{"id":"b137f210-815c-4486-94a0-59e2df7c25db","arxiv_id":"1908.09856","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A reanalysis of Planck HFI polarization with the improved SRoll2 map-making yields tau = 0.0566 (+0.0053 / -0.0062) from large-scale EE data alone, and tau = 0.059 ± 0.006 in combination with Planck temperature and high-l polarization, the tightest constraint to date.","lead":"This paper uses a new processing of Planck satellite data to measure the thickness of the early universe's fog, called the reionization optical depth, with 10 percent accuracy. Generalists should care because this number controls how much cosmic structure we infer and is the least well measured of the standard cosmological parameters.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The error budget rests on self-generated SRoll2 simulations that validate the same ADCNL model they are built from; the persistent TE null anomaly suggests this model may be incomplete, so the quoted tau uncertainty could be underestimated.","rationale":"I agree with the reader's conditional verdict. The paper is a careful reanalysis with publicly released data and likelihood code, and the result is supported by many internal consistency tests, a TE-only cross-check, a null BB test, and consistency with the independent WMAP+353 GHz tau estimate. The shift toward that independent value gives some external support to the central claim. However, the quantitative error budget rests on N+S+F-MC simulations generated by the same SRoll2 code and ADCNL model that the analysis is designed to validate; this is a real circularity in the systematics assessment. The persistent TE null anomaly is a concrete sign of model incompleteness, and while the paper tests stability across masks and foreground templates, those tests do not break the circularity. The half-mission cross-validation I propose would directly test whether the ADCNL model is overfit or incomplete, and whether the simulation-based uncertainties capture the true scatter. This concern does not overturn the central claim, but it justifies conditional acceptance rather than full acceptance, exactly as the reader concluded.","tokens_in":17937,"tokens_out":7991,"duration_ms":88072,"concrete_test":"Perform a half-mission cross-validation of the SRoll2 ADCNL correction: fit the bolometer ADCNL splines using the first half of the mission timelines and apply them to the second half, and vice versa. Compute the 100x143 EE spectrum and tau posterior for each half with the same likelihood. If the two tau values disagree by more than the expected scatter (approximately sqrt(2) times the quoted 0.006 error) or if the l=2-8 EE residuals exceed the N+S+F-MC noise level, the ADCNL model is not stable and the simulation-based error bars miss a real systematic.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim of tau = 0.059 ± 0.006 with 10% accuracy depends on the assertion that SRoll2 reduces the second-order ADCNL dipole-leakage systematic below the noise level. This assertion is validated in Figures 1-3 and Section 2 using simulated timelines that contain exactly the ADCNL model fitted by SRoll2 (Delouis et al. 2019) and are then processed with the same code. This is a consistency check: it shows the algorithm removes its own assumed systematic, but it does not demonstrate that the fitted ADCNL spline is complete or that the simulated sky and noise input are realistic. The N+S+F-MC simulations, used to build the empirical likelihood and error bars in Section 4, inherit the same assumed model. The paper itself notes that the poor PTEs of the null TE spectra persist in this data version, albeit with slightly less significance; if the TE anomaly arises from residual polarization systematics rather than from the temperature map, the EE-based tau could be biased by an amount not captured in the N+S+F-MC scatter. The statement in Section 6 that the accuracy of the ADCNL simulation is now less critical is predicated on those same simulations and is partly circular. The consistency tests across masks, multipole cuts, and foreground templates are valuable, but they test internal stability of the pipeline, not the fidelity of the systematics model against real instrument behavior. If the ADCNL model is incomplete, the ~0.008 upward shift in the central value relative to Planck 2018 could be only partially corrected, and the quoted 0.006 error would be underestimated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents a new low-multipole polarization analysis of Planck HFI data using the SRoll2 map-making algorithm, which is designed to remove