Pith. sign in

REVIEW 4 major objections 6 minor 39 references

Young-star variability fingerprints form a stable continuum, allowing model light curves to be placed and judged against 240 real stars.

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 ↗

Principal component analysis of variability fingerprints creates a stable, continuous map of young star light curves, where the main axis measures when large brightness changes begin.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection Solid methods paper on PCA of variability fingerprints; the robust-comparison claim is only tested for one-at-a-time insertions, not for the model populations it is meant to compare. the 4 major comments →

arxiv 2509.07710 v1 pith:SF2NM2WP submitted 2025-09-09 astro-ph.SR astro-ph.GAastro-ph.IM

A survey for variable young stars with small telescopes: X -- Comparing stochastic YSO light curve

Benjamin W. Ryan , Holly Stokes-Geddes , Dirk Froebrich This is my paper
classification astro-ph.SR astro-ph.GAastro-ph.IM
keywords young stellar objectsstochastic variabilityvariability fingerprintsprincipal component analysispre-main-sequence starsphotometric monitoringT Tauri variabilitylight-curve classification
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 reading

This paper develops a way to compare stochastic light curves of young stars with each other and with model-generated curves, without flattening a light curve to a single number. Each light curve is converted into a 'variability fingerprint' — a two-dimensional probability map of brightness change versus time lag. Applying principal component analysis to these fingerprints yields a stable two-dimensional landscape in which 240 highly variable young stars form a continuous distribution with no discrete clusters. The stability means a synthetic model light curve can be added to the observed sample and its position read as a statistical test of whether the model reproduces real YSO variability. The main axis of the landscape is identified as the timescale on which variability above 0.3 mag sets in — most critically 1 to 3 months — with long-term dimming or brightening over 1.5 years as the secondary axis.

Core claim

Using light curves of 240 highly variable young stars observed over up to ten years, the paper constructs variability fingerprints — maps of the probability of a brightness change Δm over a time lag Δt — and applies PCA to the 144 pixels of each fingerprint. The first two principal components capture under half the variance, and clustering metrics plus visual structure show a continuum, not discrete clusters. Adding one model-generated fingerprint with bootstrapped photometry, random phases, and shuffled cadences leaves the original points nearly fixed, while t-SNE-based clustering shifts the landscape. The loadings show the main variance axis is the timescale at which variability above 0.3

What carries the argument

The variability fingerprint is the central object: a two-dimensional histogram of all pairs of observations, with columns in log time lag (1 day to ~8.7 yr) and rows in adaptive magnitude-change bins (±0.05 to ±1.8 mag), normalized column-wise so each pixel gives the probability P(Δt, Δm) that a star changes by Δm over a lag Δt. Principal component analysis with standard scaling reduces the 144-pixel fingerprint to two components that define the 'fingerprint landscape' for a sample. The loadings matrices — the weights each fingerprint pixel contributes to PC1 and PC2 — let the authors read physical meaning back out of the landscape: PC1 tracks the onset timescale of >0.3 mag variability, and

Load-bearing premise

Every light curve is long enough and densely sampled enough that all variability timescales and amplitudes that matter are actually seen; the paper itself notes this fails for rare bursts and long dimming events.

What would settle it

Truncate each observed light curve to a two-year window and recompute the PCA landscape and loadings; if the 1–3 month peak in PC1 weakens or model placements shift by more than the typical nearest-neighbour distance, the landscape and the timescale conclusion depend on the ten-year baseline rather than on intrinsic variability.

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

If this is right

  • Model light curves can be fingerprinted and placed into the observed PCA landscape; if a model falls inside the observed continuum it is statistically consistent with real YSO variability, and if it falls outside, its variability is not seen in the sample.
  • A single added or replaced object does not materially shift the landscape, so comparisons do not require re-clustering the full sample each time.
  • The dominant variance axis means observations and models should be designed to resolve the 1–3 month onset of >0.3 mag variability; this timescale, not overall amplitude, is what most separates highly variable YSOs.
  • Because the sample forms a continuum, discrete YSO variability classes along these axes are conveniences rather than separated populations.
  • Photometric errors, timing, and observing cadence add only small scatter to a model's placement, so a model mismatch in the landscape likely reflects real variability differences rather than observing artifacts.

