Pith. sign in

REVIEW 3 major objections 3 minor 47 references

Spectroscopically confirmed Little Red Dots concentrate in two well-defined regions of a data map of 242,000 JWST sources, built with no colour cut; selecting there reaches ~78% purity at ~82% completeness and yields ~100 new candidates.

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

2026-08-01 04:22 UTC pith:BZ6TOAB3

load-bearing objection A careful, honest method paper whose headline purity/completeness numbers are partly in-sample; the approach is novel and worth engaging, but the quantitative edge over colour cuts isn't yet proven. the 3 major comments →

arxiv 2607.22835 v1 pith:BZ6TOAB3 submitted 2026-07-24 astro-ph.GA astro-ph.IM

Unsupervised selection and characterisation of Little Red Dots in JWST surveys with manifold learning

Michele Ginolfi , Filippo Mannucci , Alessandro Marconi , Francesco D'Eugenio , Giacomo Venturi , Francesco Belfiore , Giovanni Cresci , Caterina Bracci
show 4 more authors
Stefano Carniani Alessandra Cozzi Roberto Maiolino Guido Risaliti
This is my paper
classification astro-ph.GA astro-ph.IM
keywords Little Red DotsUMAPmanifold learningJWSThigh-redshift galaxiesactive galactic nucleiselection functionunsupervised 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 paper tries to establish that Little Red Dots — the compact, red, high-redshift sources whose physical nature and selection function remain open — can be found and characterised without hand-designed colour boundaries. The authors embed 242,327 well-measured, isolated sources from a homogeneous six-field JWST photometric catalogue into a two-dimensional map using the unsupervised manifold-learning method UMAP, so that objects with similar colours, compactness and photometric redshift lie close together, then anchor the map with spectroscopically confirmed LRDs classified from their V-shaped continuum rather than from broadband colours. They find the anchors concentrate in two compact, strongly over-dense regions that no colour cut was used to define, separated mainly by redshift; the main region reaches a directly measured purity of about 0.78 at about 0.82 completeness on the spectroscopically classified subset, cleaner than or competitive with published colour criteria re-applied to the same sample, and contributes about 100 previously uncatalogued candidates. If this holds, the selection function stops being the dominant unknown in LRD studies, and the same anchoring strategy becomes a general way to find and vet rare populations in large photometric surveys.

Core claim

On a two-dimensional UMAP embedding built from eleven features per source — seven F356W-normalised broadband colours, reference flux, two morphology indicators and photometric redshift — spectroscopically confirmed LRDs are strongly localised: a core enclosing 90% of the anchors holds fewer than a thousand of the 242,000 sources. The anchors form two clusters — a main locus of 56 (median z≈5.1) and a secondary locus of 11 (median z≈3.4) — enclosed by a Mahalanobis ellipse (an iso-probability contour of the anchor distribution) and a minimum-volume ellipse. The main region contains 282 sources, recovers 82% of the in-sample anchors at a purity of about 0.78 over the PRISM-classified subset, a

What carries the argument

The central machinery is the UMAP manifold itself: a two-dimensional projection of 242,327 sources in an eleven-dimensional feature space of broadband colours, morphology (stellarity, half-light radius) and photometric redshift, computed before any labels enter. The procedure turns semi-supervised only at the anchoring step: the handful of spectroscopically confirmed LRDs — classified from the V-shape of their continuum rather than from broadband colours — are clustered by density, and each cluster is enclosed by an ellipse whose size is the tunable completeness knob; a Mahalanobis iso-probability contour does this for the main locus, a minimum-volume ellipse for the elongated secondary one.

Load-bearing premise

The demonstration rests on the assumption that the spectroscopically confirmed LRDs used as the anchor are free of colour selection; the authors concede (Section 5.1) that the classification rests on the V-shaped continuum, a continuum-based counterpart of a colour selection, so the manifold region inherits the colour prior of the spectroscopic targeting, and the headline purity and completeness are measured relative to that V-shape-selected population rather than to all LRDs

