{"id":"e6934797-aaa2-43ab-8a05-33d5086be831","arxiv_id":"2607.18775","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Corrected RACS-Mid1 and RACS-High1 source positions achieve about 0.25 arcsecond accuracy for compact sources across most of the southern sky.","lead":"This paper applies a refined astrometric correction framework to two higher-frequency epochs of the ASKAP RACS survey, reducing systematic positional errors to near zero and achieving about 0.25-arcsecond accuracy for compact sources. It matters because precise southern-sky radio positions underpin fast radio burst localizations and multiwavelength cross-identification.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"0.25'' accuracy claim is based on a heavily filtered compact-source subsample; the paper does not validate fainter or resolved sources, so the unqualified headline overstates catalogue-wide precision.","rationale":"The reader's weakest assumption—that calibration errors decompose cleanly into beam and scan offsets—is plausible but not the most load-bearing issue. Even if unmodeled direction-dependent structure exists within a beam, the residual scatter measured after correction (against WISE, and independently against FIRST and RFC) would still capture that structure as part of the quoted uncertainty. The additive decomposition is therefore not a fatal flaw for the empirical error bars; it only limits interpretability of the fitted components. The deeper problem is the mismatch between the validation sample and the headline claim. All validation is performed on compact, isolated, bright sources, and the RFC comparison even restricts to sources passing filters in both epochs. The paper then states 0.25''/0.35'' as default catalogue-wide uncertainties. Extended, faint, or blended sources—which constitute a large fraction of real catalogues—could plausibly have larger centroid errors due to noise, morphology, and primary-beam asymmetries, and the paper provides no evidence to the contrary. This is a correctness risk that affects the practical use of the data products, not just the mathematical completeness of the model. The paper does have independent support: FIRST and RFC cross-checks, FRB localisation simulations, and a favourable comparison against polynomial corrections. The concern is addressable by qualifying the claim or by re-running the validation on unfiltered subsamples. Since the reader already recommended conditional acceptance and our concern reinforces that recommendation rather than changing it, the verdict remains unchanged.","tokens_in":17456,"tokens_out":6310,"duration_ms":66329,"concrete_test":"Compute RFC and FIRST residuals for the full RACS-Mid1/RACS-High1 source lists without imposing the Section 2.1 compactness or SNR filters, and split the sample by SNR (6–10 vs >10) and by integrated-to-peak flux ratio (<1.2 vs >1.5). If the 68% confidence intervals for the low-SNR or extended subsamples exceed the claimed 0.25″ (off-plane) or 0.35″ (on-plane) by more than 0.05″, the headline accuracy claim must be qualified to compact sources.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—'individual corrected RACS source positions are accurate to ~0.25″ across the bulk of the sky'—is supported only for a narrowly selected subpopulation. Step 2 of Section 2.1 removes sources with SNR≤6, extended or blended sources (integrated/peak flux ratio >1.5 or multiple components within 30″). The RFC validation in Section 3.1 goes further, requiring sources to survive the same filtering in both RACS-Mid1 and RACS-High1 (reducing ~9,000 to just under 6,000 matches). FIRST and RFC comparisons are therefore purely on compact, isolated, bright sources. The correction model itself (Section 2.1, step 4) only solves for constant beam and scan translations; it cannot correct morphology-dependent centroid shifts, primary-beam asymmetries, or noise-dependent biases in fainter sources. The abstract and conclusion state the 0.25''/0.35'' uncertainties as catalogue-wide default values without the compact-source qualification. Users cross-matching extended, blended, or low-SNR RACS sources would be applying an accuracy claim that has no direct empirical support. This is the most load-bearing gap because it affects the practical interpretation of every corrected position, not just the model's completeness.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper extends a previously developed astrometric correction framework to the RACS-Mid1 and RACS-High1 epochs. It operates on per-beam source lists, uses a hierarchical crossmatching strategy (RACS-Low1/VLASS for Stage 1, WISE for Stage 2), and models calibration-induced offsets as the sum of a scan-independent beam offset and a beam-independent scan offset. The authors report that median offsets are removed and the 68% confidence interval of residuals averaged over ~1 sq.deg. regions improves from ≳0.4″ to ≲0.18″ relative to WISE. Independent validation against FIRST and RFC yields per-source 68% intervals of roughly 0.17–0.20″ for a filtered population of compact, isolated, SNR>6 sources, and the paper recommends 1σ systematic uncertainties of 0.25″ off the Galactic plane and 0.35″ on the plane. The paper also compares with polynomial corrections, simulates FRB localization improvements, and discusses incorporation into the CELEBI pipeline.","tokens_in":17842,"tokens_out":6620,"duration_ms":55442,"significance":"If