{"id":"b42337e6-3b07-4675-b99a-bd083b380ed4","arxiv_id":"2505.16509","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"An unsupervised self-organizing map reveals that AGN selection methods trace distinct dwarf galaxy SED regions, with WISE-selected candidates splitting into a starburst-contaminated clump and a more AGN-like clump.","lead":"The paper trains a self-organizing map on 30,344 nearby dwarf galaxies and maps 438 known active-galactic-nucleus candidates onto it, showing that each detection method samples a different part of galaxy color space. It matters because the map splits infrared-selected AGN candidates into two groups, one of which may be mostly star-forming impostors, which could improve black hole censuses in dwarf galaxies.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"WISE clump dichotomy may be an artifact of the [O iii]/([O ii]×z) ratio: low [O ii] in quiescent left-clump galaxies inflates the proxy regardless of AGN content.","rationale":"The reader's weakest assumption correctly identified the line-ratio proxy maps as the load-bearing support for the WISE dichotomy. My concern sharpens this: beyond being undefined and uncorrected, the specific ratio [O iii]/([O ii]×z) has a denominator that is itself a star-formation tracer, creating a built-in anticorrelation between the proxy and star formation. Red/quiescent galaxies (the left clump) will therefore score high on this proxy even with no AGN, potentially manufacturing the observed distinction. This is a concrete, testable mechanism that the paper does not address. The proposed test—replacing the compound ratio with a standard AGN diagnostic—would directly determine whether the left clump's elevated proxy values trace actual AGN activity. Because the paper has other supporting evidence (SDSS image inspection, color–mass diagrams), the verdict remains conditional rather than reject; the concern strengthens the need for quantitative validation but does not prove the central claim false. Agreement with the reader is partial because my concern is a specific statistical pathology of the proxy, while the reader's concern was more general about missing definitions and corrections.","tokens_in":16417,"tokens_out":5050,"duration_ms":44237,"concrete_test":"Recompute the Figure 5 SOM color map using [O iii]/Hβ (or dust-corrected [O iii] luminosity) instead of [O iii]/([O ii]×z), using SDSS emission-line measurements for the training galaxies. If the left WISE clump is no longer elevated relative to the right clump, the AGN/starburst interpretation is an artifact of dividing by low [O ii].","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that WISE-selected AGNs split into a starburst-contaminated right clump and an AGN-dominated left clump hinges on the proxy maps in Figures 4 and 5. The paper uses [O ii]/z as a star-formation tracer and [O iii]/([O ii]×z) as an AGN luminosity tracer, but the latter is structurally biased: dividing by [O ii] (a star-formation tracer) means galaxies with weak [O ii]—typically red, quiescent systems, exactly the population in the left clump—will have large ratios even if their [O iii] flux is modest. The paper never reports the separate [O ii] and [O iii] fluxes or any standard AGN diagnostic (e.g., [O iii]/Hβ) for the two clumps. Moreover, the emission-line measurements are undefined: no description of continuum subtraction, dust-extinction correction, or aperture matching is given. Thus the observed elevation of [O iii]/([O ii]×z) in the left clump could simply reflect low star formation rather than genuine AGN dominance. The right-clump starburst interpretation is similarly dependent on these unvetted proxies. Without confirming that [O iii] itself is enhanced in the left clump, the paper's most novel finding—that the two WISE clumps separate AGN hosts from starburst contaminants—is not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper trains a 51×51 self-organizing map on 30,344 dwarf galaxies from the NSA catalog (z < 0.055, M* < 10^9.5 Msun) using seven colors from nine bands (GALEX FUV/NUV, SDSS ugriz, WISE W1/W2). A sample of 438 previously identified dwarf AGNs, compiled by Wasleske & Baldassare (2024), is mapped onto the SOM and the locations of AGNs selected by BPT, [O i], [S ii], He ii, broad-line, X-ray, variability, and WISE methods are compared. The central claim is that different AGN diagnostics occupy distinct and partially overlapping