{"id":"fc055eed-42e7-4136-a696-a3cefbfdcf8a","arxiv_id":"2608.10577","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"An XGBoost model assigns infrared-photometry-based distances to 36,000 AGB stars, recovering the Galactic bar and a bulge-to-disk vertical dispersion transition.","lead":"This study trains an XGBoost model on infrared photometry to assign distances to more than 36,000 dust-obscured AGB stars in the inner Milky Way, then uses the resulting catalog to trace the Galactic bar and vertical structure. The catalog is a useful new resource, but the 6% stated accuracy measures agreement with the training distances, not absolute distance accuracy, so the results should be read with that caveat.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The MSX-to-AKARI transfer is asserted but not quantitatively demonstrated; the model is applied to far-side AKARI sources outside the MSX training distribution, so far-side distance-dependent results may be artifacts.","rationale":"We independently reviewed the manuscript. The central deliverable is a statistical distance catalog for more than 36,000 AKARI AGB stars, and the paper's structural conclusions (bar, period-dependent Mira distribution, vertical dispersion) are only as trustworthy as those distances. The most load-bearing assumption is that the XGBoost model trained on MSX-selected sources with SED distances transfers to the AKARI sample. The MSX sample has a known far-side deficit (Fig. 1), while the AKARI sample is deeper (Sect. 2.1) and includes more far-side sources; the paper asserts in Sect. 3.1 that the feature spaces are similar, but presents no quantitative evidence (e.g., feature-space overlap or closest-training-sample distances). This is especially important because the reported 6% MAPE is measured against SED distances on a random 20% test split of the same MSX sample, not against the AKARI application set. The external validations (Gaia, PL) are on different or overlapping subsets, and the PL comparisons show offsets up to 36% (OGLE) and ~12% (Galactic mid-IR), so the absolute distance scale remains uncertain. The authors are honest about the far-side enhancement (they decline to interpret it as physical) and acknowledge the distances are statistical. Nevertheless, the bar morphology and vertical-dispersion results rely on the same extrapolated distances. We therefore recommend the same conditional verdict: the catalog is potentially useful, but the transferability and error budget need to be demonstrated with a side-split or feature-space validation before the far-side structural claims are accepted. Our concern aligns with the reader's weakest-assumption identification.","tokens_in":24932,"tokens_out":6494,"duration_ms":57409,"concrete_test":"Re-train the XGBoost regressor on the MSX/SED training sample restricted to Galactic longitudes l > 0 only, and evaluate it on the held-out l < 0 MSX sources (the far-side subset, which is sparse in the full training set); compare the far-side MAPE and median bias to the reported ~6% MAPE, particularly for SED distances > 8 kpc. If the far-side MAPE substantially exceeds 6% or the bias grows with distance, the MSX-to-AKARI transfer claimed in Sect. 3.1 is not validated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central product is a distance catalog for 36,134 AKARI-selected AGB stars, and its headline results (bar morphology, period-dependent Mira structure, vertical dispersion) depend on those distances. The load-bearing assumption is stated in Sect. 3.1: \"The AKARI-selected sources occupy a near- and mid-IR color-magnitude space similar to that of the MSX-selected training sample, so applying the model to the AKARI catalog does not require a large extrapolation.\" This is not quantitatively demonstrated. Fig. 1 shows the MSX training sample has a pronounced far-side deficit (-30 <= l <= 0), while the AKARI sample provides substantially more far-side coverage; Sect. 2.1 states AKARI reaches ~2-3 mag fainter. Thus the model is applied to sources that are systematically fainter and more distant than most training examples, and to a side of the Galaxy poorly represented in training. XGBoost is an interpolator; when features fall outside the training manifold, predictions are unconstrained and can inherit the training set's selection bias. The paper's own Fig. 15 reveals a far-side enhancement for short-period Miras; the authors explicitly disclaim it as a physical result (Sect. 5.3.2), but the same extrapolation risk applies to the longer-period Mira bar signal and the vertical-dispersion map, which are presented as primary findings. Without a quantitative feature-space overlap check or a side-split validation, the transfer from the MSX-deficient training set to the AKARI far-side sample is the weakest link in the chain.