What would settle it

Re-anchor the same UMAP embedding with LRDs selected orthogonally to the V-shape — e.g., compact X-ray sources or broad Balmer lines only — and check whether the regions move; if they do, the locus is an artefact of the continuum-shape prior. A cheaper test: take PRISM spectra of the 107 new candidates and the ~546 z≈8 candidates; if most lack the V-shape or broad lines, the ~0.78 purity does not extend beyond the already-classified subset and the z≈8 extension is refuted.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Photometric LRD selection no longer needs hand-tuned colour boundaries: on one common parent sample, the data-driven main region reaches the highest combined completeness–purity (Q≈0.80) of the compared selections, while the best two-branch colour cut reaches Q≈0.79.
  • The two loci are genuinely distinct populations, separated mainly by redshift and rest-UV luminosity rather than by emission-line properties; the lower-redshift secondary locus is where the four newly confirmed broad-line AGN sit, marking it as a follow-up target.
  • Brown dwarfs, the classic LRD contaminants, separate by themselves: the manifold reproduces the standard colour rejection without being told to, so contamination is visible and measurable rather than assumed.
  • The 107 new candidates are compact, at LRD-like redshifts, with bluer rest-optical colours — exactly the sources the strictest redness cuts exclude — implying that published LRD samples built on those cuts miss this bluer tail of the population.
  • The 546-source neighbourhood around the single outlying anchor, with photometric redshifts piled at z≈8 and a V-shaped stacked SED, is a concrete, spectroscopically testable prediction of a higher-redshift continuation of the main LRD locus.

Where Pith is reading between the lines

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

  • If the paper is right, the same anchored-manifold recipe should transfer to other surveys and populations — but the purity numbers are only as good as the anchor: an anchor built from emission-line- or X-ray-selected objects could shift the locus, so the 'no colour cut' claim should be read with the qualifier 'no colour cut beyond the spectroscopic targeting's own colour dependence.'
  • A natural extension is to run the same embedding on shallower but wider surveys, or directly on spectra, to test whether the two loci persist at lower flux limits; the paper itself notes its bluest-NIRCam-band detection requirement and isolation cut remove a preferentially high-redshift slice, so a censored-flux version of the feature space is the most direct upgrade.
  • The secondary locus is the fragile part of the result: re-including the single discrepant bright anchor inflates its region from 110 to 698 sources and drops purity from ~41% to ~7%, so a targeted spectroscopic campaign over the grating-only sources in that region — where the four new AGN were found — would either stabilise or dissolve it.

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

3 major / 3 minor

Summary. The paper proposes an unsupervised, label-anchored manifold-learning method for selecting Little Red Dots (LRDs) in JWST surveys and demonstrates it on the ASTRODEEP-JWST catalogue. ~242,000 isolated, well-measured sources are embedded in two dimensions with UMAP using eleven features — seven F356W-normalised broadband colours, the reference flux, stellarity, half-light radius, and photometric redshift (Table 1). The embedding is anchored by the 68 spectroscopically selected de Graaff et al. (2025) LRDs that pass the pre-processing; the anchors split into two groups, and Mahalanobis ellipses define a main region (56 anchors, f=1.0) and a secondary filament (10 anchors). The main region is reported to reach ~0.78 purity at ~0.82 completeness over the PRISM-classified subset, yields 107 new candidates, and is compared with four literature colour cuts re-applied to the same parent sample under common compactness and brown-dwarf rejection (Table 2). Auxiliary results include: feature-space verification of the localisation, a cross-validated supervised classifier as a diagnostic, a feature-ablation showing the colours alone carry the LRD signature, natural brown-dwarf separation, two populations differing mainly in redshift, four new broad-line AGN from archival grating spectroscopy, and a speculative z~8 extension of the main locus. The paper is transparent about its main weakness — the region is defined using the anchors and evaluated against them (§4.3; §5.1) — and about the co

Significance. If the central quantitative claim holds, the paper delivers a useful framework rather than merely another LRD catalogue: a common completeness–purity basis on which any selection can be compared (literature criteria re-applied to the identical parent sample, §3.3), a clean separation of intrinsic from end-to-end completeness (§2.4), and a tunable, transparent operating curve instead of a fixed cut. The ancillary results are credible and well-hedged: the brown-dwarf separation, the redshift-driven two-locus structure, the four grating-spectroscopy AGN, and the explicitly speculative z~8 prediction. The paper ships unusual methodological care: hyperparameter robustness checks (§3.1), bootstrap candidate stability (Jaccard 0.94), a label-permutation null test, and explicit disclosure of the in-sample evaluation. The unresolved point is whether the headline 0.78/0.82 is an independent measurement or partly a restatement of the anchor's V-shape colour prior; because the comparison against literature cuts is asymmetric in this respect, the quantitative advantage claimed in the abstract is not yet established at full strength. The gap is fixable within the paper's scope.