Where Pith is reading between the lines

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

  • If 1–3 months is truly the critical onset timescale, then short monitoring campaigns of under about six months will misplace objects on PC1; future surveys should favour multi-season baselines over dense single-season runs.
  • The continuum result suggests some previously reported YSO variability classes may be projections of a single smoothly graded distribution; applying the same fingerprint-plus-PCA pipeline to other photometric surveys would test this.
  • The same stable-landscape procedure could be applied to other stochastic variables, such as AGN or FUor outbursts, to compare observed variability statistics with physical models without imposing cluster structure.
  • Adding colour information as extra fingerprint dimensions might separate extinction-driven from accretion-driven variability, which are currently blended in PC1 and PC2.
Share X Bluesky LinkedIn Reddit HN

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

4 major / 6 minor

Summary. Using 10 years of HOYS photometry for 240 highly variable young stellar objects, the authors construct two-dimensional variability fingerprints - probabilities of magnitude change over time lag - and apply PCA to compare the resulting high-dimensional distributions. They report that the fingerprint landscape is a continuum with no discrete clusters, that PCA (unlike t-SNE/DBSCAN) yields a stable landscape under single-object modifications and photometric bootstrapping, and that synthetic sine/dipper/burster light curves occupy compact, interpretable regions of the landscape. From the PCA loadings they conclude that the primary variance axis is the timescale of onset of significant (>0.3 mag) variability, with 1-3 month timescales dominating, and the secondary axis is long-term (>1.5 yr) brightening or fading. The paper's stated goal is a robust quantitative method to compare observed stochastic YSO variability with model-generated light curves.

Significance. The paper makes a useful methodological contribution to a real problem: comparing inhomogeneously sampled, aperiodic YSO light curves with simulations. Its strengths are the careful treatment of fingerprint uncertainties (30,000 bootstrap iterations validated against Poisson statistics), the use of multiple clustering diagnostics (DBI, silhouette, outlier fractions) before concluding a continuum, and the explicit stability tests with synthetic light curves. Public data availability supports verification. If the stability claims are made quantitative and extended to the intended model-population comparison, the method will be a valuable tool. The physical interpretation of PC1/PC2 is interesting but currently more qualitative than the abstract implies.

major comments (4)
  1. [§4.4, Figs. 8-9] Stability is demonstrated for one-at-a-time insertion of a single synthetic fingerprint into the 240-object sample. The abstract and §4.4 motivate the method as a way to compare model-generated light curves with observations 'to assess statistical realism.' A real model comparison will typically require inserting a population of stochastic model fingerprints; when many model light curves are added and PCA is refit on the combined sample, the eigenvectors can rotate and the coordinates of all observed objects can shift. The current tests do not bound this effect because the injected signals are deterministic, periodic, and (as §3.1 concedes) do not include rare bursts or long dimming events - precisely the outliers to which PCA is sensitive. Please test with ensembles of stochastic model light curves added at a range of number fractions and quantify axis rotation and object displacement.
  2. [Abstract and §4.5, Fig. 10] The physical interpretation of PC1/PC2 rests on visual inspection of the loadings matrices. The claim that the 'largest contributions' to PC1 occur at 1-3 month timescales and that PC2 is dominated by behaviour above 1.5 yr is not supported by any quantitative summary or uncertainty estimate. Please report, for example, integrated absolute loadings within timescale bands, bootstrap or jackknife confidence regions for the loadings, and, ideally, validation against light curves with known morphology. This is needed to make the abstract's physical conclusion reliable.
  3. [Abstract, §4.3.2, §4.4.1] The phrase 'does not significantly alter the distribution' is used without a quantitative criterion. The evidence is visual ('changes ... not change by more than a small fraction', 'significantly smaller than the overall spread') and via plotted ellipses. Please define a stability metric with a pre-specified threshold - e.g. median displacement of observed points in units of local nearest-neighbour distance, or a Procrustes comparison of the PC axes before and after insertion - and report it for all tests. This would make the central claim falsifiable and easier to assess.
  4. [§3.1, §4.4.2] The fingerprint assumption that the observing baseline covers all relevant timescales and amplitudes is acknowledged to fail for rare events. The synthetic tests, however, use signals that repeat within the 10-year window (sine waves; dippers every 150 d; bursts every 2 yr). Therefore the conclusion that 'photometric uncertainties, timing, and observing cadence have minimal impact on model placement' does not yet cover the rare-event component of stochastic models. This limitation should be stated explicitly in the conclusion, or tested with realistic rare-event light curves; otherwise the robustness claim is broader than the evidence.
minor comments (6)
  1. [§4.4.1, §4.5] Use 'principal component' instead of 'principle component' in several places.
  2. [§2.3] The percentages '8,5, and 6%' are difficult to parse; write 8%, 5%, and 6%.
  3. [§3.2] The statement that values of X 'between zero and a few' do not significantly influence results is vague; specify the range actually tested or remove the claim.
  4. [Table 1] The representation of periods as reciprocal fractions of π yr (e.g., 1/(3.8π) yr) is confusing; give decimals or conventional units.
  5. [§4.1] The sentence 'A DBI of under 0.40 suggest highly compact clusters' has a grammatical error; also clarify whether 0.45 is close enough to the threshold to matter given the silhouette score.
  6. [Data availability] The paper would benefit from a statement on code availability; the data are public, but the fingerprint/PCA pipeline is described in text only.