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops an XGBoost regression model that maps 2MASS and AKARI infrared photometry to heliocentric distances for oxygen-rich AGB stars, trained on MSX-selected stars with SED-based distances from Bhattacharya et al. (2024). The model achieves R^2 ≈ 0.98 and a MAPE of about 6% on a held-out test set drawn from the same MSX sample, and is then applied to 36,134 AKARI-selected AGB candidates that do not overlap the training sample. The resulting distance catalog is validated by comparison with Gaia parallaxes (for 166 sources at |b| > 5 deg), with several period-luminosity relations, and with OH/IR bolometric luminosities. Using these distances, the authors report that the combined AKARI+MSX sample recovers the Galactic bar through an asymmetry in near/far distance distributions, finds a roughly constant vertical dispersion in the bulge that is about 30% larger than in the disk, and shows that long-period Miras (P > 400 d) trace a barred morphology more strongly than short-period Miras. The paper is careful to describe the distances as statistical, with a quoted per-source uncertainty of about 35.5%, and explicitly declines to interpret a far-side enhancement of short-period Miras as a physical result.","tokens_in":25204,"tokens_out":3810,"duration_ms":37425,"significance":"If the distance catalog is reliable at the population level, it would be a useful resource for mapping the inner Milky Way in heavily obscured regions where Gaia astrometry and kinematic distances are unavailable or biased. The methodology is straightforward, reproducible in principle, and the paper includes several diagnostic checks: 5-fold cross-validation, Optuna-based hyperparameter optimization, a held-out test set, binned 1:1 diagnostics, and comparisons with multiple independent distance estimators. The paper also explicitly acknowledges that the ML distances are tied to the SED-based distance scale of the training set and cannot be more accurate than those training labels. The main scientific claims, that long-period Miras preferentially trace the bar and that bulge and disk vertical dispersions differ, are consistent with prior work and are presented with appropriate caution. However, the central transfer from the MSX training sample to the AKARI application sample, and the interpretation of the 6% MAPE as a measure of accuracy, need stronger quantitative support before the catalog and its structural inferences can be fully endorsed.","major_comments":[{"comment":"The load-bearing assumption that the AKARI-selected sources occupy a color-magnitude space 'similar' to the MSX training sample is asserted without a quantitative demonstration. Figure 1 shows that the MSX training set has a pronounced far-side deficit for -30 deg < l < 0 deg, while the AKARI sample provides substantially more far-side coverage, and Section 2.1 states that AKARI reaches ~2-3 mag fainter than MSX. Because XGBoost is an interpolator, predictions for AKARI sources that are fainter or on the far side can be unconstrained and may inherit the training set's selection bias. I recommend adding a quantitative feature-space overlap analysis (e.g., density distributions of AKARI sources versus the training set in the five-dimensional photometric space, or a nearest-neighbor distance statistic) and a side-split or far-side validation in which the model is trained on one longitude range and tested on the other. Without such a check, the far-side structural results in Figures 8, 12, 13, 14, and 15 may be artifacts of extrapolation.","section":"Section 3.1, with implications for Sections 4.2, 5.2, and 5.3.2"},{"comment":"The headline '6% MAPE' measures the model's ability to reproduce the SED-derived training labels, not the absolute accuracy of the distances. The paper states this in Section 3.1, but the abstract and Section 4.1.3 present the 6% value in a way that can easily be misread as the total distance error. In addition, the per-source error budget of 35.5% is obtained by summing the maximum SED uncertainty (~35%) and the mean ML MAPE (~6%) in quadrature. This is not a proper propagation of per-source uncertainties: the 35% value is an upper limit, not a random error, and the 6% is a mean absolute percentage error, not a standard deviation. I recommend rephrasing the abstract to say that the model reproduces SED-based distances with a MAPE of about 6%, and either providing per-source error estimates or explicitly labeling the 35.5% as a conservative ensemble-level envelope rather than an individual 1-sigma uncertainty.","section":"Table 3 and Section 4.2.2"},{"comment":"The paper correctly