major comments (3)
  1. [§3.2, §4.3, Table 2 (and abstract)] The headline purity/completeness is an in-sample evaluation, and the comparison with the literature is asymmetric. The main-region ellipse is fit to all 56 DBSCAN-grouped anchors (§3.2); the f=1.0 completeness of 0.82 is, by construction, the fraction of the 68 in-sample anchors in the main cluster. The four literature 'criteria' points in Table 2 are genuinely out-of-sample: their thresholds were fixed without the anchor. The neural-network cross-validation (§4.2, Appendix A) tests a supervised classifier, not the geometric region, and no version of the Mahalanobis region is defined on a training half of the anchors and scored on the complementary half. I request a leave-half-out version of the region construction, reporting completeness and PRISM-subset purity of the region as a function of f on held-out anchors. Without this, 'competitive with, or cleaner than, literature colour cuts'
  2. [§4.3 (purity definition); cf. §1] Purity is measured only over the grade-3 PRISM-classified subset: the denominator is spectroscopically observed sources, the numerator is V-shape-classified LRDs. The paper correctly states that this is not the purity of the whole photometric sample and that classification is exhaustive within the subset. The composition of the subset is itself a selection, however: as §1 notes, archival spectroscopic samples 'deliberately targeted colour-selected candidates', so the PRISM-covered fraction of the region is enriched in LRD-like colours, and the quoted 0.78 is conditional on that enrichment, with the bias direction likely toward higher purity. This caveat should be stated where 0.78 is quoted in §4.3 and ideally quantified, e.g. by recomputing purity after assigning the interloper fraction (11 sources) to the spectroscopically unobserved members, or by reporting purity among PRISM-classifi
  3. [Abstract; §5.1, §5.3] The abstract's 'no colour cut imposed' is stronger than what is demonstrated and is in tension with the paper's own analysis. §5.1 concedes that the de Graaff et al. anchor 'rests on the continuum V-shape... a refined, continuum-based counterpart of a colour selection rather than one orthogonal to it', and §5.3 shows the seven broadband colours alone reproduce essentially the full localisation (classifier AUC 0.999), with morphology and photo-z largely redundant for identification. The manifold region is therefore a learned, higher-dimensional version of a colour selection, not a selection orthogonal to colour cuts. This is not an internal inconsistency — §5.1 is admirably clear — but the abstract and §1 framing ('without relying on predefined colour cuts', 'without human-induced priors') should carry the qualification (e.g., 'without hand-designed colour thresholds'), and the implicatio
minor comments (3)
  1. [§3.3, Table 2] State explicitly whether the common compactness proxy (ClassStarSE>0.8) and the F115W−F200W>−0.5 brown-dwarf cut are applied to the data-driven ellipse selection before the Table 2 entry is computed. The literature 'criteria' points include these cuts; if the data-driven row does not, the like-for-like comparison is not exactly symmetric (though §5.3 suggests applying the proxy would raise, not lower, the data-driven purity).
  2. [Figure 3] The filled 'catalogue' symbols are on a different evaluation basis (positional matches, not the common parent sample) from the open 'criteria' symbols; the caption should state that only the open symbols participate in the like-for-like comparison.
  3. [§5.2] The circular neighbourhood around the discarded outlier uses a radius equal to the mean semi-axis of the main-locus ellipse. This is flagged as a proxy, but the sensitivity of the 546-source count and the z_phot≃8 pile-up to that choice (e.g., semi-major axis, or a 90%-anchor contour) is not explored; a one-sentence robustness note would calibrate how speculative the z~8 prediction is.

Circularity Check

2 steps flagged

Main-region completeness is fixed by the f=1.0 ellipse choice (56/68 = 0.82 by construction), and purity is measured on the same anchors used to define the region; part of the headline comparison is in-sample.

specific steps
  1. fitted input called prediction [Section 3.2 (ellipse sizing) and Section 4.3 (reported completeness)]
    "The completeness with respect to the anchor is therefore a design parameter that we set and vary (we consider f between 0.5 and 1.0, and adopt f=1.0, which encloses all 56 main-region anchors, as the fiducial value) ... At its fiducial size the main-region selection recovers 82% of the in-sample anchors at a purity of ≈0.78"

    The f=1.0 Mahalanobis ellipse is explicitly sized to enclose all 56 main-region anchors. Since 68 de Graaff anchors enter the sample, 56/68 = 0.82, so the reported 82% completeness is the design fraction f restated as a measured recovery. The ellipse was built on those same objects, so this headline number is true by construction rather than an independent test of the selection.

  2. self definitional [Section 4.3 and Section 5.1]
    "the region is defined using the anchor and then evaluated against it, which favours the data-driven selection, and our purity is measured only over the PRISM-classified subset."