Circularity Check

0 steps flagged

No significant circularity: the PCA fingerprint landscape and its interpretation are data-driven, and no fitted parameter or self-citation chain forces the conclusions.

full rationale

The paper's derivation chain is: light curves -> variability fingerprints (defined in Sect. 3.1 via pair-counting and column normalization) -> PCA (Sect. 4.3) -> interpretation of loadings (Sect. 4.5). Each step is either a definition or an empirical computation, and no step defines a quantity in terms of the target conclusion. The stability tests in Sect. 4.4 insert synthetic fingerprints with fixed parameters (sine waves, dippers, bursters) and measure how much existing points move; the injected fingerprints are not fitted to preserve the observed distribution, so the stability result is not forced by construction. The loadings interpretation (PC1 = onset timescale of >0.3 mag variability, PC2 = long-term trends) is an empirical description of the PCA loadings, not a redefinition of the loadings to match the claim. Self-citations (Evitts et al. 2020; Froebrich et al. 2022) supply the fingerprint algorithm and photometric calibration, but the algorithm is fully described in this paper and its robustness is tested here; this is standard method reuse, not an unverified load-bearing self-citation. The acknowledged limitation about rare events (Sect. 3.1) affects generalizability but does not make the derivation circular. The skeptic's concern that the stability test only addresses single insertions while model-population comparisons might rotate the PCA axes is a scoping/correctness issue, not a circularity issue, and therefore does not raise the circularity score.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

No new physical entities are introduced. The free parameters are choices in the analysis pipeline; the paper argues the results are robust to these choices, but the insensitivity is not exhaustively proven. The axioms are domain assumptions inherited from the fingerprint methodology and the sample selection.

free parameters (3)
  • Adaptive Delta m bin scaling X = X=1
    Controls the width of magnitude bins in the fingerprint; paper states results are insensitive to X between 0 and a few, but it is a hand-chosen value.
  • Fingerprint resolution = 9 x 16 adaptive pixels (V band)
    Results are presented for this configuration; other resolutions give similar results, but it is a specific choice.
  • Welch-Stetson index threshold = I=2 in all three filter pairs
    Threshold used to select the 240 highly variable YSOs; paper asserts the specific choice does not affect subsequent results.
axioms (4)
  • domain assumption The total length and cadence of each light curve are sufficient to sample all relevant variability timescales and amplitudes.
    Explicitly stated in Sect. 3.1 as an assumption of the fingerprint method; if violated, fingerprints are incomplete and the landscape would be biased.
  • domain assumption The pair-difference histogram (fingerprint) is a valid statistical representation of stochastic YSO variability.
    Basis of the method, built on Evitts et al. (2020) and earlier works; the paper does not independently justify this representation.
  • domain assumption Photometric calibration and outlier rejection do not introduce systematic biases in the fingerprints.
    Relies on calibration from non-variable stars and CMD outlier removal; no independent validation of the cleaned sample is provided.
  • domain assumption The 240 selected highly variable YSOs are representative of the broader population of variable YSOs for defining the landscape.
    PCA loadings and interpretations are derived from this sample only; generalization to all YSOs is assumed.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of A survey for variable young stars with small telescopes: X -- Comparing stochastic YSO light curve." pith.science (2026). https://pith.science/paper/SF2NM2WP