disclaims the far-side enhancement of short-period Miras as a physical result, but the same extrapolation risk applies to the long-period Mira bar signal and to the vertical-dispersion map. Although period is not an explicit model feature, period correlates with mid-IR colors and circumstellar dust emission, so period-correlated SED differences can induce apparent period-dependent spatial structure through the model's feature mapping. The argument that the model 'has no explicit mechanism' to assign period-dependent distances is therefore not sufficient. I recommend testing whether the bar morphology (Figure 14) and the vertical-dispersion profile (Figures 12-13) persist when the analysis is restricted to the distance and longitude range where the training sample has good coverage, or when the model is retrained on a longitude-balanced subset of the MSX sample.","section":"Section 5.3.2, Figures 14-15"}],"minor_comments":[{"comment":"The author names for the Gaia OH/IR catalogs are inconsistent: the text uses 'B. L. Marti et al. (2025)' and 'B. Lopez Marti et al. (2025)', while the reference list has both 'Lopez Marti, B., Jimenez-Esteban, F. M., Engels, D., & Garcia-Lario, P. 2025' and 'Marti, B. L., Jimenez-Esteban, F., Engels, D., & Garcia-Lario, P. 2025'. Please unify the citation style and remove the duplicate reference.","section":"References and Section 5.1.1"},{"comment":"The abstract quotes 'a 36% total error margin' while Section 4.2.2 derives 35.5%; please make these numbers consistent or explain the rounding.","section":"Abstract and Section 4.2.2"},{"comment":"The paper reports a 'MAPE-based accuracy metric' of ~94% and then translates it to 'a mean deviation of ~6%'. Reporting the MAPE directly (about 6%) would be clearer and would avoid the impression that the model achieves 94% accuracy in an absolute sense.","section":"Equation (1) and Section 4.1.1"},{"comment":"The caption says 'the blue points show the binned median trend, indicating the 16th-84th percentile spread within each bin,' but it is unclear whether the blue points are the medians and the shaded band represents the percentile spread, or whether both are shown. Please clarify the caption and ensure the figure legend matches.","section":"Figure 5 caption"},{"comment":"The bolometric luminosities derived from ML distances are compared with Feast et al. (1989) P-L expectations without explicitly stating in the comparison paragraph that the bolometric fluxes are not corrected for interstellar extinction; the paper later notes that extinction correction would raise the luminosities, but this caveat should appear before the comparison is interpreted.","section":"Section 5.1.2"},{"comment":"The colorbar units are given as N kpc^-2 in Figure 14 and N kpc^-3 in Figure 16; if these are intentional (projected versus volume-corrected densities), please state this in the captions, but currently the units appear inconsistent at a glance.","section":"Figures 14 and 16"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within scope for an astrophysics journal and the central idea is useful, but the review hinges on whether the MSX-to-AKARI transfer can be validated quantitatively. The authors' own Section 5.3.2 shows awareness of extrapolation risk for the short-period Mira far-side enhancement, but the same risk is not addressed for the long-period Mira bar morphology and vertical dispersion, which are presented as primary results. A revision that adds a feature-space overlap check and a side-split or far-side validation would materially strengthen the paper. I also note that the training labels (Bhattacharya et al. 2024) and one of the main P-L validation samples (Lewis et al. 2023b) share authors with the present paper, so the 'independent' validation is only partially independent; the Gaia and Sanders (2023) comparisons provide useful external checks, but they are limited to small subsamples."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nBottom line: the distance catalog is a real product, the ML is competently done, and the structural results are consistent with prior work. But the headline 6% MAPE is a reproduction-of-labels metric, not an absolute accuracy, and the transfer from the MSX training sample to the fainter, more far-side-complete AKARI sample is asserted rather than demonstrated. Both issues are fixable.\n\nWhat's new: 36,134 AKARI-selected O-rich AGB stars with distances, extending the SED-based distances from Bhattacharya et al. 2024 into a new photometric space. The model uses five bands (2MASS JHKs + AKARI 9/18), properly tuned with Optuna, 5-fold CV, held-out test. That part is solid. The paper is also honest: it explicitly says ML distances are tied to the SED scale, that the 6% is model reproduction, and that individual errors are ~35.5%. It disclaims the short-period Mira far-side enhancement. That honesty matters.