    The same de Graaff spectroscopic classification supplies both the anchors used to centre and size the Mahalanobis region and the labels used to count confirmed LRDs inside it. The quoted purity (≈0.78) is therefore an in-sample self-evaluation, not an out-of-sample prediction. The paper points to the Section 4.2 cross-validation as a fairer test, but that test is a supervised classifier on held-out anchors and never redefines the Mahalanobis region; it does not provide an out-of-sample purity for the headline selection.

full rationale

Most of the derivation is not circular: the UMAP embedding is built without labels; the concentration of the spectroscopic LRDs is supported by a label-permutation test and by ~520x k-NN over-density in the original 11-D feature space; and a five-fold cross-validated neural classifier recovers ~90% of held-out anchors. These tests genuinely support the claim that LRDs occupy a distinct locus. The circularity is confined to the headline quantitative comparison in Section 4.3/Table 2. The main-region ellipse is sized with f=1.0 so as to enclose all 56 main-region anchors, so the reported 82% completeness is simply 56/68 of the in-sample anchors, i.e. the design parameter rather than a measured recovery rate. The 0.78 purity is measured over the PRISM-classified subset using the same de Graaff classification that supplied the anchors, making it an in-sample estimate; the paper explicitly concedes that 'the region is defined using the anchor and then evaluated against it, which favours the data-driven selection.' The cross-validation offered as a fairer test does not rebuild the Mahalanobis region or report its purity, so it does not rescue the 0.78/0.82 numbers from being partly by construction. Additionally, the abstract's 'without relying on predefined colour cuts' framing is weakened by the authors' own admission that the de Graaff anchor is 'a refined, continuum-based counterpart of a colour selection rather than one orthogonal to it,' so the unsupervised locus partially re-encodes the known V-shape selection. These are partial circularities, not a wholesale collapse: the 107-candidate list, the two-population redshift split, and the four new broad-line AGN are genuine outputs that do not reduce to the inputs.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 1 invented entities

The free parameters are selection/region-definition choices, not fitted physical constants; the central claim depends on f=1.0 and the compactness proxy. The axioms are standard domain assumptions for photometric surveys; the most important is that the de Graaff anchor is a valid ground truth. One invented entity (the z~8 extension) is a genuine prediction with a testable handle.

free parameters (6)
  • Enclosed anchor fraction f (Mahalanobis radius scale) = 1.0 (fiducial); varied 0.5–1.0
    Sets the size of the main LRD region ellipse (Section 3.2); directly trades completeness against purity in Figure 3. The headline 0.82/0.78 numbers are quoted at f=1.0.
  • UMAP hyperparameters (n_neighbors=15, min_dist=0, Manhattan metric) = 15, 0, Manhattan
    Chosen after exploring a grid; the paper reports the qualitative structure is robust, but exact exploration values are not given (Section 3.1).
  • Compactness proxy ClassStarSE > 0.8 = 0.8
    Empirically calibrated to retain 99% of the spectroscopic anchor (Section 3.3); used in all literature comparisons and in the purity-restricted variant.
  • Signal-to-noise threshold S/N > 2 per band (F814W excepted) = 2
    Defines the parent sample and sets the redshift ceiling; the paper tested nearby values and found conclusions unchanged (Section 2.3).
  • F444W < 26 luminosity cut = 26 (mag)
    Optional purity lever in Section 4.3 that raises purity to ~0.85–0.90; post-hoc and not part of the fiducial selection.
  • Brown-dwarf rejection colour F115W−F200W > −0.5 = -0.5
    Adopted from Greene et al. (2024) and applied identically to all methods in the common-sample comparison (Section 3.3).
axioms (6)
  • domain assumption ASTRODEEP-JWST catalogue photometry, morphology and EAZY photometric redshifts are accurate and homogeneous across the six fields.
    The entire feature space is built from this catalogue (Section 2.1); any systematic photometric error or photo-z bias propagates directly into the manifold and the LRD region.
  • domain assumption The de Graaff et al. (2025) spectroscopic sample provides correct LRD labels and is a valid ground truth.
    Used as the anchor to define the LRD region and as the label for purity/completeness (Sections 2.5, 4.3). If the V-shape classification is incomplete or biased, the region is biased.
  • standard math UMAP embedding preserves the local neighbourhood structure of the 11-dimensional feature space.
    The method assumes local structure is informative; the authors verify the anchor concentration in the original feature space (Section 4.2), which supports this, but the region ellipses are still drawn in the projected plane.
  • domain assumption The grade-3 PRISM-classified subset is representative of the whole spectroscopically followed population.
    Purity is measured only over this subset (Section 4.3); if the subset skews towards or away from LRDs, the measured 0.78 purity is not the true purity.
  • domain assumption The isolation criterion (neighbour within 0.5″ removed) does not remove a significant fraction of LRDs.
    The authors test this against the Baggen et al. companion sample and find no significant bias (Section 2.3), but the test covers only one external catalogue, not all LRDs.
  • domain assumption A two-dimensional Gaussian/Mahalanobis ellipse is an adequate description of the LRD locus.
    Used to define the selection region (Section 3.2); a non-Gaussian or disconnected locus would change the completeness–purity curve.
invented entities (1)
  • High-redshift (z~8) extension of the main LRD locus independent evidence
    purpose: Proposed continuation of the LRD population at z~8, identified as the neighbourhood of a single discarded anchor (546 sources, median z_phot ~ 8.2)
    The paper explicitly flags this as a 'concrete, spectroscopically testable prediction' (Section 5.2) that requires dedicated spectroscopy; thus it has a falsifiable handle. All other regions (main/secondary loci) are empirical groupings of known anchors, not invented entities.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Unsupervised selection and characterisation of Little Red Dots in JWST surveys with manifold learning." pith.science (2026). https://pith.science/paper/BZ6TOAB3