@misc{pith2026250907710,
  author       = {Pith},
  title        = {Pith review of: A survey for variable young stars with small telescopes: X -- Comparing stochastic YSO light curve},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SF2NM2WP}},
  note         = {Machine review of arXiv:2509.07710}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Light curves of young stars exhibit photometric variability over hours to decades and across a wide range of amplitudes. On time scales beyond a few rotation periods, these light curves are typically stochastic. The variability arises from a combination of accretion rate changes, line-of-sight extinction variations, and evolving spotted stellar surfaces. We aim to develop a methodology to quantitatively compare the full variability statistics of these inhomogeneously sampled light curves with model calculations. To achieve this, we converted the light curves into variability fingerprints. They map the probability of variation by a given amount over a given timescale. Applying principal component analysis to these fingerprints produces a stable distribution of the first two principal components. We show that this distribution is a continuum without clusters. Adding a model-generated fingerprint to an observational sample does not significantly alter the distribution of the sample, allowing a robust comparison between the model and observed light curves to assess statistical realism. We show that photometric uncertainties, timing, and observing cadence have a minimal impact on model placement within the observational distribution. The main source of variance among highly variable light curves of young stars is the timescale of the onset of significant variability (above 0.3mag), with 1-3month timescales being the most critical. The secondary cause of variance are long-term (above 1.5yr) dimming or rising trends.

Figures

Figures reproduced from arXiv: 2509.07710 by Benjamin W. Ryan, Dirk Froebrich, Holly Stokes-Geddes.

Figure 1
Figure 1. Figure 1: Left: Example light curve of the variable source FHK 116. We show the 𝑉 (green triangles), 𝑅 (red diamonds), and 𝐼-band (black circles) light curves. The points highlighted in blue have been flagged up as potentially erroneous and have been removed from any subsequent analysis. Right: Example 𝑉 − 𝑅 vs 𝑉 colour magnitude plot for the same source. The best fitting slope is over plotted and the 3-sigma outlie… view at source ↗
Figure 2
Figure 2. Figure 2: Welch-Stetson index of our YSO sample determined from all three filter combinations, I𝑉,𝑅, I𝑉,𝐼 , and I𝑅,𝐼 from left to right. The cut-off for highly variable objects used in this study is marked at I = 2 with a solid line. We also indicate I = 1 and 0.5 with a dashed and dotted line, respectively. 3 VARIABILITY FINGERPRINTS In this section, we describe how our YSO light curves are converted into variabili… view at source ↗
Figure 3
Figure 3. Figure 3: Example 𝑉-band variability fingerprints of the source FHK 116. From left to right we show the resolutions 40 x 40 , 20 x 20, and 9 x 16 adaptive pixels. The range along the time axis are identical in all panels, but the adaptive pixel map on the right has a larger magnitude range than the other two maps. variability range from −1.8 mag to +1.8 mag with only 16 of these adaptive pixels. The range and pixel … view at source ↗
Figure 4
Figure 4. Figure 4: Example 𝑉-band variability fingerprint error maps of the source FHK 116. From left to right we show the resolutions 40 x 40 , 20 x 20, and 9 x 16 adaptive pixels. The top row shows the maps determined from bootstrapping the light curve, and the bottom row when using the Poisson counting statistics. In total only about 10 percent of pixels have such a low signal-to￾noise, and these are exclusively confined … view at source ↗
Figure 5
Figure 5. Figure 5: Left: DBSCAN clustering results for the 9 x 16 𝑉-band adaptive pixel fingerprints of 240 variable YSOs. Right: The same data and process as in the left panel but with one model fingerprint added which corresponds to a sine wave like light curve with an amplitude of 1 mag and a period of 2 yr. The minimum DBI obtained was approximately 0.45. The same value was obtained for a number of combinations of parame… view at source ↗
Figure 6
Figure 6. Figure 6: Left: Clustering results with PCA and k-means using the 240 sized sample the 9 x 16 𝑉-band adaptive pixel fingerprints and 4 clusters. Right: The same process as the left figure with one artificial fingerprint added which corresponds to a sine wave like light curve with an amplitude of 1 mag and a period of two years. SCAN parameters (𝜖, PP, N). In all cases we obtain the same qualitative and quantitive re… view at source ↗
Figure 8
Figure 8. Figure 8: Stability test using PCA for our sample of 240 variable objects plus light curves representing sine waves. The approximate periods (see text for details) are indicated in the legend. The amplitudes of the sine waves are 1 mag in all cases. For each period 960 repeats with bootstrapped photometry, random phase shifts, and varying cadences are used to determine the mean positions (points) and their standard … view at source ↗
Figure 9
Figure 9. Figure 9: Placement test using PCA for our sample of 240 variable objects plus model light curves representing bursters and dipper. See text for details on the model parameters. The mean positions (points) and their standard deviations (shaded ellipses) for the models and data are determined in the same way as for [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Loadings matrices of the 9 x 16 adaptive pixel fingerprints for principle component 1 (left) and principle component 2 (right). The colour scale shows how much each pixel value contributes to the positioning in the fingerprint landscape. other hand, all variability in the light curve above an amplitude of 0.3 mag pushes the object towards smaller PC 1 values. Thus, a theoretically non-variable light curve… 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

