{"id":"460958ca-ca82-4067-9d83-5191d3a037b8","arxiv_id":"2508.21069","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Reionization is inferred to be rapid and late (midpoint z≈7, duration Δz50≈1.1), yielding an optical depth τ=0.0492 from Lyman-alpha + BAO + BBN, independent of CMB data.","lead":"This paper uses hydrogen gas measurements from the era of reionization to show the universe became ionized late and quickly, around redshift 7, with a matching low CMB scattering depth. It sharpens the ongoing tension between different cosmological datasets and strengthens hints that dark energy may not be constant.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Baseline low τ from Eq. (5) is set by the zmax=15 GP boundary; the zmax=30 test replaces it with a kSZ prior calibrated to monotonic simulations, so the claim that xHI data alone imply rapid late reionization is not yet secured.","rationale":"The paper is transparent and technically strong: it releases its likelihood code, applies a flat-τ prior correction, compares parametric and non-parametric reconstructions, and explicitly flags the zmax=15 assumption. The reader's conditional verdict is appropriate, and my stress test targets the same load-bearing premise: the high-redshift tail of xHI(z) is unobserved, and the baseline low-τ result is enforced by the zmax=15 boundary. The zmax=30 robustness test is weakened by the fact that the added kSZ prior is built from a monotonic-reionization fitting formula (Eq. 8) calibrated to amber simulations. Since the paper admits that non-monotonic and double-reionization histories are not reliably constrained by this prior, the extended test does not close the window it is meant to close. A direct computation with zmax=30 and no kSZ prior would settle whether the low-τ result is actually data-driven or prior-dominated. The proposed test is feasible because the public likelihood code can be rerun with modified settings. I do not see a reason to reject the paper or even to move it from CONDITIONAL; the concern is a concrete robustness check that the authors should add before the headline constraint is treated as final.","tokens_in":25351,"tokens_out":7798,"duration_ms":87951,"concrete_test":"Re-run the released reionlik code with the baseline Table 1 data, BBN+DESI BAO, the same GP wrapping function and hyperparameters, and flat-τ reweighting, but set zmax=30 and omit the kSZ prior. This isolates the effect of the boundary condition. If the resulting τ posterior shifts by more than ~2σ from Eq. (5) or its upper tail exceeds ~0.07, the rapid-late-reionization claim depends on the zmax=15 boundary rather than on the xHI data alone.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that current xHI(z) data plus BAO/BBN imply τ=0.0492 (Eq. 5) and exclude τ~0.09 is not fully data-driven. In Sec. 3.1, reionization is forced to begin only at zmax=15 by the boundary condition xHI(zmax)=1−ε; the data in Table 1 extend only to z≈11.5. The z>12 contribution to τ is therefore controlled by the GP prior and the boundary, not by observations. The paper's only extended test (Sec. 3.3) moves zmax to 30 but simultaneously imposes a kSZ prior on (Δz90, zmid) via Eq. (8). That relation is calibrated to the amber simulations for monotonic reionization histories; the paper itself states that non-monotonic/double-reionization histories produce larger kSZ signals and are not reliably constrained ('more work is needed'). Thus the zmax=30 exercise does not demonstrate robustness to early or non-monotonic reionization; it substitutes one model-dependent prior for another. In addition, the z=8–11.5 points (Umeda et al. 2025; Curtis-Lake et al. 2023) are JWST-era damping-wing measurements whose mapping to the volume-averaged neutral fraction carries systematic assumptions; if those points are biased toward high xHI, τ would be biased low.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper derives new constraints on the reionization optical depth, tau, using several independent routes. The primary result is a non-parametric Gaussian-process reconstruction of the volume-averaged neutral hydrogen fraction, xHI(z), from a compilation of Lyman-alpha and damping-wing constraints (Table 1). Combined with a BBN prior on Omega_b h^2 and DESI BAO data, the reconstruction yields a rapid and late reionization history, z_mid = 7.00(+0.12,-0.18), Delta z50 = 1.12(+0.12,-0.29), and tau = 0.0492(+0.0014,-0.0030) (Eq. 5). The paper also compares this astrophysical bound with tau constraints from large-scale CMB polarization (three likelihood implementations) and from small-scale CMB data combined with lensing, BAO, and SNe. In LambdaCDM, small-scale