{"id":"75edfe5d-ce48-489e-9f68-3d2eb1e2c4cf","arxiv_id":"2412.15383","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"DESI Mg II absorbers produce average E(B-V) of about 0.04 for intervening systems and can bias quasar redshifts by up to about 0.005 at z > 1.5, with masking of the absorption lines partially recovering the bias.","lead":"Using 50,674 DESI quasar spectra with Mg II absorbers, this paper measures how much these absorbers redden the quasars and shows that associated absorbers can bias measured quasar redshifts by about 0.005 at high redshift. The work offers a pixel-masking fix that could improve quasar redshifts for DESI cosmology.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The masking claim lacks a same-version unmasked control and excludes the largest-redshift-shift systems, so the observed voff narrowing could be from Redrock template updates or sample trimming.","rationale":"The reader's weakest assumption points to possible bias in absorber redshifts. I agree that z_ALS validation is important, but the more directly load-bearing gap is internal to the masking experiment: the paper's own text says the unmasked v0.20.0 re-run was computed as a control, and yet no result from it is presented. That omission is decisive because the comparison actually shown (catalog versus masked v0.20.0) conflates masking with a template update. The post-hoc exclusion of systems that move out of the associated window compounds the problem: those are precisely the systems with the largest |delta z|, so removing them guarantees some narrowing of the voff distribution. I therefore do not think the central causal claim is established by the current evidence. This does not undermine the extinction measurements, which are a large-sample contribution with a matched control and a clear E(B-V) trend; nor does it call into question the authors' honesty. The fix is straightforward: show the unmasked re-run and apply identical cuts, so the appropriate action is to keep the conditional verdict rather than reject or accept. My choice of UNCHANGED reflects that the reader's conditional verdict already captures the need for validation.","tokens_in":17107,"tokens_out":6480,"duration_ms":57219,"concrete_test":"Re-run Redrock v0.20.0 on the 28,178 associated-absorber spectra in three modes: (i) unmasked, (ii) Mg II masked as in Section 3.3, and (iii) a placebo mask shifted by ±2000 km/s from the catalog line centers, with no post-hoc exclusion, applying any sample cuts identically. Plot voff distributions for all three plus the original catalog. If unmasked v0.20.0 already reproduces the narrowing seen in Figure 8, the masking claim fails; if the placebo mask narrows as much as the real mask, the effect is not specific to Mg II. Also report the distributions before excluding objects that leave voff<3500 km/s.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.3's central demonstration that masking Mg II pixels corrects QSO redshifts is not a controlled masking experiment. The plotted comparison is between the original catalog redshifts (Redrock v0.15.3/v0.17.0) and a masked re-run with Redrock v0.20.0, which uses QSO templates trained on a sample roughly 400 times larger (Brodzeller et al. 2023). The authors state they also re-ran v0.20.0 without masking 'to ensure we can separate the effect of our masking strategy from that of the new QSO templates,' but no unmasked v0.20.0 voff distribution is shown anywhere. The improvement in Figures 8 and 9 could therefore be caused by the template update rather than by masking. In addition, systems whose re-run redshift moved outside voff<3500 km/s are excluded (784 and 1052 systems for masked and unmasked runs). Since these are the objects with the largest redshift shifts, dropping them mechanically narrows the remaining voff distribution; this post-hoc selection is never quantified. Together these gaps mean the paper has not established that masking, rather than template improvements or sample trimming, recovers the z<1.5 voff population. This is load-bearing because the proposed mitigation and the claim that DESI QSO redshifts are biased at z>1.5 both depend on that causal attribution.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies Mg II absorption-line systems in DESI EDR/DR1 QSO spectra. The sample consists of 50,674 QSOs with a single detected Mg II absorber at voff < 20,000 km/s, plus a matched control sample of 50,674 QSOs with no detected Mg II absorbers. The authors fit each spectrum with a blue QSO composite template and an SMC extinction curve to estimate E(B-V), finding that intervening absorbers have an average E(B-V) of about 0.04 magnitudes, with E(B-V) increasing at lower absorber redshift and higher rest-frame equivalent width. Associated absorbers (voff < 3500 km/s) show a strong rise in E(B-V) toward voff = 0, peaking near 0.15 magnitudes. The paper then shows that the voff distribution of associated absorbers broadens and bifurcates at z > 1.5, and proposes that this is caused by Mg II absorption biasing the QSO redshifts. It attempts to mitigate this by masking the Mg II doublet pixels and rerunning Redrock, reporting that this narrows the voff distributions and shifts QSO redshifts by Delta z of roughly +/-0.005.","tokens_in":17365,"tokens_out":6822,"duration_ms":56109,"significance":"If the causal attribution is correct, this paper