\n\nThe soft spots. First, the 6% MAPE is not an accuracy claim against truth; the validation against Lewis et al. (2023b) mid-IR P-L is same-group, and the OGLE P-L shows 36% offset. The Sanders (2023) comparison is independent and looks okay, and the Catchpole (2016) mean-distance comparison is reassuring, but none of this directly quantifies the MSX-to-AKARI feature-space shift. Fig. 1 shows the MSX sample has a far-side deficit; the AKARI sample reaches 2-3 mag fainter. The model is applied to sources that are systematically fainter and more distant than most training examples. The paper states in Sect. 3.1 that no large extrapolation is needed, but doesn't show a feature-space overlap plot or a side-split error analysis. That's the load-bearing assumption, and it deserves a direct test.\n\nSecond, the 35.5% error is an upper bound applied uniformly. That's conservative, but it means the per-source errors in the catalog are not actually characterized. For population studies that's probably fine; for anyone using individual distances it's misleading. The paper should either calibrate the error as a function of magnitude/color or state clearly that the catalog is only for ensemble use.\n\nThe Galactic structure results—bar traced by long-period Miras, vertical dispersion difference—are confirmatory, not new. They do serve as a sanity check on the distances, which is useful.\n\nWho this is for: anyone working on inner-Galaxy AGB populations, maser follow-up, or bulge/bar structure. The catalog will get used. It deserves a serious referee, but the referee should push for the domain-shift analysis and a re-framed accuracy statement. I'd accept it conditional on those revisions.\n\nRecommendation: send to peer review.","headline":"A genuinely useful distance catalog for 36k obscured AGB stars, with an honest but under-supported extrapolation from the MSX training set that needs a quantitative check before the 6% accuracy claim is taken at face value.","tokens_in":25829,"tokens_out":2475,"would_cite":true,"duration_ms":23899,"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":"A machine-learning model trained on infrared magnitudes alone assigns statistical distances to more than 36,000 dust-obscured AGB stars, and the resulting catalog recovers the Milky Way's bar and separates young and old bulge Miras.","keywords":["Milky Way Galaxy","Asymptotic giant branch stars","Galactic structure","Infrared photometry","Machine learning","Mira variables","AKARI survey","Statistical distances"],"falsifier":"Measure VLBI parallaxes for 20-30 AKARI-selected O-rich AGB stars with SiO masers located beyond 5 kpc and compare them with the ML distances: if the median residual exceeds the quoted ~36% uncertainty or correlates with Galactic longitude, the transferability assumption is falsified.","tokens_in":24656,"feed_emoji":"🌌","tokens_out":17858,"duration_ms":153893,"temperature":0.7,"pith_summary":"This paper tries to show that statistical distances to heavily obscured oxygen-rich asymptotic giant branch (AGB) stars can be derived from raw infrared photometry alone. A gradient-boosted regression model trained on about 10,000 stars from the BAaDE survey's MSX-selected sample, whose distances come from fitting their spectral energy distributions (SEDs), reaches a test-set accuracy of about 6% and is then applied to 36,134 stars selected from the AKARI infrared satellite survey, producing a distance catalog with a combined uncertainty of about 36% over 0.5-20 kpc. The authors treat the catalog as a population-level product rather than precise per-star distances. If the distances are trustworthy, they recover the expected bar-like asymmetry of the inner Galaxy, show a bulge vertical dispersion about 30% larger than the disk, and show that long-period Miras outline the bar while short-period Miras form a smoother, older component. This matters because direct astrometry and period-luminosity relations are unreliable for dust-obscured AGB stars, so a photometric distance scale would make a large hidden population usable for mapping Galactic structure.","feed_headline":"Maps 36,000 obscured stars and finds the Milky Way's bar","feed_subtitle":"A photometry-only distance catalog lets dust-hidden giant stars trace the Galactic bar and disk.","key_machinery":"The mechanism is the empirical photometry-to-distance mapping learned by the XGBoost regressor. The feature vector is five apparent magnitudes (2MASS $J,H,K_s$ and AKARI 9 and 18 $\\mu$m) with no extinction correction; the training labels are SED-based distances for BAaDE/MSX stars, so the