Pith. sign in

REVIEW 1 major objections 5 minor 21 references

A morphology-blind neural network with spectral mixing beats template fitting for 79 million J-PLUS sources, and WISE cuts quasar outliers from 40% to 23%.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-10 23:47 UTC pith:ZF26F7TO

load-bearing objection Solid DR4 catalog paper: LeMoNNADE beats LePhare on a large held-out set, WISE helps QSOs, and the public products are ready to use; the EM prior shift is the softest secondary claim but is checked hard enough. the 1 major comments →

arxiv 2607.06662 v1 pith:ZF26F7TO submitted 2026-07-07 astro-ph.GA astro-ph.CO

J-PLUS: Spectral classification and photometric redshifts for 79 million sources in the fourth data release

A. Hern\'an-Caballero , H. V\'azquez Rami\'o , C. L\'opez-Sanjuan , D. Muniesa , T. Civera , J. A. Fern\'andez-Ontiveros , H. Dom\'inguez-S\'anchez , A. del Pino
show 15 more authors
J. A. L. Aguerri S. Zarattini V. Marra A. J. Cenarro D. Crist\'obal-Hornillos C. Hern\'andez-Monteagudo A. Mar\'in-Franch M. Moles J. Varela J. Alcaniz R. A. Dupke A. Ederoclite L. Sodr\'e Jr. R. E. Angulo J. Zaragoza-Cardiel
This is my paper
classification astro-ph.GA astro-ph.CO
keywords photometric redshiftsspectral classificationJ-PLUSmachine learningquasarsgalaxy surveysWISE photometrydata augmentation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper delivers spectral classifications and photometric redshifts for 79.2 million sources (r < 22) across the J-PLUS DR4 footprint. It shows that LeMoNNADE, a fully connected network trained only on aperture photometry and enlarged by spectral mixing augmentation, systematically beats a carefully tuned LePhare template fit in precision, scatter and catastrophic-outlier rate for both galaxies and quasars. Adding WISE W1/W2 photometry further breaks the optical star-quasar degeneracy, dropping the quasar outlier rate from ~40% to ~23% and holding median redshift bias within ±1% out to z = 4. Because spectroscopic training sets under-represent stars, an Expectation-Maximization prior shift is used to recalibrate class probabilities; once corrected, the extragalactic number counts match literature values down to the survey completeness limit and the redshift distribution peaks near z ~ 0.3. The catalogues therefore supply probabilistically calibrated distances and star/extragalactic probabilities for local-Universe science while flagging faint-end incompleteness and residual contamination.

Core claim

LeMoNNADE, trained exclusively on multi-band photometry and enlarged by spectral mixing augmentation, consistently outperforms optimised LePhare template fitting in f1, σ_NMAD and η_15 for J-PLUS galaxies and quasars; adding WISE infrared bands reduces the quasar catastrophic-outlier rate from ~40% to ~23% and keeps median redshift bias inside ±1% up to z = 4.

What carries the argument

LeMoNNADE: a morphology-blind, fully connected network that maps normalised multi-band SEDs to a three-component Gaussian-mixture redshift PDF (and a star probability), regularised by spectral-mixing data augmentation and adaptive variance scaling of photometric noise.

Load-bearing premise

The Expectation-Maximization prior shift fully removes the spectroscopic bias against stars so that the calibrated star probabilities, applied to the magnitude-limited catalogue, yield unbiased extragalactic counts and N(z).