the claims are accepted, this is a valuable contribution: it would establish RACS-Mid1 and RACS-High1 as the highest-precision arcsecond-resolution all-sky astrometric references in the southern hemisphere above 1 GHz, with direct benefits for FRB localization and multiwavelength cross-identification. The methodology is transparent, the code is available on GitHub, and the use of independent FIRST and RFC comparisons is a genuine strength. The honest characterization of non-Gaussian residual tails and the recommendation to inspect sky maps are also positive features. However, the headline accuracy is presented without the sample-selection caveats that the analysis itself reveals, and one supporting statement about the confidence intervals is numerically inconsistent. These issues need to be addressed before the published claims can be used safely by the community.","major_comments":[{"comment":"The abstract and conclusion state that 'individual corrected RACS source positions are accurate to a 1σ confidence level of ~0.25″ across the bulk of the sky' without explicitly limiting this to compact, isolated, SNR>6 sources. The RFC validation in Section 3.1 restricts the sample to sources that survive the Section 2 filtering in both RACS-Mid1 and RACS-High1, reducing the sample from ~9,000 to just under 6,000. The FIRST comparison similarly relies on the same filtering. The correction model (Section 2.1, step 4) only solves for constant beam and scan translations; it cannot correct morphology-dependent centroid shifts, blending, or noise-dependent biases in fainter sources. The accuracy claim therefore has no direct empirical support for extended, blended, or low-SNR sources. The abstract and conclusion should either qualify the claim as applying to the filtered compact population o","section":"Abstract / Section 3.1 / Section 5"},{"comment":"The headline reduction from ≳0.4″ to ≲0.18″ is the 68% confidence interval of residuals measured against WISE, which was also used in Stage 2 to fit the per-beam/per-scan offsets. This is therefore a partly in-sample statistic and does not independently establish the accuracy of the corrections. The independent FIRST and RFC comparisons (Tables 3 and 4) yield per-source 68% intervals of roughly 0.17″–0.20″ for the filtered compact sample, and the paper's adopted systematic uncertainties are 0.25″/0.35″. The abstract should not present the WISE-based 0.18″ as the achieved accuracy without clarifying that WISE was used to fit the corrections; the independent validation gives a more conservative, but more trustworthy, per-source accuracy.","section":"Section 3.1, Tables 1 and 2, Figures 1, 2, 4, 5"},{"comment":"The statement that the adopted 1σ uncertainties of 0.25″ (off-plane) and 0.35″ (on-plane) are 'consistent with, or slightly conservative relative to, the 95% confidence intervals reported in Tables 3 and 4' is numerically inconsistent. The 95.4% confidence intervals for the RFC comparison in Tables 3 and 4 are approximately ±0.5″–0.6″, substantially wider than 0.25″. If the intent is to provide values that account for the non-Gaussian tails, the justification should be restated in terms of, for example, the standard deviation of the RFC residuals or a specific quantile of the empirical distribution. As written, the reasoning is unclear and could lead users to underestimate the impact of the tails.","section":"Section 3.3"},{"comment":"The correction model assumes a strict additive decomposition of calibration errors into a scan-independent beam offset and a beam-independent scan offset. The paper does not provide a direct test of whether residual errors contain additional direction-dependent or time-varying structure within a beam. The independent FIRST/RFC validation only samples the filtered compact population, and the WISE-based residual maps (Figures 1 and 4) may still show spatial structure that could indicate model incompleteness. A diagnostic such as comparing residuals from overlapping scans, splitting the data by time or beam, or fitting and removing the model and then checking for remaining spatial correlations would help assess this assumption. Without such a test, the systematic uncertainty budget may be underestimated for directions where the additive model does not hold.","section":"Section 2.1, step 4"}],"minor_comments":[{"comment":"The data availability statement contains an empty placeholder after 'PASA Datastore:' and needs the actual URL or DOI to be inserted before publication.","section":"Data Availability"},{"comment":"The phrase 'closepack36' is spelled inconsistently: 'close pack36' in Section 2.2 and 'closepack36' in the introduction. Standardize to the ASKAP convention.","section":"Section 1 and 2.2"},{"comment":"The phrase 'mean residuals of RACS source positions averaged over ~1 sq.deg. regions' is ambiguous. It is unclear whether the reported 68% intervals describe per-source residuals or the distribution of beam-averaged means. Please define the statistic in the text.","section":"Abstract and Section 3.1"},{"comment":"The captions contain minor typographical issues: 'VLT R-band' and 'VLT g-band' should be consistent with the main text's 'VLT/FORS2' usage, and the final sentence in each caption appears truncated.","section":"Figures 11 and 12"},{"comment":"In the conclusion, 'RACS-high1' should be capitalized as 'RACS-High1' for consistency with the rest of the manuscript.","section":"Section 5"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid, useful extension of the authors' earlier work and the core methodology appears sound. The main obstacle to acceptance is the mismatch between the abstract/conclusion's unqualified accuracy claim and the heavily filtered validation sample used to support it. The numerical inconsistency about the 95% confidence intervals in Section 3.3 should also be corrected. I believe these issues are fixable within the manuscript's scope, but they are load-bearing for the practical interpretation of the delivered catalogues."