ML outputs inherit that distance scale. The transfer works because the AKARI-selected sources occupy similar near- and mid-IR color-magnitude space to the training sample, and the paper checks this by comparing predicted distances with independent estimates. A secondary mechanism is the dimensionless vertical dispersion $\\tilde{\\sigma}_z = \\sigma_z / Z_{\\max}$, which normalizes the measured vertical dispersion by the maximum accessible height under the latitude cut, allowing bulge and disk thickness to be compared despite the $|b| \\lesssim 6^\\circ$ selection.","core_discovery":"The central claim is that apparent multi-band infrared photometry - 2MASS $J,H,K_s$ and AKARI 9 and 18 $\\mu$m - contains enough information to predict statistical distances to oxygen-rich AGB stars in the inner Milky Way. The paper builds an XGBoost (gradient-boosted decision tree) regressor on BAaDE/MSX stars whose distances come from an earlier SED-based analysis; the model reaches $R^2 \\approx 0.98$ and a mean absolute percentage error near 6% on an independent test set, with a binned fit of predicted versus SED distance consistent with slope $1.01 \\pm 0.01$ and zero offset. Applying the model to the AKARI/IRC color-selected sample yields distances for 36,134 sources with a combined ~35.5% uncertainty. The authors validate the scale against Gaia parallaxes for high-latitude OH/IR stars, bolometric luminosities from an OH/IR catalog, and several Mira period-luminosity relations, finding broad agreement with Galactic calibrations and systematic offsets from LMC-calibrated ones that they attribute to known distance-scale and metallicity differences. They use the catalog to recover the bar's near-far distance asymmetry, measure a dimensionless vertical dispersion $\\tilde{\\sigma}_z \\approx 0.40$ in the bulge versus about 0.30 in the disk, and show that Miras with periods longer than 400 days trace the bar while shorter-period Miras do not; they explicitly decline to treat the far-side enhancement in short-period Miras as a physical result.","pith_inferences":["If the MSX-to-AKARI transfer holds, the same photometry-to-distance training strategy could be extended to other infrared surveys to reach fainter and more crowded sightlines, including the nuclear stellar disk that this sample cannot adequately cover.","The paper's Gaia validation is limited to 166 high-latitude sources outside the $|b|<5^\\circ$ footprint, so the in-plane transferability is not directly tested by Gaia; VLBI maser parallaxes for AKARI-selected stars inside the footprint would close that gap and could also settle whether the short-period Mira far-side enhancement is real.","Because pulsation period is not a model input, the period-stratified bar morphology is an emergent result, but period-correlated changes in dust-shell emission could still leak into the distances; comparing ML distances for Miras of different periods at fixed apparent magnitudes would separate a population effect from an SED-driven artifact.","Combining the catalog with line-of-sight velocities from the BAaDE survey could upgrade the projected structural maps into kinematic constraints on the bar, since density alone fixes neither the pattern speed nor the three-dimensional orientation."],"forward_implications":["The published catalog extends statistical distance coverage for dust-obscured O-rich AGB stars from about 0.5 to 20 kpc, roughly tripling the number of inner-Galaxy AGB stars with distance information compared with the MSX-based training sample.","If the distance scale is correct, the near-far asymmetry in the longitude-selected distance distributions is an independent confirmation that the Milky Way's bulge is barred, with the bar's near end at positive longitudes.","Long-period Miras ($P>400$ days) become established as tracers of a younger, more massive bar population, while short-period Miras ($P\\le 400$ days) trace a smoother, older spheroidal component.","The dimensionless vertical dispersion profile, high and flat in the bulge and lower and flat in the disk with the bulge about 30% larger, provides a population-level measure of the inner Galaxy's vertical structure that is insensitive to the adopted scale-height conversion.","Agreement with Galactic-calibrated period-luminosity relations, at the ~12% level for mid-IR relations, supports using these statistical distances as a prior or cross-check for individual Mira distances in the inner Galaxy."],"supporting_citations":[{"why":"Supplies the SED-based distance labels and the BAaDE/MSX training sample on which the entire photometry-to-distance mapping is calibrated.","marker":"R. Bhattacharya et al. 2024"},{"why":"Establishes the MSX color selection of SiO-maser-bearing AGB/Mira