What would settle it

If independent deep morphological catalogues or future complete spectroscopic samples show that the calibrated P_star values still produce extragalactic number counts or N(z) that deviate systematically from literature benchmarks below r ~ 20.5, the claim of unbiased class probabilities collapses.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Public J-PLUS DR4 catalogues now supply calibrated photo-z PDFs and star probabilities for ~79 million sources usable for local-Universe galaxy and quasar studies.
  • Median photo-z estimates from LeMoNNADE suppress the redshift-aliasing artefacts that dominate mode estimates, giving smoother N(z).
  • High-purity subsets (P_star < 0.1) already contain 8.3 million galaxies (σ_NMAD ~ 0.032) and 230 thousand quasars (σ_NMAD ~ 0.043) ready for statistical analyses.
  • Faint-end incompleteness and residual high-z contamination from spurious detections are quantified, setting practical limits for science use.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same spectral-mixing + adaptive-noise strategy should transfer directly to other multi-band surveys whose spectroscopic training sets are sparse or biased.
  • Once SPHEREx near-IR spectra become available, the remaining quasar outlier rate should drop further because the continuum and emission-line leverage will increase.
  • The residual excess of photo-z sources at z > 0.6 is largely an artefact of low-S/N detections; a simple S_det cut could clean the high-redshift tail without new spectroscopy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 5 minor

Summary. The manuscript delivers spectral classifications and photometric redshifts for 79.2 million J-PLUS DR4 sources (r < 22) over an effective area of 4437.5 deg^{2}. It compares a customised LePhare template-fitting pipeline (galaxy templates only, z ≤ 1) against LeMoNNADE, a morphology-blind fully-connected neural network that predicts GMM PDFs and binary star probabilities from aperture-corrected photometry, trained with spectral-mixing augmentation on ~2 million spectroscopic objects. LeMoNNADE outperforms LePhare in f1, σ_NMAD and η15 across magnitudes; adding unWISE W1/W2 photometry reduces the QSO catastrophic-outlier rate from ~40 % to ~23 % and keeps median bias within ±1 % up to z = 4. An EM Bayesian prior-shift (Appendix A) corrects the spectroscopic under-representation of stars, after which extragalactic counts agree with the literature to r ~ 20.5 and the N(z) for r < 21 peaks at z ~ 0.3. High-purity samples (P_star < 0.1) of 8.34 × 10^6 galaxies and 2.30 × 10^5 QSOs are released with estimated σ_NMAD ~ 0.032 and 0.043.

Significance. The work supplies a large, publicly available, probabilistically calibrated catalogue that is immediately useful for local-Universe science. Strengths include a held-out spectroscopic test set of ~83 000 objects, explicit magnitude- and odds-dependent metrics (Figs. 4–12), transparent treatment of aliasing and PDF point-estimate choice, and thorough internal validation of the EM prior shift (Figs. A.1–A.3) plus the high-z excess (Appendix C). The morphology-blind design and spectral-mixing augmentation are well-motivated engineering choices that improve robustness against PSF variations and training-set gaps. The dual LePhare/LeMoNNADE release and the ColdPress PDF compression further increase the catalogue’s utility.

major comments (1)
  1. The primary performance claims (LeMoNNADE superiority on the spectroscopic test set and the WISE-driven QSO gains) rest on direct metrics that do not depend on the EM calibration and are solidly demonstrated. The secondary claim that calibrated extragalactic counts and N(z) are unbiased down to r ~ 20.5 does rely on the EM prior shift (Appendix A). The paper already supplies multiple internal checks (raw vs calibrated stellar fractions, flat extragalactic density vs Galactic latitude, agreement with Yasuda/Koushan after extinction correction, HSC morphology cross-match). These checks make the assumption controlled rather than unexamined; no additional load-bearing revision is required, but a short quantitative statement of residual stellar contamination after calibration (e.g., the ~3.5 % figure already given for |b| < 15°) should be elevated into the main text of §6.1 for clarity.
minor comments (5)
  1. §2.1: the three pre-processing steps applied only to LePhare photometry are clearly stated, but a one-sentence reminder that LeMoNNADE deliberately omits Galactic-extinction correction (because it over-corrects stars) would help readers who skip the later discussion.
  2. Fig. 3 caption and §5.1: the aliasing discussion is excellent; a brief note that z_random is the only estimator that statistically recovers the population PDF would make the later use of z_random for N(z) more self-contained.
  3. Appendix B: the ~5–12 % over-estimate of nominal photometric errors is useful; stating the exact correction factors per band (or a table) would aid reproducibility.
  4. Typographical: “que quantify” → “we quantify” (p. 8); “analised” → “analysed” (p. 2); occasional missing spaces around ~ and ± symbols.
  5. Data-availability statement is clear and complete; the ColdPress GitHub link is appreciated.