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the corrections work. For compact, isolated, bright sources, the RACS-Mid1 and RACS-High1 positions are consistent with FIRST and RFC at the ~0.25″ level, with median offsets near zero. That is the core result, and it holds.\n\nThe genuinely new pieces are the corrected mid/high-frequency source lists themselves; the hierarchical Stage 1 crossmatch against corrected RACS-Low1 (and VLASS north) that lets the authors chase 12″ initial offsets without drowning in false WISE matches; and keeping the Galactic plane in the model rather than excluding it. The comparison against the earlier declination-polynomial corrections is a nice addition—it shows a real factor-of-two improvement in residual scatter by modelling per-beam and per-scan offsets. The FRB localisation simulation and the two worked examples make the practical benefit concrete.\n\nThe soft spots are real but not fatal. The headline 0.18″ scatter comes from WISE residuals, and WISE was used to fit those same offsets, so that number is partly in-sample. The independent FIRST and RFC checks are the load-bearing validation, and they're done on a heavily filtered population: SNR>6, no extended structure, no blends, and in the RFC case only sources matching in both epochs. So the 0.25″ claim is solid for compact astrometrically clean sources, but there is no direct test for fainter, resolved, or blended sources. The authors would do well to say that in the abstract and conclusion, rather than presenting 0.25″ as a catalogue-wide default. They also note the residual distributions are non-Gaussian, and they advise checking the offset maps, so the paper is not hiding the issue.\n\nOne concrete problem: the data availability statement is empty—no link, no DOI in the text. The GitHub repo is mentioned but not versioned or described as containing the corrected catalogues. For a methods paper whose product is the catalogues, that needs fixing before publication.\n\nOverall: this is a competent, useful extension, not a breakthrough. It deserves a serious referee and probably a minor-to-moderate revision focusing on scoping the accuracy claims and making the products actually available. I'd cite it for the corrected RACS catalogues, and I'd bring it to a reading group for astro-instrumentation people.","headline":"RACS-Mid/High astrometric corrections are real and externally validated for compact sources; the catalogue-wide 0.25″ claim needs qualification, and the data release is not yet accessible.","tokens_in":753,"tokens_out":867,"would_cite":true,"duration_ms":235822,"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":"This paper claims that an additive two-component model of telescope calibration errors, applied beam-by-beam to unmosaicked source lists, brings the mid- and high-frequency epochs of the Rapid ASKAP Continuum Survey to ~0.25-arcsecond astro","keywords":["astrometry","radio continuum surveys","RACS","ASKAP","positional calibration","systematic errors","source catalogues","fast radio bursts"],"falsifier":"A direct test would compare corrected positions of many unresolved sources within a single beam against VLBI positions. If, after correction, the residuals show a spatial gradient, curl, or systematic dependence on position within the beam that exceeds the quoted 0.25″ uncertainty, the two-component model is incomplete. Repeated scans of the same field under different ionospheric conditions, examined beam-by-beam, could similarly expose time-dependent error terms that the model cannot absorb.","tokens_in":17381,"feed_emoji":"📡","tokens_out":7434,"duration_ms":61841,"temperature":0.7,"pith_summary":"The paper seeks to extend an existing astrometric correction framework, previously applied to low-frequency epochs of the Rapid ASKAP Continuum Survey (RACS), to the survey's mid- and high-frequency epochs. It argues that the dominant systematic positional errors in these data can be modelled as the sum of two calibration-driven components: a scan-independent beam offset and a beam-independent scan offset. After fitting and removing these offsets, the median positional offsets are effectively zero and the 68% confidence interval of mean residuals over ~1 square degree regions falls from ≳0.4″ to ≲0.18″ across most of the sky. Validation against two external radio reference catalogues supports a 1σ accuracy of ~0.25″ for individual corrected source positions outside the Galactic plane. If correct, the work provides the southern hemisphere with its first sub-arcsecond, arcsecond-resolution astrometric reference at decimetre wavelengths above 1 GHz, with direct benefits for fast radio burst localisation and multiwavelength source association.","feed_headline":"Southern-sky radio positions