pith. sign in

arxiv: 2510.07747 · v2 · submitted 2025-10-09 · 🌌 astro-ph.IM · astro-ph.GA· astro-ph.SR

Introducing SESHAT: A Tool for Object Classification from JWST Catalogs

B. L. Crompvoets , H. Kirk , R. Gutermuth , J. Di Francesco This is my paper

Pith reviewed 2026-05-18 09:24 UTC · model grok-4.3

classification 🌌 astro-ph.IM astro-ph.GAastro-ph.SR
keywords JWSTobject classificationmachine learningyoung stellar objectsphotometrybrown dwarfswhite dwarfsgalaxies
0
0 comments X

The pith

SESHAT uses XGBoost on synthetic photometry to classify JWST objects into young stellar objects, stars, brown dwarfs, white dwarfs, and galaxies with at least 85% recall.

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

The paper introduces SESHAT as a Python package that classifies five broad types of objects in JWST photometry catalogs. It trains the XGBoost algorithm on large sets of synthetic photometry that are adjusted at runtime to use only the filters present in any given dataset. Validation against real observations in both star-forming regions and deep cosmological fields shows the method recovers the known classes at a minimum of 85% recall per class. The classification works from photometry alone and does not require ellipticity or spatial information. The same framework can also be used to test whether a proposed set of JWST filters will be sufficient to separate the desired object classes.

Core claim

SESHAT identifies Young Stellar Objects, field stars (main sequence through asymptotic giant branch), brown dwarfs, white dwarfs, and galaxies from any JWST photometry by applying the XGBoost machine learning method to thousands of rows of synthetic photometry that is modified at run-time to match the filters available in the data. On real data from both star-forming regions and cosmological fields the tool reproduces the observed classes to a minimum of 85% recall across every class without additional information on ellipticity or spatial distribution. The package is released for general use and can verify whether chosen filters are sufficient to identify target classes.

What carries the argument

XGBoost classifier trained on synthetic photometry that is modified at runtime to match the specific JWST filters present in the input catalog.

If this is right

  • Large JWST catalogs can be automatically divided into the five object classes without manual inspection or extra morphological data.
  • Astronomers can check in advance whether a planned set of JWST filters will allow reliable separation of the objects they intend to study.
  • The same trained model works across both crowded star-forming regions and sparse cosmological fields.
  • The released Python package lets any user apply the classifier to their own JWST photometry sets.

Where Pith is reading between the lines

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

  • The approach could be retrained on data from other telescopes to create similar classifiers for existing archives.
  • Combining the photometric classes with positional information after the fact might reveal spatial patterns in young stellar populations.
  • If the tool is run on overlapping JWST and other-wavelength datasets, discrepancies could highlight objects with unusual spectral energy distributions.

Load-bearing premise

The synthetic photometry, modified at run-time to match the filters available in the data, accurately captures the photometric signatures of the target object classes in actual JWST observations.

What would settle it

Applying SESHAT to a new JWST catalog whose object classes have been independently confirmed by spectroscopy or other methods and finding recall below 85% for any class would falsify the performance claim.

read the original abstract

JWST's exquisite data have opened the doors to new possibilities in detecting broad classes of astronomical objects, but also to new challenges in classifying those objects. In this work, we introduce SESHAT, the Stellar Evolutionary Stage Heuristic Assessment Tool for the identification of Young Stellar Objects, field stars (main sequence through asymptotic giant branch), brown dwarfs, white dwarfs, and galaxies, from any JWST photometry. This identification is done using the machine learning method XGBoost to analyze thousands of rows of synthetic photometry, modified at run-time to match the filters available in the data to be classified. We validate this tool on real data of both star-forming regions and cosmological fields, and find we are able to reproduce the observed classes of objects to a minimum of 85\% recall across every class, with all available data, without additional information on the ellipticity or spatial distribution of the objects. Furthermore, this tool can be used to test the filter choices for JWST proposals by verifying whether the chosen filters are sufficient to identify the desired class of objects. SESHAT is released as a Python package to the community for general use.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces SESHAT, a Python package that applies the XGBoost classifier to synthetic photometry generated from evolutionary models and adjusted at runtime to match the specific JWST filters present in a given catalog. The tool targets five broad classes (Young Stellar Objects, field stars from main sequence to AGB, brown dwarfs, white dwarfs, and galaxies) using only photometric information. The central result is a claimed minimum recall of 85% when the trained model is applied to real JWST observations of both star-forming regions and cosmological fields, without any use of spatial morphology or ellipticity.