the second-order ADC nonlinearity (ADCNL) systematic that produces temperature-to-polarization dipole leakage. The authors build a 100x143 GHz EE cross-spectrum likelihood from an empirical distribution of signal plus N+S+F Monte Carlo simulations, and measure tau = 0.0566 (+0.0053, -0.0062) at 68% CL from the EE spectrum alone, with the external normalization 10^9 A_s e^{-2 tau} = 1.875. Combined with the Planck 2018 temperature and high-multipole polarization likelihoods, they report tau = 0.059 +/- 0.006, corresponding to z_re = 8.14 +/- 0.61 and sigma_8 = 0.8128 +/- 0.0053, and claim this is the strongest tau constraint to date with ten percent accuracy. The analysis includes extensive stability tests over sky fractions, multipole ranges, foreground templates, and null spectra.","tokens_in":18235,"tokens_out":9759,"duration_ms":103670,"significance":"If the systematics model is realistic, this is an important result: it reduces the Planck 2018 legacy-release uncertainty on tau by about 40% and breaks the A_s-e^{-2 tau} degeneracy more effectively, yielding a tight sigma_8 constraint. The paper is careful in its internal consistency checks: multiple masks, multipole cuts, alternative foreground tracers, and a large Monte Carlo suite are used, and the maps, simulations, and likelihood are made public. The main limitation is that the error budget and the claim that the ADCNL systematic is below noise are validated with simulations generated from the same ADCNL model that SRoll2 fits, so the quoted 10% uncertainty is conditional on the completeness of that model.","major_comments":[{"comment":"The central claim that SRoll2 brings the second-order ADCNL dipole-leakage systematic below the noise level is established with simulated timelines that contain exactly the ADCNL model fitted by SRoll2 and are then processed with SRoll2 itself. This is a self-consistency test, not an independent validation of the completeness of the ADCNL model or the realism of the simulated instrument noise. The empirical likelihood and the quoted 68% errors in Eq. (3) are built from N+S+F-MC simulations generated with the same model, so the error budget inherits the same assumption; the data-based PTE tests in Table 2 are also computed against the same simulations. The persistent poor PTEs of the null TE spectra reported in Section 4 are a data-side indication that the systematics model may not be complete. I therefore ask for a quantitative robustness test, for example rerunning the tau likelihood with residual maps from a deliberately different ADCNL model (varied spline flexibility or amplitude, or SRoll1 maps) and reporting the induced shift in tau, or adding an explicit model-error contribution to the error budget. Without such a test, the headline 10% error bar rests on a single assumed systematics model.","section":"§2, Figs. 1–3; §4 (N+S+F-MC)"},{"comment":"The TE-only measurement tau = 0.057 (+0.012, -0.013) is presented as confirmation of the EE result, but the same paragraph states that the poor PTEs of the null TE spectra persist in this data version. Because the TE analysis uses the Commander temperature map based on SRoll1, the TE anomaly may not directly affect the EE pipeline, but the claim of an independent confirmation is weakened. Please quantify the sensitivity of the headline tau to the multipoles or angular scales most affected by the TE anomaly, for example by truncating the EE likelihood at l_min = 4 or 6, or by masking the regions where the TE null PTE is lowest, and state explicitly whether the anomaly is attributed to the temperature map or to residual polarization systematics, with a supporting test.","section":"§4, TE-only result"}],"minor_comments":[{"comment":"The beta column header appears inconsistent with the entries: with the label beta x 10^2, the 100 GHz entry 1.86 gives beta = 0.0186, while the 143 GHz entry 0.0394 would give beta = 3.94 x 10^-4, contradicting the text's statement that the dust tracer is scaled by beta = 0.039. Please clarify the units or correct the typo.","section":"Table 1"},{"comment":"The empirical likelihood is built from 500 simulations per tau value with piecewise-polynomial interpolation; the paper does not report how sensitive the final tau interval is to the number of simulations or to the interpolation order. A brief convergence statement would strengthen the result.","section":"§4 and Eq. (2)"},{"comment":"The correlation-matrix values in Figure 9 are very hard to read in the printed version; a color map with labeled axes would be much clearer and would make the claimed near-diagonality more immediately