regions of SED space, reflecting host-property selection biases. The most novel finding is that WISE-selected AGNs form two clumps: a redder, more AGN-like left clump and a bluer, starburst-like right clump, suggesting a way to reduce starburst contamination in WISE-selected dwarf AGN samples.","tokens_in":16718,"tokens_out":3687,"duration_ms":32068,"significance":"If the central claims hold, the paper offers a useful data-driven visualization of AGN selection functions in the dwarf galaxy regime and a potential route to decontaminate WISE-selected dwarf AGN samples. The use of an external AGN compilation and the authors' explicit test of the WISE-input circularity are commendable. However, the most load-bearing interpretation—that the left WISE clump is AGN-dominated and the right clump is starburst-contaminated—rests on emission-line proxy maps that are not quantitatively defined, and the clump split itself is based on visual inspection without robustness tests. The paper therefore presents a plausible and interesting atlas, but the novel quantitative claims are not yet established to the standard required for a journal publication.","major_comments":[{"comment":"The central interpretation that the left WISE clump contains genuine AGNs while the right clump is contaminated by starbursts rests almost entirely on the [O ii]/z and [O iii]/([O ii]×z) maps. The manuscript never defines how the [O ii] and [O iii] fluxes are measured: are they from SDSS spectroscopy, are they continuum-subtracted, are dust-extinction corrections applied, and how are the line fluxes matched to the photometric apertures used for the SOM? Moreover, the [O iii]/([O ii]×z) ratio is structurally sensitive to low [O ii] flux: red, quiescent galaxies—exactly the population identified with the left clump—will show large values of this ratio even if their [O iii] flux is modest. The paper should report the underlying [O ii] and [O iii] fluxes or equivalent widths for the two clumps separately, and demonstrate the AGN nature of the left clump using a standard diagnostic such as [O iii]/Hβ or the BPT classification where available.","section":"§4.2, Figures 4 and 5"},{"comment":"The division of the 83 WISE-selected AGNs into 55 left-clump and 22 right-clump objects is based on visual inspection of the overplotted SOM without a quantitative definition of the clump boundaries. There is no specification of which cells belong to each clump, no test of sensitivity to SOM hyperparameters (grid size, initialization, training length), and no significance test that the two clumps are statistically distinct in their SED properties. A reproducible clump-membership criterion and a stability analysis of the split are needed before the two-component interpretation can be evaluated.","section":"§4.2, Figure 6"},{"comment":"The authors acknowledge that WISE bands are part of the SOM input and that this can enhance the apparent clustering of WISE-selected AGNs. They state that after retraining the SOM without WISE photometry, the WISE-selected sources still occupy broadly similar regions with a less concentrated clumpy structure, but they provide no figure, quantitative measure, or comparison of clump membership for this retrained map. Since the W1−W2 color is itself a WISE AGN selection criterion, the persistence of the large-scale structure is essential to separate a genuine host-property selection function from a built-in color cut. The retraining result should be shown and quantified.","section":"§4.2, page 11"},{"comment":"The comparison of spatial distributions across selection methods (e.g., the claims that X-ray and variability-selected AGNs are 'uniformly distributed' or that BPT-selected AGNs cluster in high-mass regions) is made by visual inspection, with no statistical test or uncertainty estimates. Given the small numbers in some categories, the authors should provide at least simple quantitative measures—such as the distribution of SOM cell densities relative to the training sample, or a two-dimensional Kolmogorov–Smirnov test—to support the stated differences between selection methods.","section":"§4.1 and §4.2"}],"minor_comments":[{"comment":"The NSA catalog version is given as nsa_v1_0_2 in §2.1 and as nsa_v1_0_1 in §5; please correct the inconsistency.","section":"§2.1 vs §5"},{"comment":"The reference to Wasleske & Baldassare (2024) appears as 'compiled by ?' in the summary section; the citation