@misc{pith2026260722835,
  author       = {Pith},
  title        = {Pith review of: Unsupervised selection and characterisation of Little Red Dots in JWST surveys with manifold learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BZ6TOAB3}},
  note         = {Machine review of arXiv:2607.22835}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Little Red Dots (LRDs) are compact, red sources discovered at high redshift by JWST whose physical nature and selection function remain debated. We investigate whether an unsupervised machine-learning approach applied to multi-band photometry can identify LRD-like objects, and other populations, without relying on predefined colour cuts. Using UMAP, a manifold-learning (dimensionality-reduction) method, we place ~242,000 isolated, well-measured sources from the ASTRODEEP-JWST catalogue on a two-dimensional map, where objects with similar broadband colours, morphology, and photometric redshift lie close together. We then use spectroscopically confirmed LRDs to identify where LRD-like objects lie within this map, compare the resulting areas with published colour cuts, and validate our data-driven selection against archival NIRSpec spectra from the DJA. We find that the spectroscopically selected LRDs concentrate in two well-defined regions with no colour cut imposed, tracing populations that differ mainly in redshift, a difference imprinted in their broadband colours. The main region reaches a purity of ~0.78 at ~0.82 completeness on the spectroscopically classified subset, competitive with, or cleaner than, literature colour cuts, and yields ~100 additional candidates. We also test the method as a general tool for population discovery: the manifold recovers the locations of brown dwarfs and broad-line AGN with no explicit criterion, and isolates rare pathological outliers. Overall, unsupervised manifolds, anchored by sparse high-confidence spectroscopic labels, provide an efficient, assumption-light framework for characterising populations, comparing selection methods on a common basis, and discovering rare objects in large photometric datasets.

Figures

Figures reproduced from arXiv: 2607.22835 by Alessandra Cozzi, Alessandro Marconi, Caterina Bracci, Filippo Mannucci, Francesco Belfiore, Francesco D'Eugenio, Giacomo Venturi, Giovanni Cresci, Guido Risaliti, Michele Ginolfi, Roberto Maiolino, Stefano Carniani.