CMB + BAO + lensing gives tau = 0.094 +/- 0.011, in ~3.7 sigma tension with the Ly-alpha bound. The paper then explores extended cosmological models (running spectral index, A_lens, curvature, N_eff, effective neutrino mass, wCDM, w0waCDM) and reports that the tension is reduced or removed in some of them, with w0waCDM and an unphysical negative neutrino-mass parameter restoring concordance.","tokens_in":25726,"tokens_out":9053,"duration_ms":95257,"significance":"If the central low-tau result is robust, it has considerable significance: it sharpens the well-known tension between DESI BAO combined with small-scale CMB+lensing and the standard late-reionization picture, and it strengthens the case for dynamical dark energy or other beyond-LambdaCDM explanations. The paper is careful in several respects: it uses a flat-tau prior correction, includes redshift nuisance parameters for each xHI point, compares three different large-scale polarization likelihoods, and publicly releases the likelihood code. These are concrete strengths. The main caveat is that the central claim is not as data-driven as the abstract suggests: the high-redshift contribution to tau is strongly influenced by the adopted GP boundary and hyperparameters, and the only extended test (zmax=30) relies on a kSZ prior calibrated to monotonic reionization simulations. The paper itself acknowledges that non-monotonic histories are not reliably constrained. Properly qualified, the result is credible and interesting; in its current abstract form it is somewhat overclaimed.","major_comments":[{"comment":"The central claim that xHI data plus BAO/BBN imply tau = 0.0492 and rule out tau ~ 0.09 is not fully data-driven at high redshift. Table 1 contains no data beyond z = 11.49; the baseline sets xHI(zmax=15) = 1 - epsilon, so the z > 15 contribution is zero and the z ~ 12-15 contribution is governed by the GP prior and boundary. The zmax = 30 test replaces this boundary with a kSZ prior from Eq. (8) that is calibrated to monotonic amber simulations. As the paper states, non-monotonic/double-reionization histories produce larger kSZ signals and are not reliably constrained ('more work is needed'). Hence the abstract's wording 'independent of CMB data' and the exclusion of tau ~ 0.09 require qualification: the result is conditional on a chosen prior family and a monotonic kSZ calibration. A sensitivity analysis varying zmax and the kSZ calibration, or a conservative early-reionization templat","section":"Sec. 3.1, 3.3; Eq. (5), (8), (9)"},{"comment":"The GP hyperparameters are fixed by hand (sigma = 0.3, mu = 0, inverse-gamma p=3, b=1.5, beta=8). Because the z > 10 constraints are sparse, the posterior predictive there is dominated by these choices. The paper does not test variations of sigma or mu; the quoted error bars for Eq. (5) therefore include only data and fixed-prior uncertainty, not prior uncertainty. At a minimum, report results for e.g. sigma in {0.15, 0.6} and mu in {-0.5, 0.5} (in latent space) or an explicit check that the high-z tail is insensitive to these choices.","section":"Sec. 3.1, Appendix A"},{"comment":"The claim that adding xHI data increases the preference for w0waCDM from 4.2 to 4.5 sigma inherits the same prior-dependence as the baseline tau result. The xHI likelihood effectively replaces lowE in the combination, but the high-z part of that likelihood is prior-dominated. The model-comparison conclusion should therefore be presented with the same caveat as the tau constraint; otherwise readers may take the 4.5 sigma as a data-driven result independent of the GP boundary assumptions.","section":"Sec. 4.2.6, Eq. (15)"}],"minor_comments":[{"comment":"The sentence 'For z > 12, we reasonably assume that xHI(z) ≈ 0 with an uncertainty of sigma = 0.3' appears inconsistent with the boundary condition xHI(zmax) = 1 - epsilon and with high-z neutrality. Should this read xHI(z) ≈ 1, or is the mean mu = 0 in the latent space mapped to near unity by the wrapping function? Please clarify.","section":"Sec. 3.1"},{"comment":"The wrapping function uses sign(2x - 1) where x is described as an unbounded latent variable. This expression is unusual for a map from R to (0,1); please define the argument more carefully and verify the value of F(0) used in the prior mean.","section":"Appendix A, Eq. (A1)"},{"comment":"The abstract states that the baseline tau constraint is 'independent of CMB data.' This is true for Eq. (5) itself, but the robustness test in Sec. 3.3 explicitly adds a kSZ prior derived from SPT CMB observations. The wording should be qualified to avoid the impression that all stated robustness checks are CMB-free.","section":"Abstract