needs to be completed.","section":"§5"},{"comment":"There is a typo in 'low metallicites' in the introduction, and the title contains a spurious space in 'T echniques' due to formatting; please fix both.","section":"§1 and title"},{"comment":"The caption begins with 'T op row' instead of 'Top row'; this typo should be corrected.","section":"Figure 2 caption"},{"comment":"The text says the cross-match used the WISE All-Sky Source catalog and the nine-band photometry includes only W1 and W2, but the earlier sentence mentions four WISE bands; please clarify why W3 and W4 are not used in the SOM input.","section":"§2.2"},{"comment":"The coronal-line emitter section gives 57 cross-matched sources and 48 unique ones, but the matching radius and selection details are not described; adding a sentence on the cross-match and completeness would help.","section":"§4.3"},{"comment":"The color–mass diagrams would benefit from a brief description of how the training galaxies in the left/right clump cells are selected and how many galaxies each contour represents, since the current caption relies on visual comparison with Schawinski et al. (2014) reference lines.","section":"Figure 9"}],"recommendation":"major_revision","confidential_remarks":"The paper has a useful descriptive core, but the novel WISE-clump interpretation currently rests on unquantified emission-line proxies and a visually defined clump split. I considered whether this warrants rejection, but the issues are fixable within the paper's scope: the authors can add line-flux definitions, standard AGN diagnostics for the clumps, quantitative clump membership, and a shown retraining test. If those additions are made, the paper would be a solid contribution to dwarf-galaxy AGN selection methods."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper applies self-organizing maps to ~30,000 local dwarf galaxies and maps 438 previously identified AGNs onto the resulting SED manifold. The genuinely new contribution is the empirical separation of AGN diagnostics in SED space and the observation that WISE-selected AGNs split into two clumps. The authors also do a useful control: retraining the SOM without the WISE bands and finding that the large-scale structure persists. That is real evidence against the worry that the WISE clumps are purely a training artifact.\n\nWhere the paper is soft is the interpretation of those clumps. The claim that the right clump is starburst contamination and the left clump is AGN-dominated rests on the [O iii]/([O ii]×z) proxy in Figures 4 and 5. The stress-test concern is right: dividing by [O ii], a star-formation tracer, means low-[O ii] quiescent galaxies in the left clump get inflated ratios regardless of AGN content. The paper never reports the separate [O ii] and [O iii] fluxes for the two clumps, and it never defines continuum subtraction, extinction correction, or aperture matching. Without evidence that [O iii] itself is enhanced in the left clump, the starburst/AGN dichotomy is not established. That is the load-bearing weakness.\n\nThere are also no significance tests or error bars anywhere; the 55 vs 22 clump counts come from a visually drawn boundary, and the line-ratio maps are compared by eye. The paper is honest about the WISE-input circularity and addresses it, but the line-ratio issue is unacknowledged and more serious. Minor reproducibility point: no code or data release is mentioned.\n\nWho gets value: anyone working on dwarf AGN selection, survey design, or SOM-based SED analysis. The framework is reusable and the question matters. It deserves a serious referee rather than a desk rejection, but the referee should push for a robust treatment of the line-ratio proxies, significance testing, and public code/data. I would not cite the clump interpretation as established, but the diagnostic-comparison framework may be worth citing once the proxies are validated.","headline":"A useful SOM-based map of dwarf AGN selection, but the WISE clump interpretation rests on an unvalidated line-ratio proxy.","tokens_in":17281,"tokens_out":2707,"would_cite":false,"duration_ms":23010,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Unsupervised machine learning on 30,344 dwarf galaxy SEDs reveals that each AGN detection method selects a distinct host population, with WISE-selected AGNs splitting into a starburst-like clump and a redder, AGN-like clump.","keywords":["AGN selection","dwarf