Figure 1
Figure 1. Figure 1: The two-dimensional embedding colour-coded by (a) photometric redshift, (b) reference-band (F356W) flux, (c) stellarity, and (d) spectro￾scopic coverage with the locations of the spectroscopic LRDs. Smooth gradients in (a)–(c) show that the embedding encodes physical structure; panel (d) shows where the spectroscopic follow-up and the labelled anchor sit on the manifold. For the main region the size of the… view at source ↗
Figure 2
Figure 2. Figure 2: (a) The two anchor regions on the embedding. The spectroscopic anchors are shown on top (de Graaff et al. 2025 main and secondary loci); the photometrically selected samples of Barro et al. (2026a) (orange squares) and Kokorev et al. (2024) (purple triangles) are overlaid beneath them, illustrating that the latter are more dispersed across the manifold, and the black cross marks the lone anchor discarded b… view at source ↗
Figure 3
Figure 3. Figure 3: ), because the fainter members carry noisier photometry and admit more contaminants. Compactness provides a further, independent purity lever, which we return to in Section 5.3. 4.4. A sample of new candidates At the fiducial size the main region contains 282 sources. Of these, 164 are already identified as LRDs in at least one pub￾lished catalogue, and a further 11 have a high-quality PRISM spectrum but w… view at source ↗
Figure 4
Figure 4. Figure 4: The new candidates (blue) compared with the spectroscopic LRDs (red) in the two colour–colour planes used by the literature, with the corresponding selection boundaries overplotted, and in photometric redshift and F444W magnitude. The candidates are compact and at LRD-like redshifts but are bluer in the rest-optical, below the strict redness thresholds. sources, 22 have Hα in the reliable regime (Hα S/N ≳ … view at source ↗
Figure 5
Figure 5. Figure 5: Rest-frame stacked SED of the two loci (left: main; right: secondary, the fiducial ten-anchor filament). In each panel the photometric candidates (coloured line: median with 16–84th percentile band) are compared with the spectroscopically confirmed members (black points, with error bars giving the standard error of the binned median). Each source is normalised to its median band flux so the stack reflects … view at source ↗
Figure 6
Figure 6. Figure 6: (a) Brown dwarfs (green) relative to the LRD loci. (b) Zoom on the brown-dwarf region, with nearby photometric LRD candidates highlighted as possible contaminants. (c) The same in colour space; brown dwarfs are blue in F115W−F200W. The de Graaff et al. (2025) selection is not line-based: the low￾resolution PRISM cannot resolve broad Hα, so their classifica￾tion rests on the continuum V-shape (the UV and op… view at source ↗
Figure 7
Figure 7. Figure 7: The two loci are two populations that differ mainly in redshift and rest-ultraviolet luminosity. (a) Spectroscopic anchors, from the public de Graaff et al. (2025) fits: the secondary island (blue) lies at lower spectroscopic redshift and is fainter in MUV than the main island (red). (b) Photometric candidates: their photometric redshifts follow the same two-population split, indicating they are drawn from… view at source ↗
Figure 8
Figure 8. Figure 8: The four new broad-line AGN with no prior catalogue identification (Section 4.7). The top two panels are the two objects that also fall within the fiducial ten-anchor secondary region (the outlier-excluded filament; see [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The discarded de Graaff et al. (2025) outlier and its neighbourhood (speculative; a single spectroscopic redshift). (a) UMAP zoom: the field colour-coded by photometric redshift, the outlier (star), the assumed circular region (mean semi-axis of the main-locus f=1 ellipse), and the main locus (dashed). (b) Binned rest-frame stacked SED of the region members (normalised to the median band flux) with the out… view at source ↗
Figure 10
Figure 10. Figure 10: Feature ablation. For each subset of the input features, the feature-space k-nearest-neighbour over-density of the spectroscopic LRDs (left) and the cross-validated classifier separability (right). The broadband colours alone carry the LRD signature; morphology and photometric redshift add little and are weak on their own. A concrete demonstration is that the same representation that localises the LRDs al… view at source ↗
Figure 11
Figure 11. Figure 11: The colours-only embedding (morphology, reference flux and photometric redshift switched off). (a) The spectroscopic LRDs still form a single concentration and the brown dwarfs remain separate. (b) Photometric redshift is recovered as a smooth gradient. (c) In this colours-only projection the two fiducial populations merge into one concentration, although they remain separable in the underlying feature sp… view at source ↗
Figure 12
Figure 12. Figure 12: Broad-line AGN from the Baccus & Xu (2025) JWST/NIRSpec census located on the manifold (the 149 of 252 that fall in our parent sample), colour-coded by spectroscopic redshift, with the main and secondary LRD loci ellipses for reference (the dotted curve is the full secondary region). The view is zoomed on the lower part of the plane, where the BLAGN and the loci lie. The BLAGN are not randomly scattered: … view at source ↗
Figure 13
Figure 13. Figure 13: Three peculiar regions that the manifold isolates without any prior definition, one per row: top, saturated bright stars; middle, catastrophic photometric-redshift failures (Galactic stars fitted with high-redshift galaxy templates, so their catalogue zphot piles up near z ≃ 6); bottom, sources contaminated by the diffraction spikes of a nearby bright star. Columns: (a) the region’s location on the embedd… 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

47 extracted references · 9 linked inside Pith

  1. [1]

    B., Casey, C

    Akins, H. B., Casey, C. M., Lambrides, E., et al. 2025, ApJ, 991, 37

    2025

  2. [2]

    A., Siudek, M., Eriksen, M., et al

    Alcolea, J. A., Siudek, M., Eriksen, M., et al. 2026, arXiv e-prints, arXiv:2607.07329 Arrabal Haro, P., Dickinson, M., Finkelstein, S. L., et al. 2023, Nature, 622, 707 Astropy Collaboration, Price-Whelan, A. M., Lim, P. L., et al. 2022, ApJ, 935, 167

    Pith/arXiv arXiv 2026

  3. [3]

    Baccus, C. & Xu, X. 2025, arXiv e-prints, arXiv:2512.03281

    arXiv 2025

  4. [4]

    Baggen, J. F. W., Scoggins, M. T., van Dokkum, P., et al. 2026, ApJ, 1002, L4

    2026

  5. [5]

