit on baryonic feedback. Validation includes 1240 covariance realizations, leave-one-out emulator cross-validation with covariance inflation, a systematics-free SLICS mock, and a systematics-contaminated mock.","tokens_in":2290,"tokens_out":2427,"duration_ms":134065,"significance":"If the headline claims hold, this is a substantive methodological advance: it is the first demonstration that a non-Gaussian (HOWLS-type) statistic can be modelled end-to-end with systematics and applied to real lensing data, and that clipping extracts additional information while preferentially using scales least affected by baryons. The paper's strengths include the large and carefully constructed mock suites, the explicit inflation of the covariance with measured emulator error (Eq. 10), the check of Gaussianity of the data vector, the use of both Gaussian-with-Hartlap and Sellentin-Heavens likelihoods, the open-source release of the clipping and likelihood codes, and an independent simulation-based consistency check of the KiDS-1000 A21 result. The recovered consistency with LCDM and with A21 for the cosmological parameters, together with the agreement of the Delta-z posteriors with A21, gives reasonable confidence that the cosmology result is not dominated by a gross pipeline error.","major_comments":[{"comment":"The intrinsic-alignment bias model is the main load-bearing assumption of the combined-probe claim, and it is currently untested in precisely the two directions that matter. Equation (11) defines B_IA as the difference between AIA != 0 and AIA = 0 measurements from the IA Set, which spans full 10x10 deg^2 lightcones without the KiDS-1000 footprint and is run only at the fiducial cosmology. This single bias shape is scaled by AIA and added to the cosmology emulator at every point of the parameter space (Sec. 3.3). Clipping is a nonlinear, mask-dependent operation: the mask is reapplied to the reconstructed kappa map before thresholding, and the residual map Delta-kappa is masked before interpolation to galaxy positions (Sec. 3.1, Eqs. 1-3). The assumption that the IA bias of the clipped statistic is unaffected by the ~23% removed area is asserted in Sec. 2.2(iii) but not demonstrated, and the GI term in B_IA depends on cosmology through the lensing efficiency and the matter power spectrum. The Appendix C1 validation cannot catch a bias in B_IA because the systematics-contaminated mocks are generated with the same emulators and linear models, so that validation tests internal consistency only. The 3.2-sigma offset of the inferred AIA from the A21 value is exactly the signature a mask- or cosmology-dependent B_IA would produce, and since the combined probe is claimed to improve the AIA constraint by 27%, this matters for both the headline precision gains and the IA secondary result. I recommend that the authors (i) compute B_IA with a KiDS-like mask applied to the IA Set realizations and quantify the change in both the clipped and unclipped bias; (ii) propagate a conservative systematic error on B_IA and demonstrate that the 16% S8 and 24% w0 gains survive; and (iii) state explicitly that the Appendix C1 contaminated-mock test does not validate the bias model.","section":"Sec. 2.2(iii) and Eq. (11)"},{"comment":"The baryonic feedback model linearly rescales the magneticum dark-matter-only versus hydro difference with a single parameter bbary in [0,2], with the upper half of the prior representing an extrapolation to twice the magneticum feedback level. The comparison in Fig. 8 with hmcode for log10(T_AGN/K) = [7.6, 7.8, 8.0] shows that the fractional suppression of the unclipped correlation functions is not linear in feedback strength, and the clipped auto-correlations behave non-monotonically: stronger feedback reduces small-scale power, which decreases the amount of clipping and can increase power at 5-10 arcmin. A one-parameter linear rescaling therefore imposes a specific shape for the feedback effect, and the headline secondary result bbary < 0.97 depends on that shape. Please validate the linear-scaling ansatz against intermediate feedback strengths (for example, an additional magneticum or BAHAMAS run, or the multiple AGN-temperature nodes already available through hmcode for the unclipped part), or quantify the sensitivity of the upper limit to relaxing the linearity assumption.","section":"Sec. 2.2(v) and Sec. 3.3"},{"comment":"The two mock