Circularity Check

0 steps flagged

No significant circularity: performance metrics and population statistics are evaluated against held-out external spectroscopy and independent literature counts.

full rationale

The central claims (LeMoNNADE superiority over LePhare on f1/σ_NMAD/η15, WISE-driven reduction of QSO outliers, and calibrated extragalactic N(z)/counts) rest on supervised evaluation against a held-out spectroscopic test set drawn from DESI/SDSS/HectoMAP/HETDEX and on post-hoc comparison of the magnitude-limited catalogue to external benchmarks (Yasuda et al. 2001, Koushan et al. 2021, SHELS, COMBO-17, HSC morphology). The EM prior-shift (Appendix A, Eq. A.1) is an unsupervised re-weighting of class priors that does not force agreement with those benchmarks; agreement is an empirical outcome that is further stress-tested by Galactic-latitude flatness, extinction-corrected counts, and HSC extendedness cross-matches (Appendix C). No equation reduces a claimed prediction to a quantity defined by the same data used to fit it, no uniqueness theorem is imported from the authors, and self-citations (Hernán-Caballero et al. 2021, ColdPress, LeMoNNADE in prep.) supply only methodological scaffolding, not load-bearing results. The derivation chain is therefore self-contained.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 2 invented entities

The central performance claims rest on standard photometric-redshift practice plus a handful of tunable hyperparameters and the assumption that the EM prior shift recovers the true stellar fraction. No new physical entities are postulated; the invented pieces are algorithmic.

free parameters (6)
  • z_scale = 0.005
    Width of the redshift window used in the GMM loss; set to 0.005 after experimentation on the validation set.
  • P_floor = 0.01
    Floor probability added to the loss to protect against label noise; chosen as 0.01.
  • GMM components = 3
    Number of Gaussians in the output PDF model; fixed at three.
  • inference iterations / adaptive variance scaling = 100 iterations, scale factor fitted
    100 Monte-Carlo flux perturbations per source whose variance is scaled empirically so that the final PDFs are statistically calibrated.
  • LePhare prior parameters (α, γ, absolute-magnitude cuts) = optimized on training set
    Contrast-correction parameters and M_i limits for early/intermediate/late types are optimized on the spectroscopic sample.
  • EM magnitude-bin width and convergence threshold = 0.5 mag, 1e-6
    0.5-mag bins and |Δπ|<10^{-6} used for the Saerens-style prior shift.
axioms (5)
  • domain assumption Spectroscopic redshifts and classifications from DESI/SDSS/HectoMAP/HETDEX are sufficiently pure that they can serve as ground truth after quality cuts.
    Section 2.3; residual ~0.1% label impurity is acknowledged but treated as negligible.
  • domain assumption Aperture-corrected 3-arcsec photometry (scaled to AUTO total flux) preserves colors while supplying the correct luminosity prior.
    Section 2.1; justified by earlier J-PLUS calibration papers.
  • ad hoc to paper Linear spectral mixing of normalized SEDs at fixed redshift produces physically plausible training examples that fill color-redshift gaps without introducing bias.
    Section 4.3; core of the mixaugmentation strategy.
  • domain assumption The Saerens EM procedure recovers the true stellar prior of the magnitude-limited sample from the biased spectroscopic prior.
    Appendix A; standard technique but its success is verified only by post-hoc latitude and count comparisons.
  • domain assumption Summing (or Monte-Carlo sampling) well-calibrated individual PDFs yields the true population N(z).
    Section 6.2; standard photo-z assumption, tested against SHELS and literature counts.
invented entities (2)
  • LeMoNNADE independent evidence
    purpose: Morphology-blind neural network that outputs both star probability and a three-component GMM photo-z PDF.
    Introduced in Section 4; performance is demonstrated on held-out data, so independent_evidence is true in the sense of external spectroscopic validation.
  • Spectral mixing augmentation no independent evidence
    purpose: Generate synthetic SEDs by linear combination of real training SEDs at the same redshift to combat sparse sampling of color-redshift space.
    Section 4.3; the technique is algorithmic and validated by equal train/validation loss, but has no external physical prediction.

pith-pipeline@v1.1.0-grok45 · 29703 in / 3352 out tokens · 41864 ms · 2026-07-10T23:47:46.359761+00:00 · methodology

0 comments
read the original abstract

We present spectral classifications and photometric redshifts for 79.2 million sources up to an r-band magnitude of 22 in Data Release 4 of the Javalambre Photometric Local Universe Survey (J-PLUS). Leveraging the 12-band J-PLUS filter system, we compare a template-fitting approach (LePhare) against LeMoNNADE, a morphology-blind machine learning pipeline that uses spectral mixing augmentation to overcome training set limitations. LeMoNNADE consistently outperforms template fitting in precision, robust scatter, and outlier rates. Including WISE infrared photometry breaks optical degeneracies between stars and quasars, reducing the catastrophic outlier rate for quasars from ~40% to ~23% and constraining systemic redshift bias to <1% up to z = 4. We find LeMoNNADE is also less susceptible to redshift aliasing, particularly when adopting the probability density function median. Because the spectroscopic training samples severely under-represent stars, we apply an Expectation-Maximization Bayesian calibration to recover unbiased class probabilities for the magnitude-limited sample. This reveals that extragalactic counts agree with the literature down to the r ~ 20.5 completeness limit. The inferred redshift distribution for r < 21 extragalactic sources peaks at z ~ 0.3, showing broad agreement with existing literature up to z ~ 0.6. The resulting catalogues represent a significant milestone for local Universe science, offering probabilistically calibrated classifications and distances while explicitly characterising faint-end limits and contamination.

Figures

Figures reproduced from arXiv: 2607.06662 by A. del Pino, A. Ederoclite, A. Hern\'an-Caballero, A. J. Cenarro, A. Mar\'in-Franch, C. Hern\'andez-Monteagudo, C. L\'opez-Sanjuan, D. Crist\'obal-Hornillos, D. Muniesa, H. Dom\'inguez-S\'anchez, H. V\'azquez Rami\'o, J. A. Fern\'andez-Ontiveros, J. A. L. Aguerri, J. Alcaniz, J. Varela, J. Zaragoza-Cardiel, L. Sodr\'e Jr., M. Moles, R. A. Dupke, R. E. Angulo, S. Zarattini, T. Civera, V. Marra.