and Sec. 3.3"},{"comment":"The text describes Sigma m_nu,eff = -0.120(+0.034,-0.039) eV as 'the strongest cosmological bound on the neutrino mass.' Since this parameter is an unphysical extension (as the paper itself notes), the wording should be changed to 'strongest constraint on the effective negative-mass parameter' to avoid confusion with the physical sum of neutrino masses.","section":"Sec. 4.2.5, Eq. (13)"}],"recommendation":"major_revision","confidential_remarks":"The paper is well-executed and the central low-tau result is plausible, but the abstract's strong claim is not yet fully supported by the robustness analysis. The zmax=30 test relies on a monotonic kSZ calibration that the paper itself admits does not reliably cover non-monotonic histories. I recommend major revision rather than rejection, with the expectation that a sensitivity analysis and a more cautious abstract will make the paper publishable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline result is a tau of about 0.049 from a non-parametric xHI(z) reconstruction combined with DESI BAO and BBN, deliberately excluding CMB data. That is a genuinely useful new datapoint for the reionization debate, and the paper does a lot right: it compares three low-ell polarization likelihoods, includes a flat-tau prior correction, treats redshift uncertainties with nuisance parameters, releases the likelihood code, and is unusually explicit about its own assumptions. The tension numbers (3.7-4.3 sigma in T, 4.5 sigma for w0waCDM) are quantified with public tools, and the qualitative conclusion that reionization was late and rapid is consistent with Planck lowE and several prior reconstructions. Credit is due for that.\n\nThe soft spot is exactly where the stress-test puts it. The baseline tau sits on top of the zmax=15 boundary condition, and the data stop around z~11.5. The paper's own zmax=30 test does not close this gap with xHI data alone; it imports a kSZ prior calibrated to monotonic reionization simulations. The paper says double-reionization histories produce larger kSZ signals and that reliable constraints from such models need more work. So the claim that BAO+BBN+xHI data alone imply tau=0.049 and exclude tau~0.09 is too strong. What the data do, conditional on a GP prior that starts reionization after z=15 and a kSZ prior anchored to monotonic histories, is prefer rapid late reionization. That is a meaningful result, but it is not a pure data measurement.\n\nThere is also a smaller issue with the extended-model section: the effective negative neutrino mass parameter is taken from the author's own work, and the paper itself notes it absorbs tension into an unphysical parameter. That is fine as an exploration, but the 4.5 sigma dark energy preference should be read with that in mind. A full systematic error budget on tau, including leave-one-out tests on the z>8 JWST points and marginalization over GP hyperparameters, would make the central constraint more robust.\n\nBottom line: the paper is a solid, honest constraints analysis that deserves a serious referee. The central tau value is probably close to right, but the headline phrasing overstates the independence from priors. I would recommend sending it to review with a request for the robustness tests and a softer abstract.","headline":"A careful, transparent tau constraints paper that delivers a credible low optical depth from Lyman-alpha data plus BAO/BBN, but the 'CMB-independent' wording oversells it because the GP prior boundary does the heavy lifting at high z.","tokens_in":787,"tokens_out":723,"would_cite":true,"duration_ms":23626,"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":"Cosmic reionization is pinned as rapid and late, with optical depth τ = 0.049, independent of CMB data.","keywords":["reionization","optical depth","neutral hydrogen fraction","Lyman-alpha constraints","Gaussian process reconstruction","DESI BAO","CMB polarization","dark energy equation of state"],"falsifier":"A credible measurement of a non-negligible ionized fraction at z > 12 would break the central constraint — for example, a damping-wing or dark-pixel observation of a z > 12 source showing the neutral fraction clearly below 1, a 21-cm power spectrum placing the onset of reionization well before z ≈ 10, or evidence of substantial Population III ionization. The decisive confirmation runs the other way: a CMB polarization experiment with σ(τ) ≈ 0.002 measuring τ near 0.05 would vindicate the Lyman-α bound, while a value near 0.09 would show the bound is biased by the astrophysical priors.","tokens_in":25198,"feed_emoji":"🌌","tokens_out":11792,"duration_ms":102080,"temperature":0.7,"pith_summary":"The