validations cover complementary but incomplete ground. The systematics-contaminated mock data are drawn from the trained emulators and linear models themselves, so they verify that the sampler and likelihood recover the input values given the adopted bias model, but they cannot falsify that model; the SLICS-based systematics-free test verifies emulator generalisation to a sister simulation suite but contains no IA, photo-z, or baryon contamination. The paper would be strengthened by a third test in which the contamination is generated from a model outside the inference set, for example a TATT-motivated IA signal or hydrodynamics from an independent simulation code, to provide at least one end-to-end check that the decompositions of Eqs. (11)-(13) and the linear BB rescaling recover the truth.","section":"Appendix C1"}],"minor_comments":[{"comment":"The text states that the unclipped analysis is prior-limited for w0, yet Table 2 reports a 24% precision gain for w0 without qualification. Please report the unclipped marginalised width for w0 and state clearly how much of the quoted gain reflects the prior rather than information in the data; the same caveat applies to the FoM factors of 1.6 and 4.0 in Sec. 4.","section":"Sec. 3.4 and Table 2"},{"comment":"In Eq. (4) the argument of the tangential and cross components is written with theta_g,b for both galaxies in the numerator and denominator; the second position should refer to galaxy a (theta_g,a) so that the pair weighting is unambiguous.","section":"Sec. 3.1 and Eq. (4)"},{"comment":"The claimed 3.2-sigma disagreement of AIA with A21 should be justified: with the quoted values AIA = 0.39+0.32-0.37 (A21) and AIA = -0.50+0.19-0.20 (this work), a naive quadrature combination gives roughly 2.1-2.3 sigma depending on how the asymmetric errors are combined.","section":"Sec. 4"},{"comment":"There is a typographical error in the abstract: 'complimentary information' should be 'complementary information'.","section":"Abstract"},{"comment":"The comparison with A21 for Omega_m is made for different models (A21 fixes w0 = -1, this work varies w0); one sentence noting the implication of that difference for the 0.8-sigma consistency statement would help the reader interpret the comparison.","section":"Sec. 4"},{"comment":"The clipping threshold (kappa_c = 0.010) and smoothing scale (sigma_s = 6.6 arcmin) are inherited from G18 and revalidated with a single test; since these settings control the ~20% clipped area and the depth of the clipping trough, a brief exploration of the sensitivity of the final constraints to (kappa_c, sigma_s) would strengthen the robustness picture.","section":"Sec. 3.1"}],"recommendation":"major_revision","confidential_remarks":"The main risk is that the central claim depends on the separability and mask-independence assumptions embedded in Eq. (11), which the authors acknowledge in Sec. 2.2(iii) but validate only self-referentially via Appendix C1. If the masked-IA robustness test is feasible, the paper could be acceptable after revision; otherwise the 3.2-sigma AIA discrepancy with A21 remains a legitimate concern for the combined-probe gains. The paper is a good fit for MNRAS, and the machinery (emulator validation, covariance construction, code release) is solid."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First, what you should know: the paper delivers the first systematics-controlled clipped lensing constraints on real data. That claim stands. It includes tomographic bins and the clipped ξ− for the first time, forward-models the KiDS-1000 footprint into the mocks, and validates on 1240 covariance realisations plus external SLICS mocks. The precision gains — 16% on S8, 24% on w0 — are relative to their own unclipped emulator analysis, not to A21's pipeline, but the mock tests support the relative gain. The cosmological constraints are consistent with A21, which is an independent pipeline; that is real validation.\n\nWhat is genuinely new: clipping as a usable statistic rather than a proof of concept, with systematics models for IA, baryons, photo-z, and source-lens clustering. The emulator error inflation is carefully done, and the robustness tests across error definitions are reassuring.