Figure 1
Figure 1. Figure 1: Magnitude distribution for galaxies, quasars and stars in the J￾PLUS DR4 subsample with spectroscopy. The notable increase in star counts in the range 16.2 < r < 19.2 is due to the selection criteria of the DESI Milky Way Survey (Cooper et al. 2023). 2.2. WISE photometry As discussed in Section 5.5, the J-PLUS photometry alone (here￾after the jplusonly dataset) presents challenges for LeMoN￾NADE in separat… view at source ↗
Figure 2
Figure 2. Figure 2: Cumulative fraction of sources with redshift error |∆z| = |zphot￾zspec|/(1+zspec) smaller than a given threshold for different point esti￾mates. The sample consists of galaxies in the jplus+wise test set with zspec < 1. Carlo sample drawn randomly from the cumulative distribution (zrandom). The choice of the optimal point estimate depends on the intended application. As shown in [PITH_FULL_IMAGE:figures/f… view at source ↗
Figure 3
Figure 3. Figure 3: Redshift distributions derived from different PDF point estimates compared to the spectroscopic distribution (grey shaded area) for r<21 galaxies in the test sample. 5.2. Photo-z accuracy for galaxies We compare the performance of the template-fitting code LeP￾hare against the machine-learning pipeline LeMoNNADE. To ensure a fair comparison, we restrict the evaluation set to the subset of the test sample w… view at source ↗
Figure 4
Figure 4. Figure 4: Photometric vs spectroscopic redshift for z < 1 galaxies in the test sample. The colour coding indicates the logarithmic density of galaxies in bins of 0.01 × 0.01. The solid diagonal marks the 1:1 relation, while dashed lines indicate the catastrophic outlier threshold |∆z| = 0.15. 0.0 0.2 0.4 0.6 0.8 f1 LePhare LeMoNNADE (jplusonly) LeMoNNADE (jplus+wise) 0.00 0.02 0.04 0.06 0.08 σNMAD 16 17 18 19 20 21 … view at source ↗
Figure 5
Figure 5. Figure 5: Magnitude dependence of the f1 metric (top), σNMAD (middle), and η15 (bottom) for LePhare and the two LeMoNNADE configura￾tions. Solid symbols represent the full sample; open symbols corre￾spond to the 50% of sources with the highest odds in each magnitude bin [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Observed outlier rate as a function of the mean odds parameter. The dotted line represents the theoretical expectation for perfectly calibrated probabilities. 0.0 0.2 0.4 0.6 0.8 f1 galaxies QSOs 0.00 0.02 0.04 0.06 0.08 σNMAD 16 17 18 19 20 21 22 r 0 5 10 15 η15 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Photometric versus spectroscopic redshift for QSOs in the test sample. The colour coding indicates the odds parameter correspond￾ing to the point estimate zmedian. The solid diagonal marks the 1:1 re￾lation, while dashed lines indicate the catastrophic outlier threshold |∆z| = 0.15. erties of the full J-PLUS extragalactic population using LeMoN￾NADE with the jplus+wise configuration. Applying models traine… view at source ↗
Figure 7
Figure 7. Figure 7: Magnitude dependence of the f1 metric (top), σNMAD (mid￾dle), and η15 (bottom) for the photometric redshifts of QSOs (triangles) and galaxies (circles) from the spectroscopic test sample, obtained with LeMoNNADE in the jplus+wise configuration. Solid symbols repre￾sent the full sample; open symbols correspond to the 50% of sources with the highest odds in each magnitude bin. 6. Properties of the J-PLUS ext… view at source ↗
Figure 9
Figure 9. Figure 9: Redshift dependence of the photometric redshift bias, measured in bins of width 0.01(1+z). Broken lines represent the median ∆z within each bin, excluding catastrophic outliers, while shaded regions denote the corresponding 1-σ confidence intervals. 16 17 18 19 20 21 22 r 0.0 0.2 0.4 0.6 0.8 1.0 hPstari stars galaxies QSOs [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Classification uncertainty and sample abundance by spectral type. Top panel: Mean probability of being a star, ⟨Pstar⟩, as a function of r-band magnitude for spectroscopically confirmed stars, differentiated by spectral class. Bottom panel: Number of sources per magnitude bin for each spectral class in the spectroscopic training sample. 10−4 10−3 10−2 10−1 1 2 1 − 10−11 − 10−21 − 10−3 mean predicted Pstar… view at source ↗
Figure 10
Figure 10. Figure 10: Mean Pstar as a function of r-band magnitude for stars (green), galaxies (red), and QSOs (blue). Open and solid symbols denote the jplusonly and jplus+wise datasets, respectively. sequent analysis on a randomly selected subsample comprising 10% of the data2 . 6.1. Number density In [PITH_FULL_IMAGE:figures/full_fig_p009_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: Calibration test for Pstar values: fraction of spectroscopic stars versus predicted Pstar. rate of spurious detections (see Appendix C). The excluded tiles represent 20% of the total J-PLUS DR4 footprint. The extinction-corrected J-PLUS extragalactic counts show excellent agreement with the SDSS galaxy counts from Yasuda et al. (2001), except in the 16 < r < 18 interval where SDSS counts are slightly high… view at source ↗
Figure 13
Figure 13. Figure 13: Differential number counts of J-PLUS extragalactic sources as a function of extinction-corrected r-band magnitude. For comparison, we include galaxy counts from SDSS (Yasuda et al. 2001) and Koushan et al. (2021). Additional reference counts are derived from McCracken et al. (2003) (CFH12K-VIRMOS deep field), Kashikawa et al. (2004) (Subaru Deep Field), and Capak et al. (2004) (Hawaii HDF-N). The referenc… view at source ↗
Figure 14
Figure 14. Figure 14: Differential number counts as a function of redshift for the full r < 21 J-PLUS sample, obtained by weighting each source by its probability of being extragalactic. Top panel: The z < 1 distributions obtained from the full PDFs (zrandom) from LePhare (thin solid line) and LeMoNNADE (using the jplus+wise dataset, thick solid line). For comparison, counts from the spectroscopic SHELS survey are also shown. … view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages · 3 internal anchors