39 extracted references · 22 canonical work pages · 1 internal anchor

  1. [1]

    Audard M., et al., 2014, @doi [Protostars and Planets VI] 10.2458/azu_uapress_9780816531240-ch017 , http://adsabs.harvard.edu/abs/2014prpl.conf..387A pp 387--410

    doi:10.2458/azu_uapress_9780816531240-ch017 2014

  2. [2]

    Bacher A., Kimeswenger S., Teutsch P., 2005, @doi [ ] 10.1111/j.1365-2966.2005.09329.x , https://ui.adsabs.harvard.edu/abs/2005MNRAS.362..542B 362, 542

    arXiv 2005

  3. [3]

    Bertin E., Arnouts S., 1996, @doi [ ] 10.1051/aas:1996164 , https://ui.adsabs.harvard.edu/abs/1996A&AS..117..393B 117, 393

    doi:10.1051/aas:1996164 1996

  4. [4]

    Bouvier J., et al., 1999, , https://ui.adsabs.harvard.edu/abs/1999A&A...349..619B 349, 619

    work page 1999

  5. [5]

    Bouvier J., Alencar S. H. P., Harries T. J., Johns-Krull C. M., Romanova M. M., 2007, in Reipurth B., Jewitt D., Keil K., eds, Protostars and Planets V. p. 479 ( @eprint arXiv astro-ph/0603498 ), @doi 10.48550/arXiv.astro-ph/0603498

    Pith/arXiv arXiv doi:10.48550/arxiv.astro-ph/0603498 2007

  6. [6]

    P., Mohanty S., Scholz A., Stassun K

    Bouvier J., Matt S. P., Mohanty S., Scholz A., Stassun K. G., Zanni C., 2014, @doi [Protostars and Planets VI] 10.2458/azu_uapress_9780816531240-ch019 , http://adsabs.harvard.edu/abs/2014prpl.conf..433B pp 433--450

    doi:10.2458/azu_uapress_9780816531240-ch019 2014

  7. [7]

    M., Hillenbrand L

    Carpenter J. M., Hillenbrand L. A., Skrutskie M. F., 2001, @doi [ ] 10.1086/321086 , https://ui.adsabs.harvard.edu/abs/2001AJ....121.3160C 121, 3160

    doi:10.1086/321086 2001

  8. [8]

    M., Stauffer J., Baglin A., Micela G., Rebull L

    Cody A. M., Stauffer J., Baglin A., Micela G., Rebull L. M., Flaccomio E., et al. 2014, @doi [ ] 10.1088/0004-6256/147/4/82 , http://adsabs.harvard.edu/abs/2014AJ....147...82C 147, 82

    doi:10.1088/0004-6256/147/4/82 2014

  9. [9]

    M., Hillenbrand L

    Cody A. M., Hillenbrand L. A., Rebull L. M., 2022, @doi [ ] 10.3847/1538-3881/ac5b73 , https://ui.adsabs.harvard.edu/abs/2022AJ....163..212C 163, 212

    doi:10.3847/1538-3881/ac5b73 2022

  10. [10]

    L., Bouldin D

    Davies D. L., Bouldin D. W., 1979, @doi [IEEE Transactions on Pattern Analysis and Machine Intelligence] 10.1109/TPAMI.1979.4766909 , PAMI-1, 224

    work page arXiv 1979

  11. [11]

    AAAI Press, pp 226--231, https://www.aaai.org/Papers/KDD/1996/KDD96-037.pdf

    Ester M., Kriegel H.-P., Sander J., Xu X., 1996, in Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD'96). AAAI Press, pp 226--231, https://www.aaai.org/Papers/KDD/1996/KDD96-037.pdf

    work page 1996

  12. [12]