Significance. If the validation chain is shown to be robust, SESHAT would supply a practical, filter-agnostic classification utility that could be run on any JWST photometric catalog and could also serve as a forward-modeling aid for proposal filter selection. The open release of the code is a clear strength for community adoption and reproducibility.

major comments (2)
  1. [Validation procedure (abstract and §4)] Validation procedure (abstract and §4): the headline claim of ≥85% recall on real data is presented without any description of training-set construction from the evolutionary models, the cross-validation scheme, class-imbalance mitigation, or the precise train/test split between synthetic and real objects. These omissions make it impossible to assess whether the reported performance is statistically reliable or merely reflects optimistic partitioning.
  2. [Synthetic photometry fidelity (§3)] Synthetic photometry fidelity (§3): the manuscript provides no quantitative comparison (e.g., magnitude or color residuals, Kolmogorov-Smirnov statistics, or overlap metrics) between the runtime-modified synthetic photometry and actual JWST observations of literature-classified objects. Without such diagnostics, it remains possible that the classifier is learning model-specific features rather than observational signatures, directly undermining the transferability claim.
minor comments (2)
  1. The abstract states that the tool works 'with all available data' but does not report the number of real objects, the specific catalogs, or the filter sets used in the validation experiments; adding these numbers would improve clarity.
  2. XGBoost hyperparameter values and the feature list (magnitudes, colors, or both) should be stated explicitly rather than left as 'hyperparameters' so that readers can reproduce the exact model.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which highlight important areas for improving the clarity and robustness of the SESHAT manuscript. We address each major comment below and confirm that revisions will be incorporated in the next version.

read point-by-point responses

  1. Referee: [Validation procedure (abstract and §4)] Validation procedure (abstract and §4): the headline claim of ≥85% recall on real data is presented without any description of training-set construction from the evolutionary models, the cross-validation scheme, class-imbalance mitigation, or the precise train/test split between synthetic and real objects. These omissions make it impossible to assess whether the reported performance is statistically reliable or merely reflects optimistic partitioning.

    Authors: We agree that additional methodological details are required for readers to evaluate the reliability of the reported performance. In the revised manuscript we will add a new subsection to §4 that describes: (i) the construction of the synthetic training set from the specific evolutionary models used (BT-Settl, PARSEC, and MIST grids, with the exact parameter ranges and filter transmission curves applied at runtime); (ii) the stratified 5-fold cross-validation performed during hyperparameter tuning on the synthetic data; (iii) class-imbalance handling via XGBoost’s scale_pos_weight parameter tuned per class; and (iv) the explicit train/test protocol in which the classifier is trained exclusively on synthetic photometry and evaluated on independent, literature-classified real JWST catalogs from both star-forming regions and cosmological fields. These additions will make the validation chain fully reproducible and address the concern about optimistic partitioning. revision: yes

  2. Referee: [Synthetic photometry fidelity (§3)] Synthetic photometry fidelity (§3): the manuscript provides no quantitative comparison (e.g., magnitude or color residuals, Kolmogorov-Smirnov statistics, or overlap metrics) between the runtime-modified synthetic photometry and actual JWST observations of literature-classified objects. Without such diagnostics, it remains possible that the classifier is learning model-specific features rather than observational signatures, directly undermining the transferability claim.

    Authors: We acknowledge the importance of demonstrating that the runtime-adjusted synthetic photometry reproduces real observational signatures. In the revised §3 we will add quantitative fidelity diagnostics, including: (i) magnitude and color residuals computed for a set of literature-classified objects observed with JWST; (ii) two-sample Kolmogorov-Smirnov tests on the distributions of key colors (e.g., F115W–F200W, F277W–F444W); and (iii) overlap metrics reporting the fraction of synthetic points lying within the 68 % and 95 % contours of the observed color-color distributions. These diagnostics will be shown both before and after the runtime filter adjustment to confirm that the synthetic data capture the essential observational features rather than model-specific artifacts, thereby supporting the transferability claim. revision: yes

Circularity Check

0 steps flagged

No significant circularity: validation uses independent real observations

full rationale

The paper generates synthetic photometry from evolutionary models (modified at runtime to match available filters) and trains an XGBoost classifier on it. Validation is performed by applying the trained model to real JWST photometry from star-forming regions and cosmological fields, reporting minimum 85% recall against the observed classes in those independent datasets. Because the test set consists of external real observations rather than model outputs or fitted parameters, the performance metric does not reduce to the training inputs by construction. No self-citations, uniqueness theorems, or ansatzes are shown as load-bearing in the derivation chain. The fidelity of synthetic photometry to real data is an assumption affecting correctness, not a circularity in the reported validation procedure.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that synthetic photometry can be generated and adjusted to stand in for real JWST observations of the five target classes, plus standard machine-learning generalization from training to test data.

free parameters (1)
  • XGBoost hyperparameters
    Model parameters are learned from synthetic data but no specific values or tuning procedure are stated in the abstract.
axioms (1)
  • domain assumption Synthetic photometry modified at run-time to match available filters accurately represents the photometric properties of young stellar objects, field stars, brown dwarfs, white dwarfs, and galaxies in real JWST data.
    This premise underpins both the training process and the claim that no ellipticity or spatial information is required.

pith-pipeline@v0.9.0 · 5748 in / 1438 out tokens · 68185 ms · 2026-05-18T09:24:29.481883+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.