paper argues that the cosmic phase transition called reionization — when ultraviolet light from the first stars and galaxies stripped the universe's hydrogen of its electrons — was rapid and late, centred near redshift 7 and essentially finished by redshift 8. It derives an optical depth of τ = 0.0492 from Lyman-α measurements of the neutral hydrogen fraction combined with DESI baryon-acoustic-oscillation data and a Big Bang nucleosynthesis prior, with no cosmic microwave background input. That matters because τ ≈ 0.09 has been proposed as the way to keep the standard cosmological model while absorbing the discord between DESI and CMB measurements of the expansion history; the paper finds such high optical depths clash with the Lyman-α data at about 4σ. Independent routes — large-scale CMB polarization, and small-scale CMB data combined with supernovae or galaxy-lensing measurements — land on the same low value, while small-scale CMB combined with BAO only reaches τ ≈ 0.09 under plain ΛCDM. If the low value holds, the DESI-CMB discord survives and points to physics beyond ΛCDM: evolving dark energy, an excess lensing amplitude, or anomalous neutrino behaviour.","feed_headline":"CMB-free data pin cosmic reionization to redshift 7","feed_subtitle":"Lyman-α, DESI BAO and BBN rule out the τ≈0.09 fix for the DESI–CMB tension — at 4σ.","key_machinery":"The engine is a Gaussian-process reconstruction of the neutral fraction xHI(z): the history is drawn from a multivariate Gaussian with a squared-exponential kernel, mapped from unbounded latent variables onto the unit interval by a wrapping function, with endpoints fixed at xHI ≈ 0 for z = 5.2 and xHI ≈ 1 for z = 15 (chosen to allow gradual and early reionization). The optical depth follows from τ = nH c σT ∫ xe(z)(1+z)²/H(z) dz, where a BBN prior on the baryon density fixes nH and DESI BAO data fix H(z) — the ingredient that lets cosmology and ionization history be varied simultaneously with no CMB input. A derived parameter T ≡ τ √(Ωm h²/0.14) factors out most of the expansion-history depe","core_discovery":"The paper's central claim is that existing Lyman-α and Lyman-β constraints on the volume-averaged neutral hydrogen fraction xHI(z) — a compilation of 19 measurements that includes new JWST-era damping-wing results at z > 8 — force a Gaussian-process reconstruction of the reionization history to be fast and late. When those constraints are combined with DESI DR2 BAO data and a BBN prior on the baryon density, which supply the expansion history and hydrogen density needed to convert the neutral fraction into a Thomson optical depth, the reconstruction yields a midpoint z_mid = 7.00 (+0.12/−0.18), a duration Δz50 = 1.12 (+0.12/−0.29), and τ = 0.0492 (+0.0014/−0.0030). The measurement is deliber","pith_inferences":["If the low optical depth is right, the DESI-CMB tension becomes a sharper discriminator among extended cosmologies: the optical depth can be used as a nearly free, CMB-independent lever arm on dark energy and neutrino physics, since a high-τ model now has to jump the Lyman-α hurdle too.","The paper's 4.5σ preference for evolving dark energy inherits the assumption that damping-wing constraints are cosmology-independent; a reanalysis of the Table 1 posteriors within each extended model (which the paper flags as unquantified) could shift that preference, so an independent determination of the reionization midpoint from 21-cm or angular-diameter measurements would be a natural arbiter","The 'standard depth' trick — using τ to constrain the matter density when paired with an independent optical-depth probe — could become routine once a CMB polarization experiment reaches σ(τ) ≈ 0.002, yielding competitive matter-density constraints without BAO data.","A testable extension of the pipeline: apply the same Gaussian-process reconstruction to mock xHI(z) compilations drawn from double-reionization simulations reaching τ ≈ 0.09, to quantify how strongly the Lyman-α + kSZ combination actually penalizes each non-monotonic history class, since the kSZ calibration assumes monotonic reionization."],"forward_implications":["If reionization is this fast and late, the ionizing photon budget is dominated by sources at z ≈ 6–8, leaving little room for a long Population III tail or for significant high-redshift contributions to the optical depth.","The τ ≈ 0.09 scenario for reconciling DESI BAO with CMB data under ΛCDM is excluded at 3.7–4σ by the Lyman-α bound; the escape hatch of exotic early reionization is narrowed by kSZ constraints on reionization's midpoint and duration.","Under ΛCDM, small-scale CMB plus lensing plus BAO gives τ = 0.094 ± 0.011, a ~4σ clash with the astrophysical bound, so the discord must be absorbed by model extensions: Alens > 1 reduces the tension to 0.9σ, negative effective neutrino mass to 0.8σ, and evolving dark energy (w0waCDM) to 1.2σ.","Adding the xHI(z) data to the most powerful CMB + BAO + SNe combination raises the preference for dynamical dark energy from 4.2σ to 4.5σ, with the result immune to large-scale CMB polarization systematics by construction.","Large-scale CMB polarization, evaluated with three independent likelihoods, agrees with the Lyman-α bound and is insensitive to the cosmological model, providing a CMB-based cross-check that does not depend on astrophysical reionization tracers."],"supporting_citations":[{"why":"Supplies the framework for modelling xHI upper limits as one-sided Gaussians and the precedent for non-parametric reionization reconstruction.","marker":"Greig & Mesinger (2017)"},{"why":"The non-parametric reconstruction method this work adapts, and a direct comparison value (τ = 0.053 ± 0.004) the baseline result is checked against.","marker":"Mason et al. (2019b)"},{"why":"The new z > 8 neutral-fraction constraints that pinpoint the beginning of reionization and push τ downward.","marker":"Umeda et al. (2025)"},{"why":"Dark pixel fraction measurements at z ≈ 5.5–5.9 that anchor the end of reionization in the compiled dataset.","marker":"Jin et al. (2023)"},{"why":"The DR2 BAO measurements that break the degeneracy between τ and matter density, enabling the absolute optical-depth value.","marker":"DESI Collaboration et al. (2025a)"},{"why":"The BBN prior on the physical baryon density, which supplies the hydrogen density nH in the optical-depth integral.","marker":"Schöneberg (2024)"},{"why":"The kSZ trispectrum prior on reionization midpoint and duration used to constrain extended histories with zmax = 30.","marker":"Raghunathan et al. (2024)"},{"why":"The large-scale polarization (lowE) likelihood and the τ ≈ 0.05 reference value that the CMB comparison is built on.","marker":"Planck Collaboration et al. (2020b)"},{"why":"The proposal that τ ≈ 0.09 could reconcile DESI and CMB data, which the Lyman-α bound is tested against.","marker":"Sailer et al. (2025)"},{"why":"The approach of extracting τ from small-scale CMB data with low-redshift probes, and the motivation for scrutinizing lowE systematics.","marker":"Giarè et al. (2024a)"}],"fun_headline_variants":["Lyman-alpha+DESI BAO imply tau=0.05, fast reionization","Reionization midpoint z=7, duration dz~1, tau=0.049","4-sigma tension: Lyman-alpha tau=0.05 vs CMB+BAO tau=0.09","JWST-era Lyman-alpha data make reionization rapid and late","CMB-independent tau=0.049 from Lyman-alpha, BAO, BBN"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The load-bearing premise is that the 19 published measurements of the neutral hydrogen fraction — especially the JWST-era points at z > 8 — faithfully represent the cosmic average, and that the prior fixing reionization as essentially complete by z = 15 (with a kSZ prior standing in when the start is pushed to z = 30) correctly handles what is unobserved above z ≈ 12; a hidden early ionized fraction or a double-reionization episode there would raise τ substantially.","fun_headline_variants_meta":{"raw":{"variants":["Lyman-alpha+DESI BAO imply tau=0.05, fast reionization","Reionization midpoint z=7, duration dz~1, tau=0.049","4-sigma tension: Lyman-alpha tau=0.05 vs CMB+BAO tau=0.09","JWST-era Lyman-alpha data make reionization rapid and late","CMB-independent tau=0.049 from Lyman-alpha, BAO, BBN"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000491,"raw_usage":{"total_tokens":2373,"prompt_tokens":991,"completion_tokens":1382,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":735,"completion_tokens_details":{"reasoning_tokens":1263}},"tokens_in":735,"tokens_out":1382,"duration_ms":12866,"temperature":1.0,"reasoning_tokens":1263,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T14:34:53.731205+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A credible measurement of a non-negligible ionized fraction at z > 12 would break the central constraint — for example, a damping-wing or dark-pixel observation of a z > 12 source showing the neutral fraction clearly below 1, a 21-cm power spectrum placing the onset of reionization well before z ≈ 10, or evidence of substantial Population III ionization. The decisive confirmation runs the other way: a CMB polarization experiment with σ(τ) ≈ 0.002 measuring τ near 0.05 would vindicate the Lyman-α bound, while a value near 0.09 would show the bound is biased by the astrophysical priors.","supporting_citations":[],"review_version":1}