    J., et al., 2020, @doi [ ] 10.1093/mnras/staa158 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.493..184E 493, 184

    Evitts J. J., et al., 2020, @doi [ ] 10.1093/mnras/staa158 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.493..184E 493, 184

    work page doi:10.1093/mnras/staa158 2020

  13. [13]

    M., Hillenbrand L., 2015, @doi [ ] 10.1088/0004-637X/798/2/89 , https://ui.adsabs.harvard.edu/abs/2015ApJ...798...89F 798, 89

    Findeisen K., Cody A. M., Hillenbrand L., 2015, @doi [ ] 10.1088/0004-637X/798/2/89 , https://ui.adsabs.harvard.edu/abs/2015ApJ...798...89F 798, 89

    work page doi:10.1088/0004-637x/798/2/89 2015

  14. [14]

    J., Hillenbrand L

    Fischer W. J., Hillenbrand L. A., Herczeg G. J., Johnstone D., Kospal A., Dunham M. M., 2023, in Inutsuka S., Aikawa Y., Muto T., Tomida K., Tamura M., eds, Astronomical Society of the Pacific Conference Series Vol. 534, Protostars and Planets VII. p. 355 ( @eprint arXiv 2203.11257 ), @doi 10.48550/arXiv.2203.11257

    Pith/arXiv arXiv doi:10.48550/arxiv.2203.11257 2023

  15. [15]

    Froebrich D., et al., 2018, @doi [ ] 10.1093/mnras/sty1350 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.478.5091F 478, 5091

    work page doi:10.1093/mnras/sty1350 2018

  16. [16]

    Froebrich D., et al., 2021, @doi [ ] 10.1093/mnras/stab2082 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.506.5989F 506, 5989

    doi:10.1093/mnras/stab2082 2021

  17. [17]

    Froebrich D., Eisl \"o ffel J., Stecklum B., Herbert C., Hambsch F.-J., 2022, @doi [ ] 10.1093/mnras/stab3450 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.510.2883F 510, 2883

    work page doi:10.1093/mnras/stab3450 2022

  18. [18]

    Froebrich D., et al., 2024, @doi [ ] 10.1093/mnras/stae311 , https://ui.adsabs.harvard.edu/abs/2024MNRAS.529.1283F 529, 1283

    work page doi:10.1093/mnras/stae311 2024

  19. [19]

    N., Melnikov S

    Grankin K. N., Melnikov S. Y., Bouvier J., Herbst W., Shevchenko V. S., 2007, @doi [ ] 10.1051/0004-6361:20065489 , http://adsabs.harvard.edu/abs/2007A

    work page doi:10.1051/0004-6361:20065489 2007

  20. [20]

    J., 2012, The Journal of the American Association of Variable Star Observers, https://ui.adsabs.harvard.edu/abs/2012JAVSO..40.1003H 40, 1003

    Hambsch F. J., 2012, The Journal of the American Association of Variable Star Observers, https://ui.adsabs.harvard.edu/abs/2012JAVSO..40.1003H 40, 1003

    work page 2012

  21. [21]

    Springer series in statistics, Springer, https://books.google.co.uk/books?id=eBSgoAEACAAJ

    Hastie T., Tibshirani R., Friedman J., 2009, The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Springer series in statistics, Springer, https://books.google.co.uk/books?id=eBSgoAEACAAJ

    work page 2009

  22. [22]

    Herbert C., Froebrich D., Scholz A., 2023, @doi [ ] 10.1093/mnras/stac3051 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.520.5433H 520, 5433

    work page doi:10.1093/mnras/stac3051 2023

  23. [23]

    Herbst W., Eisl \"o ffel J., Mundt R., Scholz A., 2007, in Reipurth B., Jewitt D., Keil K., eds, Protostars and Planets V. p. 297 ( @eprint arXiv astro-ph/0603673 ), @doi 10.48550/arXiv.astro-ph/0603673

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.astro-ph/0603673 2007

  24. [24]

    A., Kiker T

    Hillenbrand L. A., Kiker T. J., Gee M., Lester O., Braunfeld N. L., Rebull L. M., Kuhn M. A., 2022, @doi [ ] 10.3847/1538-3881/ac62d8 , https://ui.adsabs.harvard.edu/abs/2022AJ....163..263H 163, 263

    work page doi:10.3847/1538-3881/ac62d8 2022

  25. [25]