    B., Finkelstein, S

    Bagley, M. B., Finkelstein, S. L., Koekemoer, A. M., et al. 2023, ApJ, 946, L12

    2023

  6. [6]

    B., Pirzkal, N., Finkelstein, S

    Bagley, M. B., Pirzkal, N., Finkelstein, S. L., et al. 2024, ApJ, 965, L6

    2024

  7. [7]

    G., Kocevski, D

    Barro, G., Pérez-González, P. G., Kocevski, D. D., et al. 2024, ApJ, 963, 128

    2024

  8. [8]

    E., et al

    Bezanson, R., Labbe, I., Whitaker, K. E., et al. 2024, ApJ, 974, 92

    2024

  9. [9]

    B., van Dokkum, P

    Brammer, G. B., van Dokkum, P. G., & Coppi, P. 2008, ApJ, 686, 1503

    2008

  10. [10]

    2026, arXiv e-prints, arXiv:2601.22214

    Brazzini, M., D’Eugenio, F., Maiolino, R., et al. 2026, arXiv e-prints, arXiv:2601.22214

    arXiv 2026

  11. [11]

    O., Hickox, R

    Casey, Q. O., Hickox, R. C., Cleri, N. J., et al. 2026, arXiv e-prints, arXiv:2606.26098 de Graaff, A., Hviding, R. E., Naidu, R. P., et al. 2025, arXiv e-prints, arXiv:2511.21820 D’Eugenio, F., Cameron, A. J., Scholtz, J., et al. 2025, ApJS, 277, 4

    Pith/arXiv arXiv 2026

  12. [12]

    S., Abraham, R

    Dunlop, J. S., Abraham, R. G., Ashby, M. L. N., et al. 2021, PRIMER: Public Release IMaging for Extragalactic Research, JWST Proposal. Cycle 1, ID. #1837

    2021

  13. [13]

    J., Johnson, B

    Eisenstein, D. J., Johnson, B. D., Robertson, B., et al. 2025, ApJS, 281, 50

    2025

  14. [14]

    J., Willott, C., Alberts, S., et al

    Eisenstein, D. J., Willott, C., Alberts, S., et al. 2026, ApJS, 283, 6

    2026

  15. [15]

    L., Bagley, M

    Finkelstein, S. L., Bagley, M. B., Ferguson, H. C., et al. 2023, ApJ, 946, L13

    2023

  16. [16]

    2024, Astronomy and Computing, 48, 100851

    Fotopoulou, S. 2024, Astronomy and Computing, 48, 100851

    2024

  17. [17]

    2026, Nature Astronomy [arXiv:2512.02096]

    Fu, S., Zhang, Z., Jiang, D., et al. 2026, Nature Astronomy [arXiv:2512.02096]

    arXiv 2026

  18. [18]

    E., Labbe, I., Goulding, A

    Greene, J. E., Labbe, I., Goulding, A. D., et al. 2024, ApJ, 964, 39

    2024

  19. [19]

    N., Helton, J

    Hainline, K. N., Helton, J. M., Johnson, B. D., et al. 2024, ApJ, 964, 66

    2024

  20. [20]

    N., Helton, J

    Hainline, K. N., Helton, J. M., Miles, B. E., et al. 2026, ApJ, 1004, 223

    2026

  21. [21]

    R., Millman, K

    Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357

    2020

  22. [22]

    E., Watson, D., Brammer, G., et al

    Heintz, K. E., Watson, D., Brammer, G., et al. 2024, Science, 384, 890

    2024

  23. [23]

    E., Sun, Y ., & Davey, N

    Hocking, A., Geach, J. E., Sun, Y ., & Davey, N. 2018, MNRAS, 473, 1108

    2018

  24. [24]

    Hunter, J. D. 2007, Computing in Science and Engineering, 9, 90 Juodžbalis, I., Ji, X., Maiolino, R., et al. 2024, MNRAS, 535, 853

    2007

  25. [25]

    D., Finkelstein, S

    Kocevski, D. D., Finkelstein, S. L., Barro, G., et al. 2025, ApJ, 986, 126

    2025

  26. [26]

    I., Greene, J

    Kokorev, V ., Caputi, K. I., Greene, J. E., et al. 2024, ApJ, 968, 38 Labbé, I., van Dokkum, P., Nelson, E., et al. 2023, Nature, 616, 266

    2024

  27. [27]

    2026, arXiv e-prints, arXiv:2605.21574

    Lin, X., Fan, X., Cai, Z., et al. 2026, arXiv e-prints, arXiv:2605.21574

    Pith/arXiv arXiv 2026

  28. [28]

    2025, A&A, 703, A36

    Loiacono, F., Gilli, R., Mignoli, M., et al. 2025, A&A, 703, A36

    2025

  29. [29]