galaxies","self-organizing maps","spectral energy distributions","WISE mid-infrared selection","selection bias","starburst contamination","unsupervised machine learning"],"falsifier":"Conduct a spectroscopic follow-up of WISE-selected dwarf AGNs from both SOM clumps and measure dust-corrected BPT line ratios or coronal emission lines. If the bluer clump shows the same AGN fraction as the redder clump—or if the two clumps have equal X-ray detection rates—the claimed separation into starburst contaminants and true AGN hosts fails.","tokens_in":16254,"feed_emoji":"🔭","tokens_out":12037,"duration_ms":91168,"temperature":0.7,"pith_summary":"This paper shows that the way we find active black holes in dwarf galaxies is heavily shaped by the type of galaxy each detection method can see. Using a self-organizing map (SOM, a machine-learning grid that places similar galaxy colors next to each other) trained on the photometric colors of 30,344 dwarf galaxies, the authors put 438 previously known dwarf AGNs on a single map of spectral energy distributions and find that AGNs selected by different diagnostics cluster in different, partially overlapping regions. BPT-selected AGNs (found through optical emission-line ratios) sit in higher-mass hosts, X-ray and variability-selected AGNs spread widely, and WISE-selected AGNs (found through mid-infrared colors) concentrate in low-mass hosts where they form two clumps: one bluer and starburst-like, the other redder with AGN-like spectral signatures. If true, this gives a concrete measure of the selection bias and contamination in each technique, and the WISE clump split offers a way to clean starburst contaminants out of infrared-selected dwarf AGN samples.","feed_headline":"WISE dwarf AGNs split into two clumps; one is starburst contaminants","feed_subtitle":"A map of 30,000 dwarf galaxy SEDs separates likely true AGN hosts from star-forming contaminants.","key_machinery":"The load-bearing object is a $51\\times 51$ Self-Organizing Map (SOM), an unsupervised neural network that compresses the seven-dimensional color space built from nine photometric bands (FUV, NUV, $u$, $g$, $r$, $i$, $z$, W1, W2) into a two-dimensional grid while preserving which spectral energy distributions resemble which. It is trained on 30,344 dwarf galaxies from the NSA catalog with $z<0.055$ and $M_*<10^{9.5}\\,M_\\odot$, and the 438 known AGNs are overlaid cell by cell. The argument that the two WISE clumps differ physically is carried by two proxy maps: $[\\mathrm{O\\,II}]/z$-band flux for star-formation activity and $[\\mathrm{O\\,III}]/([\\mathrm{O\\,II}]\\times z)$ for AGN luminosity relative to stellar light, supplemented by visual inspection of SDSS images and color–mass diagrams for the clump galaxies.","core_discovery":"On the paper's own terms, the central discovery is that AGN diagnostics are not interchangeable windows onto the same dwarf-galaxy population: each method samples a distinct and only partially overlapping region of the spectral energy distribution manifold. The strongest new result is that WISE-selected dwarf AGNs appear in two separated clumps on the map, with the right clump made of bluer, actively star-forming galaxies and the left clump of redder galaxies whose spectral proxies point to AGN activity; the left clump overlaps heavily with AGNs found by BPT, broad-line, X-ray, [O I], [S II], and He II selection, while the right clump overlaps with regions of high $[\\mathrm{O\\,II}]/z$ star-formation tracers. The paper interprets this as the SOM separating true AGN hosts from starburst contaminants in WISE selection, and it further finds that a subset of AGNs in low-mass galaxies occupy cells with high $[\\mathrm{O\\,III}]/([\\mathrm{O\\,II}]\\times z)$, indicating luminous AGNs hosted by faint dwarfs.","pith_inferences":["The paper stops at mapping the clumps; a direct next step would be to convert the SOM into a per-cell completeness function, so that each diagnostic's recovery fraction can be integrated over the manifold to produce method-corrected AGN occupation fractions for dwarf galaxies.","The left/right WISE clump split implies that a simple cut in SOM coordinates—roughly a particular combination of near-UV to mid-infrared colors—could serve as an empirical contamination filter, although the paper does not prescribe the cut itself.","Because the SOM preserves topology, the same trained map could be applied to higher-redshift