  1. [1]

    2022, ApJS, 259, 2,

    Abdurro’uf, Accetta, K., Aerts, C., et al. 2022, ApJS, 259, 2,

    work page 2022

  2. [2]

    Aihara, H., Arimoto, N., Armstrong, R. et al. 2018, PASJ, 70, S4 Aihara, H., AlSayyad, Y ., Ando, M., et al. 2022, PASJ, 74, 247 Arnouts, S., & Ilbert, O. 2011, LePHARE: Photometric Analysis for Redshift Estimate, ascl:1108.009 Benítez, N., Dupke, R., Moles, M., et al. 2014, arXiv e-prints, arXiv:1403.5237 Bertin, E. & Arnouts, S. 1996, A&AS, 117, 393 Ber...

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  3. [3]

    2021, A&A, 653, A31 Boquien, M., Burgarella, D., Roehlly, Y ., et al

    Bonoli, S., Marín-Franch, A., Varela, J., et al. 2021, A&A, 653, A31 Boquien, M., Burgarella, D., Roehlly, Y ., et al. 2019, A&A, 622, A103 Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682 Capak, P., Cowie, L. L., Hu, E. M., et al. 2004, AJ, 127, 1,

    work page 2021

  4. [4]

    The Pan-STARRS1 Surveys

    Cenarro, A. J., Moles, M., Marín-Franch, A., et al. 2014, Proceedings of the SPIE, 9149, 91491I. Cenarro, A. J., Moles, M., Cristóbal-Hornillos, D., et al. 2019, A&A, 622, A176. Chambers, K. C., Magnier, E. A., Metcalfe, N., et al. 2016, , arXiv:1612.05560. Cooper, A. P., Koposov, S. E., Allende Prieto, C., et al. 2023, ApJ, 947, 1,

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  5. [5]

    Data Release 1 of the Dark Energy Spectroscopic Instrument

    Cristóbal-Hornillos, D., Gruel, N., Varela, J., et al. 2012, Proceedings of the SPIE, 8451, 845116 del Pino, A., López-Sanjuan, C., Hernán-Caballero, A., et al. 2024, A&A, 691, A221. DESI Collaboration, Abdul-Karim, M., Adame, A. G., et al. 2025, , arXiv:2503.14745. Geller, M. J., Hwang, H. S., Fabricant, D. G., et al. 2014, ApJS, 213, 2,

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  6. [6]

    M., Schlafly, E

    Green, G. M., Schlafly, E. F., Finkbeiner, D., et al. 2018, MNRAS, 478, 651 Hernán-Caballero, A., Varela, J., Lopez-Sanjuan, C. et al. 2021, A&A, 654, 101 Hernán-Caballero, A. 2025, Research Notes of the American Astronomical So- ciety, 9, 7,

    work page 2018

  7. [7]

    2012, MNRAS, 421, 3,

    Hildebrandt, H., Erben, T., Kuijken, K., et al. 2012, MNRAS, 421, 3,

    work page 2012

  8. [8]

    J., et al

    Ilbert, O., Arnouts, S., McCracken, H. J., et al. 2006, A&A, 457, 841 Kashikawa, N., Shimasaku, K., Yasuda, N., et al. 2004, PASJ, 56,

    work page 2006

  9. [9]

    P., Bellstedt, S., et al

    Koushan, S., Driver, S. P., Bellstedt, S., et al. 2021, MNRAS, 503, 2,

    work page 2021

  10. [10]