    & Maiolino, R

    Madau, P. & Maiolino, R. 2026, arXiv e-prints, arXiv:2605.05074

    Pith/arXiv arXiv 2026

  30. [30]

    P., Brammer, G., et al

    Matthee, J., Naidu, R. P., Brammer, G., et al. 2024, ApJ, 963, 129

    2024

  31. [31]

    2026, arXiv e-prints, arXiv:2603.17667

    Matthee, J., Torralba, A., Pezzulli, G., et al. 2026, arXiv e-prints, arXiv:2603.17667

    arXiv 2026

  32. [32]

    2018, arXiv e-prints, arXiv:1802.03426 Article number, page 17 A&A proofs:manuscript no

    McInnes, L., Healy, J., & Melville, J. 2018, arXiv e-prints, arXiv:1802.03426 Article number, page 17 A&A proofs:manuscript no. lrd_umap_aa Fig. 13.Three peculiar regions that the manifold isolates without any prior definition, one per row:top, saturated bright stars;middle, catastrophic photometric-redshift failures (Galactic stars fitted with high-redsh...

    Pith/arXiv arXiv 2018

  33. [33]

    2024, A&A, 691, A240

    Merlin, E., Santini, P., Paris, D., et al. 2024, A&A, 691, A240

    2024

  34. [34]

    2026, arXiv e-prints, arXiv:2606.09721

    Pan, Z., Zhuang, M.-Y ., Shen, Y ., et al. 2026, arXiv e-prints, arXiv:2606.09721

    Pith/arXiv arXiv 2026

  35. [35]

    2026, arXiv e-prints, arXiv:2605.14233

    Park, K., Torralba, A., Matthee, J., et al. 2026, arXiv e-prints, arXiv:2605.14233

    Pith/arXiv arXiv 2026

  36. [36]

    2011, Journal of Machine Learning Research, 12, 2825 Pérez-González, P

    Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, Journal of Machine Learning Research, 12, 2825 Pérez-González, P. G., Barro, G., Carniani, S., et al. 2026, arXiv e-prints, arXiv:2602.20247

    arXiv 2011

  37. [37]

    Portillo, S. K. N., Parejko, J. K., Vergara, J. R., & Connolly, A. J. 2020, AJ, 160, 45

    2020

  38. [38]

    X., & Wolf, L

    Reis, I., Rotman, M., Poznanski, D., Prochaska, J. X., & Wolf, L. 2021, Astron- omy and Computing, 34, 100437

    2021

  39. [39]

    J., Robertson, B., Tacchella, S., et al

    Rieke, M. J., Robertson, B., Tacchella, S., et al. 2023, ApJS, 269, 16

    2023

  40. [40]

    2026, arXiv e-prints, arXiv:2604.07138

    Rinaldi, P., Hainline, K., D’Eugenio, F., et al. 2026, arXiv e-prints, arXiv:2604.07138

    Pith/arXiv arXiv 2026

  41. [41]

    P., et al

    Rusakov, V ., Watson, D., Nikopoulos, G. P., et al. 2026, Nature, 649, 574

    2026

  42. [42]

    2025, arXiv e-prints, arXiv:2511.05439

    Saxena, A. 2025, arXiv e-prints, arXiv:2511.05439

    arXiv 2025

  43. [43]

    H., Watson, D., et al

    Sneppen, A., Matthews, J. H., Watson, D., et al. 2026, arXiv e-prints, arXiv:2604.09399

    Pith/arXiv arXiv 2026

  44. [44]

    2022, ApJ, 935, 110

    Treu, T., Roberts-Borsani, G., Bradac, M., et al. 2022, ApJ, 935, 110

    2022

  45. [45]

    Valentino, F., Brammer, G., Gould, K. M. L., et al. 2023, ApJ, 947, 20 van der Maaten, L. & Hinton, G. 2008, Journal of Machine Learning Research, 9, 2579

    2023

  46. [46]

    E., et al

    Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Medicine, 17, 261 Wes McKinney. 2010, in Proceedings of the 9th Python in Science Conference, ed. Stéfan van der Walt & Jarrod Millman, 56 – 61

    2020

  47. [47]

    C., & Inayoshi, K

    Zhang, Z., Jiang, L., Liu, W., Ho, L. C., & Inayoshi, K. 2026, ApJ, 998, 170 Article number, page 18 M. Ginolfi et al.: Mapping LRDs in JWST surveys with UMAP Appendix A: Feature-space classifier validation As an additional check that the LRD localisation is not a pro- jection artefact, we trained a supervised classifier directly on the original feature v...

    2026

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.