dwarf galaxies in deeper surveys, transferring the measured selection biases to regimes where current diagnostics are even less reliable."],"forward_implications":["BPT-selected dwarf AGN samples are systematically biased toward the high-mass end of the dwarf population, so occupation fractions built on BPT alone undercount low-mass, metal-poor hosts.","WISE-selected dwarf AGN samples contain a measurable starburst-contaminated subpopulation that can be identified by its position on the SED manifold, making the clump split a practical contamination filter.","Variability-selected AGNs distribute almost uniformly across the SED manifold, suggesting that upcoming time-domain surveys will deliver a much less host-biased dwarf AGN census than line- or infrared-selected samples.","Emission-line, broad-line, and WISE selection all avoid hosts with strong star formation, so the union of current diagnostics is missing a population of dwarf AGNs embedded in actively star-forming galaxies.","Cells with high $[\\mathrm{O\\,III}]/([\\mathrm{O\\,II}]\\times z)$ that host AGNs in low-mass galaxies imply luminous AGNs hiding in faint dwarfs, recoverable by combining manifold position with line-ratio information."],"supporting_citations":[{"why":"Compiles the 733 dwarf AGNs across diagnostics; the 438 matched to the NSA catalog form the test set mapped onto the SOM.","marker":"Wasleske & Baldassare 2024"},{"why":"Sets the dwarf galaxy mass limit and documents why BPT line diagnostics lose low-mass, metal-poor AGNs.","marker":"Reines et al. 2013"},{"why":"Establishes WISE mid-infrared selection as effective for low-mass AGNs while warning of dusty starburst contamination.","marker":"Satyapal et al. 2014"},{"why":"Supports the interpretation that WISE selection favors compact, gas-rich dwarfs and suffers starburst contamination in this regime.","marker":"Hainline et al. 2016"},{"why":"Introduces self-organizing maps, the unsupervised learning method the paper applies.","marker":"Kohonen 1982"},{"why":"Define the BPT emission-line diagnostics whose selection region the map reveals as high-mass biased.","marker":"Baldwin et al. 1981; Kewley et al. 2006"},{"why":"Provides the WISE mid-infrared color criterion used to classify one set of dwarf AGNs.","marker":"Stern et al. 2012"},{"why":"Prior application of SOMs to quantify AGN selection bias in a spectroscopic survey, extended here to dwarf galaxies.","marker":"Sanjaripour et al. 2024"}],"fun_headline_variants":["WISE dwarf AGNs split into two clumps: true AGNs vs starbursts","SOM maps separate true dwarf AGNs from starburst impostors","Dwarf AGN selection biases exposed by machine learning map","WISE-selected dwarf AGNs show two distinct SED populations","Starburst contaminants found in WISE dwarf AGN selection"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The interpretation of the two WISE clumps as AGN versus starburst contamination assumes the $[\\mathrm{O\\,II}]/z$ and $[\\mathrm{O\\,III}]/([\\mathrm{O\\,II}]\\times z)$ proxy maps are faithful tracers of star formation and AGN strength, but the paper does not define the emission-line flux measurements, leaving open whether dust extinction or host dilution biases those proxies.","fun_headline_variants_meta":{"raw":{"variants":["WISE dwarf AGNs split into two clumps: true AGNs vs starbursts","SOM maps separate true dwarf AGNs from starburst impostors","Dwarf AGN selection biases exposed by machine learning map","WISE-selected dwarf AGNs show two distinct SED populations","Starburst contaminants found in WISE dwarf AGN selection"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000343,"raw_usage":{"total_tokens":1976,"prompt_tokens":1125,"completion_tokens":851,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":741,"completion_tokens_details":{"reasoning_tokens":756}},"tokens_in":741,"tokens_out":851,"duration_ms":6822,"temperature":1.0,"reasoning_tokens":756,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:58:39.931298+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Conduct a spectroscopic follow-up of WISE-selected dwarf AGNs from both SOM clumps and measure dust-corrected BPT line ratios or coronal emission lines. If the bluer clump shows the same AGN fraction as the redder clump—or if the two clumps have equal X-ray detection rates—the claimed separation into starburst contaminants and true AGN hosts fails.","supporting_citations":[],"review_version":1}