\n\nSoft spots, in order of importance. The IA bias model assumes separability from cosmology and no interaction with the survey mask, but the IA Set lacks the footprint. The bias B_IA is computed as a difference between full-lightcone mocks and then added to every emulator prediction. The authors flag this as an assumption, but it is not tested. Because the clipped probe is non-linear and mask-dependent, and because the inferred AIA disagrees with A21 by 3.2σ, this is the place a referee should push. It is not fatal to the central claim — the cosmological results are consistent and the mock recovery is good — but a dedicated test (footprint-infused IA mocks, or at least a mask-dependent IA bias estimate) would considerably strengthen the paper. Minor: baryonic feedback is a linear scaling of one hydro simulation (magneticum), with the upper bound driven by that extrapolation; the wording could be clearer. Also, the headline precision gains are internal to the emulator-based unclipped baseline; readers should not compare them directly to A21's errors.\n\nOverall: this is a serious, honest paper. The central result — clipping extracts extra information from two-point lensing statistics with systematics under control — is supported. The IA assumption is the main unresolved question. The paper deserves peer review, and I would send it out. A good referee will ask for the IA test or a clear discussion of its feasibility, but the analysis as it stands is publishable with revision.","headline":"A careful, genuinely new application of clipping to KiDS-1000 with credible precision gains, but the IA bias modelling is the soft spot to probe in review.","tokens_in":33504,"tokens_out":1921,"would_cite":true,"duration_ms":17268,"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":"Combining clipped and unclipped shear correlations extracts extra cosmological information from KiDS-1000 data, tightening S8 by 16% and w0 by 24%.","keywords":["clipped cosmic shear","KiDS-1000","weak lensing","intrinsic alignments","baryonic feedback","Gaussian process emulators","wCDM","S8 tension"],"falsifier":"Build intrinsic-alignment mocks that carry the same 18-tile masked KiDS-1000 footprint as the cosmology set and recompute the intrinsic-alignment bias in the clipped correlation functions; if the masked and unmasked biases differ by more than the emulator-plus-statistical error budget, the separability assumption fails and the reported $A_{\\rm IA}$, which already sits $3.2\\sigma$ from the Asgari et al. (2021) result, would flag a biased model. A complementary check is to run the identical combined pipeline on KiDS-Legacy or DES Year 6 data and ask whether the same $S_8$ and $A_{\\rm IA}$ are recovered.","tokens_in":32399,"feed_emoji":"🔭","tokens_out":13308,"duration_ms":101348,"temperature":0.7,"pith_summary":"This paper argues that a simple preprocessing step — clipping, or cutting out the densest patches of the projected matter field before measuring two-point shear correlations — recovers cosmological information that ordinary two-point statistics throw away, because the cosmic lensing field is strongly non-Gaussian. Applied to the fourth data release of the Kilo-Degree Survey (KiDS-1000) with all major systematics forward-modelled by simulation-based emulators, the clipped and unclipped statistics jointly tighten the dark-energy equation-of-state parameter $w_0$ by 24% and the clustering amplitude $S_8$ by 16% relative to the conventional analysis alone, yielding $\\Omega_{\\rm m} = 0.263^{+0.035}_{-0.038}$, $S_8 = 0.724 \\pm 0.027$, and $w_0 = -1.24^{+0.26}_{-0.28}$. The constraints stay consistent with $\\Lambda$CDM and with the earlier KiDS-1000 analysis of Asgari et al. (2021), which used a completely independent modelling pipeline. A sympathetic reader should care because the gain comes from larger, linear scales that are least contaminated by baryonic feedback: if the claim holds, existing lensing surveys already contain extra cosmological and astrophysical information that clipping can extract at no observational cost.","feed_headline":"Clipping densest lensing patches tightens S8 by 16% and w0 by 24%","feed_subtitle":"Cutting the densest regions out of KiDS-1000's lensing maps extracts information ordinary two-point statistics miss.","key_machinery":"The load-bearing object is the clipped shear correlation function $\\xi^c_\\pm$. Convergence maps are reconstructed from smoothed galaxy ellipticity maps via Kaiser-Squires inversion (Gaussian smoothing $\\sigma_s = 6.6$ arcmin); every pixel with $\\kappa \\geq \\kappa_c = 0.010$ is set to the threshold value $\\kappa_c$, clipping roughly 20% of the observed area; the residual map $\\Delta\\kappa$ is inverted back to residual ellipticities and subtracted from the observed galaxy ellipticities, so only galaxies sitting on convergence peaks are modified. The correlation function of these clipped ellipticities, measured in nine angular bins across five tomographic redshift bins, forms the $\\xi^c_\\pm$ data vector. Its cosmological and systematic dependence is predicted by Gaussian-process emulators trained on the 26-cosmology cosmoSLICS suite (sampling $\\Omega_{\\rm m}$, $S_8$, $h$, $w_0$), with the covariance estimated from 1,240 independent SLICS realisations and the intrinsic-alignment, photo-$z$, and baryonic-feedback biases modelled as differences between dedicated contaminated and uncontaminated mock sets. The emulator accuracy is validated by leave-one-out cross-validation and folded into the covariance as an error term.","core_discovery":"The central claim is that two-point shear correlation functions measured on a clipped convergence field carry cosmological information that is sufficiently independent of the unclipped measurement to sharpen parameter inference when the two are combined. The paper reports the first systematics-controlled application of clipping to real lensing data, with tomographic binning, masked footprint infusion, per-object shear calibration, and emulators trained on the cosmoSLICS, SLICS, intrinsic-alignment, photo-$z$, and magneticum simulation suites. On KiDS-1000 data the combined probe gives $\\Omega_{\\rm m} = 0.263^{+0.035}_{-0.038}$, $S_8 = 0.724 \\pm 0.027$, $w_0 = -1.24^{+0.26}_{-0.28}$, tightening $S_8$ by 16% and $w_0$ by 24% over the unclipped analysis, improving the figure of merit in the $\\Omega_{\\rm m}$–$S_8$ plane by a factor of 1.2, and reproducing the Asgari et al. (2021) cosmology through an independent forward-modelling pipeline. It further finds that the clipped statistic breaks degeneracies the unclipped probe cannot: the intrinsic-alignment amplitude is constrained 27% more tightly ($A_{\\rm IA} = -0.50^{+0.19}_{-0.20}$), and a first upper limit on baryonic feedback, $b_{\\rm bary} < 0.97$, is placed from lensing alone. The paper attributes these gains to clipping's decoupling of scale-dependent information: removing high-density peaks suppresses the small scales where baryonic feedback dominates while preserving the larger, cleaner scales.","pith_inferences":["Editorial inference: the mechanism behind the gain — clipping decouples scale-separated information — suggests the same transform could sharpen other probes, such as clipped galaxy clustering or a clipped 3x2-point analysis, and the cross-statistic between clipped and unclipped fields (excluded here to keep the covariance invertible) is a natural addition once more realisations are available.","Editorial inference: the 3.2$\\sigma$ offset in $A_{\\rm IA}$ relative to Asgari et al. (2021) could be a first hint that clipping is sensitive to scale-dependent intrinsic-alignment physics (luminosity- or redshift-dependent alignment, or tidal torquing) that the simple NLA model does not capture; a testable extension would be to fit a TATT-like model to the combined data vector.","Editorial inference: the baryon-feedback limit is set more by emulator noise on small scales than by the information content of the data — the paper shows the $b_{\\rm bary}$ constraint tightens sharply when emulator error is handled by propagating the Gaussian-process covariance — so targeted improvements to small-scale emulation accuracy (denser simulation grids) would turn clipping into a compet"],"forward_implications":["The same KiDS-1000 data, analysed with the combined clipped and unclipped probes, yields $S_8$ and $w_0$ constraints 16% and 24% tighter than the unclipped analysis alone, with the $\\Omega_{\\rm m}$–$S_8$ figure of merit improved by a factor of 1.2.","The emulator-based unclipped analysis independently reproduces the Asgari et al. (2021) cosmology — $\\Omega_{\\rm m}$ within 0.8$\\sigma$ and $S_8$ within 1.8$\\sigma$ — validating the standard hmcode-based KiDS-1000 pipeline from a completely different modelling direction.","Clipping breaks the degeneracy between intrinsic alignments and cosmic shear: $A_{\\rm IA}$ is measured at 39% precision (a 27% improvement over the unclipped probe), and a