identifies a redshift-dependent systematic in DESI QSO redshifts that could affect both absorber-host studies and cosmological analyses using DESI QSOs. The paper's strengths include the large sample, the explicit control sample, the public data release on Zenodo, and a straightforward and reproducible extinction-fitting procedure. The authors also honestly flag limitations, including the unexpected low-redshift control behavior in Figure 5 and the reduced effectiveness of masking at z > 2.0. However, the central mitigation claim--that masking Mg II pixels, rather than the concurrent Redrock template update or the post-hoc removal of large-shift systems, recovers the low-redshift voff population--is not yet demonstrated by a controlled same-version comparison.","major_comments":[{"comment":"The demonstration that masking Mg II pixels corrects DESI QSO redshifts is not a controlled masking experiment. The plotted comparison is between the original catalog redshifts (Redrock versions 0.15.3 and 0.17.0) and a masked re-run with Redrock v0.20.0, whose QSO templates were trained on a substantially larger sample. The authors state that they also re-ran v0.20.0 without masking 'to ensure we can separate the effect of our masking strategy from that of the new QSO templates,' but no unmasked v0.20.0 version of Figures 7, 8, or 9 is shown. Without that same-version unmasked control, the improvement in Figure 8 could be caused by the template update rather than by masking. Please add the unmasked v0.20.0 voff distributions and a direct masked-vs-unmasked difference (Delta z) computed with an identical selection.","section":"Section 3.3, Figures 8 and 9"},{"comment":"The post-hoc exclusion of systems whose re-run redshift moved outside voff < 3500 km/s (784 for the masked run and 1052 for the unmasked run) removes the objects with the largest redshift shifts, and dropping them mechanically narrows the remaining voff distribution. The 268-system difference between the masked and unmasked exclusions may itself account for part of the apparent improvement. Please show the full distributions before exclusion (for example, by retaining all systems at their new voff values or by displaying them as a separate population), and quantify how the trimming changes the 16th, 50th, and 84th percentiles reported in Figures 7 and 8.","section":"Section 3.3, paragraph on excluded systems"},{"comment":"The analysis assumes that the Mg II absorber redshifts in the EDR/DR1 catalog are correct and that all voff broadening at z > 1.5 arises from errors in the QSO redshift. The paper notes that Mg II falls beyond 8400 Angstroms at z > 2.3 where DESI spectra are noisier, but it never validates absorber redshifts against other transitions (for example, C IV, Fe II, or [O II]) or against an independent line list. Because the same catalog defines both the mask and the reference redshift z_ALS in the definition of voff, any systematic error in the absorber redshifts would directly produce an apparent QSO-redshift bias. A cross-check of a subset of z > 1.5 absorbers with other metal lines would substantially strengthen the causal interpretation.","section":"Sections 2.1 and 3.2"},{"comment":"Negative E(B-V) fits are treated as physical measurements in the reported medians and trends, although the paper recognizes that negative values likely reflect the spread in intrinsic QSO power-law slopes relative to the assumed blue composite. Since the fit has only two parameters (normalization and E(B-V)) and the SMC extinction curve is assumed, the absolute scale of the reported E(B-V) values and the shape of the E(B-V)-voff trend depend on the distribution of intrinsic spectral indices in the fitted sample. The claim that associated absorbers show E(B-V) rising to about 0.15 at voff = 0 would be strengthened by a test that either marginalizes over spectral index, restricts the sample to a narrow color or luminosity range, or compares with a composite-stacking approach at the same voff.","section":"Sections 2.2 and 3.1"}],"minor_comments":[{"comment":"Please state explicitly whether the unmasked v0.20.0 rerun exists as a machine-readable table or figure; if so, include a direct comparison, since the current text only summarizes the number of systems (1052 versus 784) that leave the associated sample.","section":"Section 3.3"},{"comment":"The caption says 'Bins are 0.04 units wide'; please specify that this is in absorber redshift and, for the control sample, the matched absorber redshift.","section":"Figure 5 caption"},{"comment":"The sentence 'We can note that the associated absorber sample contains 28,178 systems with E(B-V) = 0.1092 magnitudes' does not state whether this is the mean or median; please state the estimator and use consistent terminology with the abstract and Figure 3.","section":"Section 3.1"},{"comment":"Equation (1) uses the speed of light c without explicitly defining it; please add a definition.","section":"Equation (1)"},{"comment":"In the conclusion, the reported median E(B-V) = 0.052 appears without the asymmetric 16th/84th percentile uncertainties given in Section 3; please include them for consistency.","section":"Section 5"},{"comment":"The data-release statement should include the exact Redrock version and the pixel-masking recipe used for the masked re-run, so that Figure 8 is reproducible from the Zenodo data