apparent.","section":"Figure 9"},{"comment":"The standalone EE value tau = 0.0566 is conditional on the fixed external normalization 10^9 A_s e^{-2 tau} = 1.875; this is disclosed in Section 4 but should be stated explicitly in the abstract or in the sentence quoting the standalone value, to avoid the impression that it comes from polarization data alone.","section":"Abstract and §4"}],"recommendation":"major_revision","confidential_remarks":"The paper is technically careful and the public release of maps, simulations, and the likelihood is a genuine strength. My main hesitation is that the systematics validation is generated with the same SRoll2 code and ADCNL model being tested, so the 10% error claim is conditional on that model being complete. I would like to see a model-perturbation test or an explicit model-error term before endorsing the headline uncertainty as robust. This is a major-revision rather than a reject situation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read it. This is the strongest CMB-only tau measurement we have, and it is not a fluke: the paper is careful, the maps and code are public, and the consistency tests are unusually thorough. The genuinely new piece is the SRoll2 map-making, which pushes the second-order ADCNL dipole leakage below the noise level and recovers the quadrupole and octupole that Planck 2018 had to suppress. That gives tau = 0.0566 (+0.0053/-0.0062) from EE alone, and tau = 0.059 ± 0.006 combined, with sigma_8 at 0.5% precision. I think the result is likely true.\n\nWhat the paper does well: the mask ladder, multipole-cut tests, alternate dust and synchrotron tracers, jackknife-style multipole removal, and 500 N+S+F-MC simulations. They also openly report that the null-TE anomaly persists, and the TE-only tau (0.057 ± 0.012) is consistent, which strengthens the EE result. The public release of maps and likelihood is a real plus.\n\nThe soft spot is exactly where the stress-test note lands: the error bar is built from N+S+F-MC simulations that use the same ADCNL model that SRoll2 corrects. That is a consistency check, not an independent validation of the model. If the ADCNL spline is incomplete or the simulated sky input is missing a real systematic, the tau shift could be larger than the scatter suggests. The paper argues that the accuracy of the ADCNL simulation is now less critical because residuals are below noise and cosmic variance dominates, but that statement itself is validated on those same simulations, so it is partly circular. The persistent TE anomaly is a hint that the systematics model is not perfect; it may be in temperature or geometry rather than polarization, but I would want that discussed more explicitly. The foreground template scaling uncertainties are not propagated into the final intervals, although the cross-checks suggest they are stable. These are caveats, not fatal flaws. The central claim is credible, and the upward shift relative to Planck 2018 is about one sigma, so I would not treat it as a tension.\n\nFor a referee: yes, this deserves a careful review. The main request I'd make is to present the N+S+F simulations as model-internal validation, and to quantify how much the tau error would grow if the TE anomaly were treated as a residual polarization systematic. As it stands, the quoted 0.006 is a best-case uncertainty. I'd cite this and I'd take the measurement seriously, but I would pair it with an independent check before treating the central value as final.","headline":"A careful, credible reanalysis that likely gives the tightest CMB tau so far, but the quoted error bar is set by the same systematics model the mapmaker fits, so treat 0.006 as an optimistic floor.","tokens_in":18868,"tokens_out":2894,"would_cite":true,"duration_ms":29710,"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":"Re-analyzing Planck HFI data with an upgraded mapmaker, this paper measures the reionization optical depth as $\\tau = 0.059 \\pm 0.006$ (68% C.L.), a 10% measurement it argues is the strongest constraint to date.","keywords":["reionization optical depth","cosmic microwave background polarization","Planck HFI","SRoll2 map-making","ADC non-linearity","large-scale polarization","cosmological parameters","reionization redshift"],"falsifier":"Re-run the SRoll2 pipeline on simulated timelines with a known input $\\tau = 0.051$ while injecting ADCNL residuals at twice the amplitude used in the published simulation set; if the recovered $\\tau$ moves upward by the roughly 0.008 shift seen between Planck 2018 and SRoll2, the systematic estimate is under-correcting. Equivalently, an