    2005, A&A, 439, 845 López-Sanjuan, C., Tremblay, P.-E., Ederoclite, A., et al

    Le Fèvre, O., Vettolani, G., Garilli, B., et al. 2005, A&A, 439, 845 López-Sanjuan, C., Tremblay, P.-E., Ederoclite, A., et al. 2022, A&A, 658, A79. López-Sanjuan, C., Vázquez Ramió, H., Xiao, K., et al. 2024, A&A, 683, A29. Marin-Franch, A., Taylor, K., Cenarro, J., et al. 2015, IAU General Assembly, 29, 2257381. McCracken, H. J., Radovich, M., Bertin, E...

    work page 2005

  11. [11]

    2023, ApJ, 943, 2,

    Mentuch Cooper, E., Gebhardt, K., Davis, D., et al. 2023, ApJ, 943, 2,

    work page 2023

  12. [12]

    T., Strauss, M

    Richards, G. T., Strauss, M. A., Fan, X., et al. 2006, AJ, 131, 6,

    work page 2006

  13. [13]

    Adjusting the Outputs of a Classifier to New a Priori Probabilities: A Simple Procedure,

    Saerens, M., Latinne, P., & Decaestecker, C. 2002, “Adjusting the Outputs of a Classifier to New a Priori Probabilities: A Simple Procedure,”Neural Com- putation, 14, 21–41. Schlafly, E. F., Meisner, A. M., Stutz, A. M., et al. 2016, ApJ, 821, 78 Schlafly, E. F., Meisner, A. M., & Green, G. M. 2019, ApJS, 240, 2,

    work page 2002

  14. [14]

    J., Hwang, H

    Sohn, J., Geller, M. J., Hwang, H. S., et al. 2023, ApJ, 945, 2,

    work page 2023

  15. [15]

    2024, MNRAS, 527, 2,

    von Marttens, R., Marra, V ., Quartin, M., et al. 2024, MNRAS, 527, 2,

    work page 2024

  16. [16]

    D., Placco, V

    Whitten, D. D., Placco, V . M., Beers, T. C., et al. 2019, A&A, 622, A182 Wolf, C., Wisotzki, L., Borch, A., et al. 2003, A&A, 408,

    work page 2019

  17. [17]

    2009, MNRAS, 397, 1,

    Wolf, C. 2009, MNRAS, 397, 1,

    work page 2009

  18. [18]

    2026, Nature Scientific Data, 13, 1,

    Xiao, K., Huang, Y ., Yuan, H., et al. 2026, Nature Scientific Data, 13, 1,

    work page 2026

  19. [19]

    K., et al

    Yasuda, N., Fukugita, M., Narayanan, V . K., et al. 2001, AJ, 122, 3,

    work page 2001

  20. [20]

    G., Adelman, J., Anderson, John E., J., et al

    York, D. G., Adelman, J., Anderson, John E., J., et al. 2000, AJ, 120, 1579 1 Centro de Estudios de Física del Cosmos de Aragón (CEFCA), Plaza San Juan 1, 44001 Teruel, Spain 2 Unidad Asociada CEFCA-IAA, CEFCA, Unidad Asociada al CSIC por el IAA y el IFCA, Plaza San Juan 1, 44001 Teruel, Spain 3 Instituto de Física de Cantabria, Av. de los Castros, 39005 ...

    work page 2000

  21. [21]

    Such a distortion would arise if the classification un- 0.0 0.2 0.4 0.6 0.8 1.0 zmedian 101 102 103 counts / bin HSC gri extended Pstar < 0.5 Pstar < 0.1 Pstar weighted Fig

    could bias the inferred redshift distribution, even if thePstar values are statistically well- calibrated. Such a distortion would arise if the classification un- 0.0 0.2 0.4 0.6 0.8 1.0 zmedian 101 102 103 counts / bin HSC gri extended Pstar < 0.5 Pstar < 0.1 Pstar weighted Fig. C.1.Photometric redshift distribution forr<21 galaxies in the JPSV field der...

    work page 2018