stars that defines the BAaDE sample.","marker":"L. O. Sjouwerman et al. 2009"},{"why":"Describes the AKARI/IRC catalog and the 9 and 18 micron bands used as model features and for source selection.","marker":"D. Ishihara et al. 2010"},{"why":"Defines the AKARI [9]-[18] color region for O-rich AGB stars and provides an earlier AKARI AGB distribution that the paper compares with its own maps.","marker":"D. Ishihara et al. 2011"},{"why":"Supplies the XGBoost gradient-boosted regression algorithm that the whole distance-prediction pipeline is built on.","marker":"T. Chen & C. Guestrin 2016"},{"why":"Provides the mid-IR period-luminosity relation for Galactic maser-bearing Miras used to validate the ML distances.","marker":"M. O. Lewis et al. 2023b"},{"why":"Supplies a Gaia-calibrated Ks-band Mira period-luminosity relation for Milky Way Miras used as an independent distance comparison.","marker":"J. L. Sanders 2023"},{"why":"Supplies the OGLE near-IR Mira period-luminosity relation and the 20-degree bar inclination used in the Mira morphology analysis.","marker":"P. Iwanek et al. 2023"},{"why":"Provides the period-dependent Mira bulge morphology and the R0 values against which the paper's period-stratified mean distances are compared.","marker":"R. M. Catchpole et al. 2016"},{"why":"Contributes the Gaia OH/IR star catalog used for the high-latitude parallax comparison and the bolometric-luminosity cross-check.","marker":"B. L. Martí et al. 2025"}],"fun_headline_variants":["Machine-learning distances to 36,000 obscured stars trace the Milky Way's bar","Photometry-only distances recover the Galactic bar from dust-hidden AGB stars","Long-period Miras trace the bar; short-period ones outline the disk","AI predicts stellar distances from IR photometry, mapping the inner Milky Way","36,000 obscured stars get distances that reveal the Galaxy's central bar"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the photometry-to-distance relation learned on the brighter, near-side-heavy MSX sample transfers directly to the fainter AKARI sample with broader far-side longitude coverage, so any undetected difference in color-magnitude space or distance distribution between the two samples would shift the distances and the structural conclusions.","fun_headline_variants_meta":{"raw":{"variants":["Machine-learning distances to 36,000 obscured stars trace the Milky Way's bar","Photometry-only distances recover the Galactic bar from dust-hidden AGB stars","Long-period Miras trace the bar; short-period ones outline the disk","AI predicts stellar distances from IR photometry, mapping the inner Milky Way","36,000 obscured stars get distances that reveal the Galaxy's central bar"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0008,"raw_usage":{"total_tokens":3633,"prompt_tokens":1174,"completion_tokens":2459,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":790,"completion_tokens_details":{"reasoning_tokens":2361}},"tokens_in":790,"tokens_out":2459,"duration_ms":17620,"temperature":1.0,"reasoning_tokens":2361,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T21:30:57.554755+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure VLBI parallaxes for 20-30 AKARI-selected O-rich AGB stars with SiO masers located beyond 5 kpc and compare them with the ML distances: if the median residual exceeds the quoted ~36% uncertainty or correlates with Galactic longitude, the transferability assumption is falsified.","supporting_citations":[{"cited_title":"M., Pihlstr¨ om, Y","cited_arxiv_id":null,"evidence_quote":"Supplies the SED-based distance labels and the BAaDE/MSX training sample on which the entire photometry-to-distance mapping is calibrated."},{"cited_title":"O., Capen, S","cited_arxiv_id":null,"evidence_quote":"Establishes the MSX color selection of SiO-maser-bearing AGB/Mira stars that defines the BAaDE sample."},{"cited_title":"2010, Astronomy & Astrophysics, 514, A1","cited_arxiv_id":null,"evidence_quote":"Describes the AKARI/IRC catalog and the 9 and 18 micron bands used as model features and for source selection."},{"cited_title":"2011, Astronomy & Astrophysics, 534, A79","cited_arxiv_id":null,"evidence_quote":"Defines the AKARI [9]-[18] color region for O-rich AGB stars and provides an earlier AKARI AGB distribution that the paper compares with its own maps."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies a Gaia-calibrated Ks-band Mira period-luminosity relation for Milky Way Miras used as an independent distance comparison."},{"cited_title":"M., Whitelock, P","cited_arxiv_id":null,"evidence_quote":"Provides the period-dependent Mira bulge morphology and the R0 values against which the paper's period-stratified mean distances are compared."}],"review_version":1}