independent map-making code that does not use the SRoll2 spline model, applied to the same raw data, should reproduce the same EE quadrupole and octupole within the reported noise.","tokens_in":17695,"feed_emoji":"🌌","tokens_out":9322,"duration_ms":82046,"temperature":0.7,"pith_summary":"This paper argues that the residual large-scale systematic contamination in the Planck High Frequency Instrument (HFI) polarization maps, left incompletely removed in the 2018 legacy release, can be pushed below the noise level with an upgraded map-making algorithm called SRoll2. The authors measure the reionization optical depth from the cleaned 100$\\times$143 GHz EE cross-spectrum as $\\tau = 0.0566^{+0.0053}_{-0.0062}$ (68% C.L.), and, within the $\\Lambda$CDM model combined with the Planck temperature and high-$\\ell$ polarization likelihoods, as $\\tau = 0.059 \\pm 0.006$, corresponding to a mid-point reionization redshift $z_{\\rm re} = 8.14 \\pm 0.61$. Since the CMB sees the primordial fluctuation amplitude only through $A_s e^{-2\\tau}$, a 10% measurement of $\\tau$ gives about 1% accuracy on $A_s$ and breaks the main degeneracy that has made $\\tau$ the least constrained $\\Lambda$CDM parameter. The result matters because reionization is the last major phase transition of the universe and its timing is a direct probe of early galaxy formation.","feed_headline":"Reprocessed Planck data pin tau = 0.059 ± 0.006","feed_subtitle":"A new mapmaker removes the large-scale leakage that limited earlier Planck estimates, giving a 10% measurement of tau.","key_machinery":"The load-bearing element is SRoll2, an upgraded version of the SRoll map-making code described in the companion paper. It corrects the second-order ADC non-linearity by fitting a two-dimensional spline per bolometer as a function of signal value and time, so that a single gain fits the entire mission and the temperature-to-polarization dipole leakage is removed. The analysis proceeds with the 100$\\times$143 GHz EE cross-spectrum estimated with a quadratic maximum-likelihood estimator (QML) on a 50% sky mask, and a simulation-based likelihood, lowE-S2, constructed from 500 noise-plus-systematics-plus-foreground-residual Monte Carlo realizations, interpolated over a grid in $\\tau$ with $10^9 A_s e^{-2\\tau} = 1.875$ fixed.","core_discovery":"On the paper's own terms, the central discovery is that the second-order analogue-to-digital converter non-linearity in the Planck HFI readout chains, when modelled with a per-bolometer spline fit, removes the dipole-to-polarization leakage that dominated the lowest multipoles of the 2018 legacy maps; the cleaned 100$\\times$143 GHz EE spectrum then gives $\\tau = 0.0566^{+0.0053}_{-0.0062}$ from polarization alone. Combining the new lowE-S2 likelihood with Planck's low-$\\ell$ Commander temperature likelihood and high-$\\ell$ temperature and polarization likelihood gives $\\tau = 0.059 \\pm 0.006$ (68% C.L.), $z_{\\rm re} = 8.14 \\pm 0.61$, and $\\sigma_8 = 0.8128 \\pm 0.0053$. The paper states this is the strongest reionization optical depth constraint to date, with the uncertainty reduced by roughly 40% relative to the Planck 2018 legacy value $\\tau = 0.051 \\pm 0.009$; the central value shifts upward by about one $\\sigma$ while the 95% upper limit stays near $\\tau \\lesssim 0.07$.","pith_inferences":["If the systematics estimates are right, an independent re-analysis of the same raw timelines with a map-making code that does not assume the SRoll2 spline model should reproduce the recovered quadrupole and octupole in EE within the reported errors; agreement would effectively close the map-making chapter of Planck HFI large-scale polarization.","The same second-order ADC non-linearity correction strategy could be applied to other bolometric CMB instruments with similar readout chains, potentially cleaning their large-scale polarization before launch or in re-processing.","The paper notes a semi-analytical likelihood is now feasible; because cosmic variance dominates the error budget, that likelihood should give nearly identical $\\tau$ posteriors, and disagreement would flag a problem in the simulation-based estimate.","If the true $\\tau$ is near 0.059 rather than the earlier 0.051, reionization models need somewhat more ionizing photons around $z \\sim 8$; future high-redshift observations should see a correspondingly earlier or more efficient reionization history."],"forward_implications":["The $\\tau$ uncertainty drops to about 10%, which propagates to roughly 1% accuracy on the primordial fluctuation amplitude $A_s$ via the $A_s e^{-2\\tau}$ degeneracy.","The