first upper limit $b_{\\rm bary} < 0.97$ is placed on baryonic feedback from lensing alone, where the unclipped probe leaves it unconstrained.","Because the precision gains come from larger, less baryon-contaminated scales rather than from the small scales targeted by other higher-order statistics, the method ports directly to future data sets (KiDS-Legacy, Euclid, LSST) once tailored simulations exist."],"supporting_citations":[{"why":"Developed the clipping transform on KiDS-450 lensing data, set the threshold and smoothing scale reused here, and identified the masking bias that motivates footprint-infused mocks.","marker":"Giblin et al. (2018, G18)"},{"why":"The reference KiDS-1000 cosmic shear analysis whose scale cuts and redshift priors this work adopts and whose cosmology the emulator-based pipeline independently validates.","marker":"Asgari et al. (2021, A21)"},{"why":"Presents the cosmoSLICS dark-matter-only suite of 26 wCDM cosmologies whose clipped and unclipped correlation functions train the cosmological emulators.","marker":"Harnois-Déraps et al. (2019)"},{"why":"Presents the SLICS suite of independent fiducial-cosmology realisations from which the combined data covariance is estimated.","marker":"Harnois-Déraps et al. (2018)"},{"why":"Produced the intrinsic-alignment simulation set used to compute the $A_{\\rm IA}$-dependent biases to the clipped and unclipped statistics via the difference between $A_{\\rm IA}\\ne0$ and $A_{\\rm IA}=0$ mocks.","marker":"Harnois-Déraps, Martinet & Reischke (2022)"},{"why":"The magneticum hydrodynamical simulations provide the baryonic-feedback-contaminated mocks whose difference from dark-matter-only runs sets the $b_{\\rm bary}$ bias model.","marker":"Castro et al. (2021)"},{"why":"Supplies the photometric-redshift calibration whose per-bin biases and covariance define the $\\Delta z$ shift priors for the photo-$z$ nuisance parameters.","marker":"Hildebrandt et al. (2021)"},{"why":"Provides the correction factor applied to the simulated inverse covariance to remove bias in the Gaussian likelihood.","marker":"Hartlap et al. (2007)"},{"why":"The mass-mapping inversion used to reconstruct the convergence maps from smoothed ellipticity maps, the field that clipping operates on.","marker":"Kaiser & Squires (1993)"}],"fun_headline_variants":["Clipped shear maps sharpen KiDS-1000 cosmology by up to 24%","Removing dense lensing peaks lifts KiDS-1000 cosmic constraints","KiDS-1000: Clipped lensing tightens S8 16%, w0 24%","Clipping the densest sky regions reveals extra cosmic information"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The analysis assumes that the contamination from galaxies' intrinsic shape alignments and the cosmological lensing signal are separable in the clipped statistics, and that the survey's masked sky pattern does not change how that contamination enters the measurement — an assumption made because the intrinsic-alignment simulations are run on full, unmasked lightcones.","fun_headline_variants_meta":{"raw":{"variants":["Clipped shear maps sharpen KiDS-1000 cosmology by up to 24%","Removing dense lensing peaks lifts KiDS-1000 cosmic constraints","KiDS-1000: Clipped lensing tightens S8 16%, w0 24%","Clipping the densest sky regions reveals extra cosmic information"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000618,"raw_usage":{"total_tokens":3013,"prompt_tokens":1234,"completion_tokens":1779,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":850,"completion_tokens_details":{"reasoning_tokens":1704}},"tokens_in":850,"tokens_out":1779,"duration_ms":10459,"temperature":1.0,"reasoning_tokens":1704,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:27:36.230426+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Build intrinsic-alignment mocks that carry the same 18-tile masked KiDS-1000 footprint as the cosmology set and recompute the intrinsic-alignment bias in the clipped correlation functions; if the masked and unmasked biases differ by more than the emulator-plus-statistical error budget, the separability assumption fails and the reported $A_{\\rm IA}$, which already sits $3.2\\sigma$ from the Asgari et al. (2021) result, would flag a biased model. A complementary check is to run the identical combined pipeline on KiDS-Legacy or DES Year 6 data and ask whether the same $S_8$ and $A_{\\rm IA}$ are recovered.","supporting_citations":[],"review_version":2}