alone.","section":"Data availability"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the paper is a solid census of Mg II absorber reddening in DESI, and the extinction trends are likely of interest. The main risk is that the central mitigation claim is confounded by the Redrock template update and by the exclusion of large-shift systems. I would encourage the authors to add the unmasked v0.20.0 comparison and the pre-trim distributions; with those, the paper could be acceptable. The paper is within the scope of the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. The paper's genuinely new result—that associated Mg II absorbers bias DESI quasar redshifts above z~1.5 and that masking the doublet pixels and re-running Redrock pulls the velocity-offset distribution back toward the low-z shape—is plausible but not yet demonstrated cleanly. The comparison in Fig. 8 is against the original catalog redshifts, whose Redrock versions differ from the re-run version, and the authors exclude the very systems with the largest redshift shifts. The extinction part is a much more solid, incremental confirmation of known SMC-like reddening trends on a 50,674-spectrum sample.\n\nWhat is good: per-spectrum template fitting with an external blue composite, a matched control sample, and a candid discussion of the low-z control anomaly and the negative E(B-V) tail. The E(B-V) trends with voff, absorber redshift, and equivalent width are physically sensible and match Zhu & Menard and Chen et al. This is real measurement work, and the redshift-bias idea is worth taking seriously.\n\nWhere it is soft. The masking test is the load-bearing evidence for the redshift bias, but the paper does not show the unmasked v0.20.0 voff distribution even though it says that run was done. So the reader cannot separate the effect of masking from the effect of the new QSO templates. On top of that, 784 (masked) and 1052 (unmasked) systems that moved outside voff<3500 km/s are excluded; those have the biggest redshift shifts, and removing them mechanically narrows the histograms. Both issues need to be addressed head-on. A second, lesser concern: the absorber redshifts come from the same catalog used to define the mask, and at z>2.3 MgII sits in the noisy red end of DESI, so an independent check of absorber redshifts (e.g., CIV or other transitions) would strengthen the interpretation. The unexplained low-z control sample behavior in Fig. 5 is a real loose end, but it does not obviously break the relative trends between associated and intervening systems.\n\nBottom line: the redshift systematic is probably real, and the paper is honest about its limits. But the headline claim is not fully controlled, so I would not take the masking recovery as quantified at face value yet. This deserves a serious referee—send it out—with a request for the same-version unmasked comparison and a sensitivity run to the exclusions. I'd cite it for the extinction measurements and as a caution flag on DESI QSO redshifts.","headline":"A plausible redshift bias from associated Mg II absorbers, supported by a large sample, but the masking demonstration is under-controlled and needs a cleaner comparison.","tokens_in":18085,"tokens_out":3247,"would_cite":true,"duration_ms":30130,"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":"Associated Mg II absorbers bias DESI quasar redshifts at $z>1.5$, and masking the doublet pixels before re-fitting recovers velocity-offset distributions like those at lower redshift, with typical shifts of $\\Delta z \\approx \\pm 0.005$.","keywords":["Mg II absorbers","quasar redshifts","extinction","E(B-V)","DESI","associated absorption systems","velocity offset","redshift bias"],"falsifier":"Take a sample of associated absorbers at $1.51.5$ the velocity-offset distribution of associated absorbers broadens and bifurcates in a way the authors argue is nonphysical, and masking the Mg II doublet pixels and re-running the redshift fitter narrows those distributions and shifts quasar redshifts by typically $\\Delta z \\approx \\pm 0.005$. This identifies a redshift-dependent systematic in DESI quasar redshifts that can be at least partially removed by line masking.","feed_headline":"Masking Mg II lines fixes biased quasar redshifts in DESI","feed_subtitle":"Associated absorbers skew DESI quasar redshifts at z>1.5; masking the doublet restores their velocity offsets.","key_machinery":"The central diagnostic is the velocity offset $v_{\\rm off} = c(z_{\\rm QSO} - z_{\\rm ALS})/(1+z_{\\rm QSO})$, which separates associated from intervening absorbers and exposes the redshift bias through its distribution. The corrective machinery is a masking-and-refit procedure: using each absorber's fitted redshift and line width from the catalog, the authors flag every pixel within $5\\sigma$ of either line of the Mg II doublet, exclude those pixels, and rerun Redrock — DESI's template-matching redshift code — with updated quasar templates on the masked spectrum. Comparing $v_{\\rm off}$ before and after masking isolates the contribution of the absorption lines to the quasar redshift measurement.","core_discovery":"The paper's central claim is that associated Mg II absorbers (velocity offset below $3500\\,\\mathrm{km\\,s^{-1}}$) do not merely redden a quasar; they systematically change its measured redshift. The authors find that at $z>1.5$ the $v_{\\rm off}$ distribution of associated absorbers broadens, its median moving from roughly 140–240 km/s at lower redshift to 700–1000 km/s, and then splits into two peaks near $\\pm1500$ km/s, while the most reddened systems pile up at strongly negative $v_{\\rm off}$. They interpret this as the Mg II doublet distorting the Mg II emission line that the DESI redshift pipeline fits. Masking the doublet pixels and re-running the fitter with updated quasar templates shifts typical redshifts by $\\Delta z \\approx \\pm 0.005$, most effectively at $1.62$.","pith_inferences":["A step the authors do not take, but that follows naturally, would be to use the measured $\\Delta z$ as a calibration prior: build a correction map in $z$ and $v_{\\rm off}$ and apply it to the full DESI DR1 quasar sample, including systems without detected absorbers, since the bias may also affect weak or undetected absorbers.","The bifurcation pattern suggests the Mg II doublet is being fit as if it were velocity-shifted emission; injecting synthetic absorbers into spectra at known redshifts and observing the fitted $z$ would directly test this mechanism and could predict masking performance from line strength alone.","Cross-checking absorber redshifts with other transitions (Fe II $\\lambda2600$, the C IV doublet, or Mg I) in the same systems would determine whether the residual high-$z$ broadening is a quasar-side or absorber-side problem, a test the paper does not include.","The reddish control-sample behavior at $z<1$ is left unexplained; if real, it means even DESI quasars without detected Mg II have dust or template mismatches that could produce extinction and redshift errors of the same class."],"forward_implications":["DESI quasar redshifts at $z>1.5$ that host associated Mg II absorbers carry a systematic offset of order $\\Delta z \\approx \\pm 0.005$, peaking near $z\\approx 1.8$, which will propagate into any cosmology or clustering analysis that uses those redshifts.","Masking the Mg II doublet before redshift fitting is a practical correction: it recovers velocity-offset distributions that look like the $z<1.5$ population for most of the $1.52$, where C IV, C III, and Lyman-$\\alpha$ dominate the fits, further corrections that handle those broad lines will be needed before quasar redshifts in that regime are fully cleaned."],"supporting_citations":[{"why":"Constructs the EDR Mg II absorber catalog and the line-fitting/MCMC methods that this paper extends to DR1, supplying the absorber redshifts, line widths, and equivalent widths used throughout.","marker":"Napolitano et al. (2023)"},{"why":"Defines the DESI QSO target selection and Redrock-based redshift decision matrix used to assign final catalog redshifts and to interpret the masked re-runs.","marker":"Chaussidon et al. (2023)"},{"why":"Provides the blue X-shooter QSO composite and extinction-fitting approach the authors use to measure $E(B-V)$ toward each absorbed spectrum.","marker":"Fawcett et al. (2022)"},{"why":"Establishes the $\\pm3500$ km/s boundary between associated and intervening absorbers and supplies prior results on associated-absorber extinction that this paper tests and refines.","marker":"Zhu & Ménard (2013)"},{"why":"Earlier template-fitting work showing that associated absorbers are more reddened than intervening ones and that an SMC-like extinction curve applies; the paper's extinction measurements are compared against it.","marker":"Vanden Berk et al. (2008a)"},{"why":"Supplies the SMC Bar extinction curve with $R(V) = 2.74$ that the fitting model adopts.","marker":"Gordon et al. (2003)"},{"why":"Builds the improved Redrock quasar templates, trained on a sample roughly 400 times larger than earlier templates, that the authors use for both masked and unmasked redshift re-runs.","marker":"Brodzeller et al. (2023)"},{"why":"Quantifies the roughly 200 km/s intrinsic uncertainty of broad-line-only QSO redshifts, which the paper uses to interpret the observed $v_{\\rm off}$ broadening.","marker":"Shen et al. (2016)"}],"fun_headline_variants":["Mg II absorbers skew DESI quasar redshifts; masking fixes it","Masking Mg II lines restores DESI quasar redshifts","DESI quasar redshifts biased by Mg II absorbers at z>1.5","Fix for DESI quasar redshifts: mask Mg II doublet"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the catalogued Mg II absorber redshifts are accurate, so all nonphysical broadening of $v_{\\rm off}$ at $z>1.5$ is blamed on errors in the quasar redshift; if absorber redshifts are themselves biased at high $z$, the bifurcation and the correction from masking are misattributed.","fun_headline_variants_meta":{"raw":{"variants":["Mg II absorbers skew DESI quasar redshifts; masking fixes it","Masking Mg II lines restores DESI quasar redshifts","DESI quasar redshifts biased by Mg II absorbers at z>1.5","Fix for DESI quasar redshifts: mask Mg II doublet"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000385,"raw_usage":{"total_tokens":2117,"prompt_tokens":1105,"completion_tokens":1012,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":721,"completion_tokens_details":{"reasoning_tokens":933}},"tokens_in":721,"tokens_out":1012,"duration_ms":5396,"temperature":1.0,"reasoning_tokens":933,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T11:28:38.838258+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a sample of associated absorbers at $1.5