sharpened to ~0.25 arcsec","feed_subtitle":"Beam-by-beam correction cuts large-scale offsets, giving southern astronomy a sub-arcsecond radio reference for transient studies.","key_machinery":"The central mechanism is the additive offset model operating on per-beam source lists. Each scan's positional error is decomposed into a scan-independent beam offset (common to all scans sharing a bandpass calibration) and a beam-independent scan offset (common to all beams in a single scan). Working directly from unmosaicked per-beam catalogues circumvents the additional positional blur introduced by mosaicking at beam boundaries; a two-stage crossmatch—first against a lower-frequency corrected catalogue or a northern survey using a generous 12-arcsecond radius to catch large offsets, then against a dense infrared reference at 2 arcseconds to refine—keeps false associations low. Fitting the","core_discovery":"The paper establishes that a two-term additive error model captures the dominant systematic astrometric distortions in the unmosaicked per-beam source lists of the RACS-Mid1 and RACS-High1 epochs. The correction removes the mean RA and Dec. offsets, reduces the 68% scatter in mean residuals to below 0.18″ over most of the survey, and yields median offsets of ~0.01″ against the FIRST and RFC reference catalogues after compact-source filtering. It further demonstrates that this beam-resolved, per-scan/per-beam modelling outperforms a global declination-dependent polynomial correction by roughly a factor of two in residual scatter.","pith_inferences":["If the two-component error model holds, a similar decomposition may improve astrometric calibration for other wide-field interferometers that form multiple simultaneous beams, provided per-beam source lists are retained.","The non-Gaussian tails the paper reports in residual distributions suggest a small direction-dependent or time-varying error component remains; testing for spatial structure within individual beams against VLBI reference sources would reveal whether a third model term is needed.","The validation is limited to filtered compact, isolated sources, so the ~0.25″ claim should not be assumed to hold for extended or confused sources, where the paper itself flags degraded performance in the Galactic plane.","A practical consequence the paper does not spell out: users working near the Galactic plane should check the provided sky maps of residual offsets rather than applying the global 0.25″ value, since local errors can be larger."],"forward_implications":["The corrected RACS-Mid1 and RACS-High1 catalogues constitute the first sub-arcsecond astrometric reference across the southern sky at frequencies above 1 GHz, improving by more than a factor of five over earlier southern surveys.","Fast radio bursts detected by ASKAP can be localised with lower systematic error when the reference catalogue is matched to the observing frequency, sharpening host-galaxy identification.","Because the framework generalises across ASKAP's frequency range, the survey can support a uniform cross-band astrometric standard for future large-area work.","Adopting the paper's recommended conservative uncertainties of ~0.25″ (off-plane) and ~0.35″ (on-plane) gives users a practical, data-backed error budget for cross-matching and population studies."],"fun_headline_variants":["Radio positions across southern sky now accurate to 0.25 arcsec","ASKAP survey sharpens astrometry to 0.25 arcsec","RACS mid/high data get sub-arcsec positional corrections","Beam-by-beam fixes cut radio positional errors to 0.25 arcsec","Southern-sky radio astrometry hits 0.25 arcsec precision"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the calibration-induced positional error decomposes cleanly into two additive terms—a scan-independent beam offset and a beam-independent scan offset—so that any real error with additional direction-dependent or time-varying structure inside a beam will not be captured by the fit and will silently inflate the true astrometric error beyond the quoted residual scatter.","fun_headline_variants_meta":{"raw":{"variants":["Radio positions across southern sky now accurate to 0.25 arcsec","ASKAP survey sharpens astrometry to 0.25 arcsec","RACS mid/high data get sub-arcsec positional corrections","Beam-by-beam fixes cut radio positional errors to 0.25 arcsec","Southern-sky radio astrometry hits 0.25 arcsec precision"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000198,"raw_usage":{"total_tokens":1255,"prompt_tokens":846,"completion_tokens":409,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":590,"completion_tokens_details":{"reasoning_tokens":312}},"tokens_in":590,"tokens_out":409,"duration_ms":6568,"temperature":1.0,"reasoning_tokens":312,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T14:19:57.066123+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test would compare corrected positions of many unresolved sources within a single beam against VLBI positions. If, after correction, the residuals show a spatial gradient, curl, or systematic dependence on position within the beam that exceeds the quoted 0.25″ uncertainty, the two-component model is incomplete. Repeated scans of the same field under different ionospheric conditions, examined beam-by-beam, could similarly expose time-dependent error terms that the model cannot absorb.","supporting_citations":[],"review_version":1}