mid-point reionization redshift $z_{\\rm re} = 8.14 \\pm 0.61$ gives a direct CMB-side target for astrophysical models of early galaxy formation and for high-redshift galaxy surveys.","Replacing the Planck 2018 lowE likelihood with lowE-S2 leaves the other $\\Lambda$CDM parameters essentially unchanged, while tightening $\\sigma_8$ to $0.8128 \\pm 0.0053$ and slightly lowering the lensing amplitude parameter $A_L$ to $1.163 \\pm 0.064$.","Systematics no longer dominate the low-$\\ell$ EE error budget: cosmic variance does, so further tightening from the same sky requires either more sky coverage or a different observable.","The $\\tau$ upper limit remains close to $0.07$ at 95% C.L., so the minimal one-parameter extensions of $\\Lambda$CDM explored in the paper are not significantly affected."],"supporting_citations":[{"why":"Introduces the SRoll2 map-making algorithm and supplies the N+S+F-MC simulations used for error estimation and validation.","marker":"Delouis et al. (2019)"},{"why":"Provides the QML and likelihood methodology, the 2018 legacy $\\tau = 0.051 \\pm 0.009$ baseline, and the Commander low-$\\ell$ temperature likelihood that the new analysis builds on.","marker":"Planck Collaboration V (2019)"},{"why":"Establishes the simulation-based low-$\\ell$ EE likelihood approach and the previous SRoll1 analysis that SRoll2 extends.","marker":"Planck Collaboration Int. XLVI (2016)"},{"why":"Documents the first-order ADCNL correction and residual leakage in the HFI 2018 legacy maps that SRoll2 aims to remove.","marker":"Planck Collaboration III (2019)"},{"why":"Supplies the FFP8 full-focal-plane simulations and noise covariance matrices used in the power spectrum and likelihood computation.","marker":"Planck Collaboration XII (2016)"},{"why":"Sets the $\\Lambda$CDM parameter framework, best-fit values, and comparison Planck 2018 results used throughout the cosmological analysis.","marker":"Planck Collaboration VI (2019)"},{"why":"Provides the CMB lensing likelihood that, combined with BAO, is used to check that the $\\tau$ constraint does not shift in the full parameter exploration.","marker":"Planck Collaboration VIII (2019)"},{"why":"Underlies the construction of the residual noise covariance matrices used by the estimator.","marker":"Keskitalo et al. (2010)"},{"why":"Identified as the validation reference for the simulation-based likelihood approximation used in lowE-S2.","marker":"Gerbino et al. (2019)"}],"fun_headline_variants":["Planck reanalysis cuts tau error to 10%","New Planck mapmaker reduces tau uncertainty by 40%","Reprocessed Planck data yield tau = 0.059 ± 0.006","Planck's improved mapmaking pins tau with 10% accuracy","Planck HFI fix delivers tau = 0.059 ± 0.006"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The result assumes that the N+S+F-MC simulations, built with the same SRoll2 code and a simulated sky model, reproduce the real instrument's residual systematics faithfully; if the ADCNL model or the input sky is incomplete, the quoted error bars and the upward shift in $\\tau$ would be underestimated.","fun_headline_variants_meta":{"raw":{"variants":["Planck reanalysis cuts tau error to 10%","New Planck mapmaker reduces tau uncertainty by 40%","Reprocessed Planck data yield tau = 0.059 ± 0.006","Planck's improved mapmaking pins tau with 10% accuracy","Planck HFI fix delivers tau = 0.059 ± 0.006"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000395,"raw_usage":{"total_tokens":2117,"prompt_tokens":1038,"completion_tokens":1079,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":654,"completion_tokens_details":{"reasoning_tokens":983}},"tokens_in":654,"tokens_out":1079,"duration_ms":10467,"temperature":1.0,"reasoning_tokens":983,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:00:18.351966+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the SRoll2 pipeline on simulated timelines with a known input $\\tau = 0.051$ while injecting ADCNL residuals at twice the amplitude used in the published simulation set; if the recovered $\\tau$ moves upward by the roughly 0.008 shift seen between Planck 2018 and SRoll2, the systematic estimate is under-correcting. Equivalently, an independent map-making code that does not use the SRoll2 spline model, applied to the same raw data, should reproduce the same EE quadrupole and octupole within the reported noise.","supporting_citations":[{"cited_title":"Residual noise covariance for Planck low-resolution data analysis","cited_arxiv_id":"0906.0175","evidence_quote":"Underlies the construction of the residual noise covariance matrices used by the estimator."}],"review_version":1}