    W., Blanton M., Lang D., Mierle K., Roweis S., 2008, in Argyle R

    Hogg D. W., Blanton M., Lang D., Mierle K., Roweis S., 2008, in Argyle R. W., Bunclark P. S., Lewis J. R., eds, Astronomical Society of the Pacific Conference Series Vol. 394, Astronomical Data Analysis Software and Systems XVII. p. 27

    work page 2008

  26. [26]

    H., 1945, @doi [ ] 10.1086/144749 , http://adsabs.harvard.edu/abs/1945ApJ...102..168J 102, 168

    Joy A. H., 1945, @doi [ ] 10.1086/144749 , http://adsabs.harvard.edu/abs/1945ApJ...102..168J 102, 168

    doi:10.1086/144749 1945

  27. [27]

    S., et al., 2017, @doi [ ] 10.1088/1538-3873/aa80d9 , https://ui.adsabs.harvard.edu/abs/2017PASP..129j4502K 129, 104502

    Kochanek C. S., et al., 2017, @doi [ ] 10.1088/1538-3873/aa80d9 , https://ui.adsabs.harvard.edu/abs/2017PASP..129j4502K 129, 104502

    doi:10.1088/1538-3873/aa80d9 2017

  28. [28]

    S., Naylor T., 2022, @doi [ ] 10.1093/mnras/stac1477 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.514.2736L 514, 2736

    Lakeland B. S., Naylor T., 2022, @doi [ ] 10.1093/mnras/stac1477 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.514.2736L 514, 2736

    doi:10.1093/mnras/stac1477 2022

  29. [29]

    Moffat A. F. J., 1969, , https://ui.adsabs.harvard.edu/abs/1969A&A.....3..455M 3, 455

    work page 1969

  30. [30]

    Springer

    Murty M., Devi V., 2016, Pattern Recognition, An Algorithmic Approach. Springer

    work page 2016

  31. [31]

    Rigon L., Scholz A., Anderson D., West R., 2017, @doi [ ] 10.1093/mnras/stw2977 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.465.3889R 465, 3889

    doi:10.1093/mnras/stw2977 2017

  32. [32]

    J., 1987, @doi [Journal of Computational and Applied Mathematics] 10.1016/0377-0427(87)90125-7 , 20, 53

    Rousseeuw P. J., 1987, @doi [Journal of Computational and Applied Mathematics] 10.1016/0377-0427(87)90125-7 , 20, 53

    doi:10.1016/0377-0427(87)90125-7 1987

  33. [33]

    Scholz A., Eisl \"o ffel J., 2004, @doi [ ] 10.1051/0004-6361:20034022 , https://ui.adsabs.harvard.edu/abs/2004A&A...419..249S 419, 249

    work page doi:10.1051/0004-6361:20034022 2004

  34. [34]

    J., Naylor T., Littlefair S

    Sergison D. J., Naylor T., Littlefair S. P., Bell C. P. M., Williams C. D. H., 2020, @doi [ ] 10.1093/mnras/stz3398 , https://ui.adsabs.harvard.edu/abs/2020MNRAS.491.5035S 491, 5035

    work page doi:10.1093/mnras/stz3398 2020

  35. [35]

    V., et al., 2017, @doi [ ] 10.1093/mnras/stw2262 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.464..274S 464, 274

    Sokolovsky K. V., et al., 2017, @doi [ ] 10.1093/mnras/stw2262 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.464..274S 464, 274

    work page doi:10.1093/mnras/stw2262 2017

  36. [36]

    B., 1996, @doi [ ] 10.1086/133808 , https://ui.adsabs.harvard.edu/abs/1996PASP..108..851S 108, 851

    Stetson P. B., 1996, @doi [ ] 10.1086/133808 , https://ui.adsabs.harvard.edu/abs/1996PASP..108..851S 108, 851

    doi:10.1086/133808 1996

  37. [37]

    Wang Y., Huang H., Rudin C., Shaposhnik Y., 2021, Journal of Machine Learning Research, 22, 1

    work page 2021

  38. [38]

    van der Maaten L., Hinton G., 2008, Journal of Machine Learning Research, 9, 2579

    work page 2008

  39. [39]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTION or pop #1...

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.