{"id":"17b7cc0c-873c-4e7c-bda4-00647fc34755","arxiv_id":"2506.08439","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A supervised object detection pipeline plus visual checks on ASKAP EMU survey data yields five new Odd Radio Circles and a speculative new class of diffuse radio sources called GLAREs.","lead":"Astronomers used a machine learning filter trained on known radio galaxies to narrow a three-million-source sample from the first year of the EMU survey down to 1,794 visual-inspection candidates. They report five new Odd Radio Circles, two further candidates, 55 diffuse radio-emission galaxies, and 18 starburst ring galaxies.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The five new ORCs are supported only by visual ring morphology plus a positionally coincident host; the paper itself admits (§5.1) that such rings can be relic lobes or double-lobed projections, so the claimed population growth is not yet secure.","rationale":"The central claim is a catalogue claim (five new ORCs, two candidates), not a physical-model claim, so I did not require proof of the ORC formation mechanism. The detection pipeline is well described, the code is released, and the Gal-DINO metrics (AP50 73%, keypoint accuracy <3'') give some confidence that candidate selection was consistent. My concern is narrower: the step from 'round radio structure with a central WISE galaxy' to 'ORC' is exactly the step the paper's own definition makes consequential. The paper excludes relic lobes and double-lobed projections but supplies no observable that discriminates those alternatives for the new sample. The self-acknowledged ambiguities for J0402-5321, J1313-4709, and J0510-5825 are not cosmetic; they concern the physical association that defines an ORC. The later removal of previously published ORCs from the ORC list (footnote j) is direct evidence that visual morphological classification is revisable, and the new sample deserves the same scrutiny before the population-growth claim is treated as secure. I do not see an internal inconsistency in the detection methodology itself, and the GLARE/SRRG catalogues are useful regardless. Thus the verdict remains CONDITIONAL, unchanged from the reader's assessment.","tokens_in":28076,"tokens_out":7487,"duration_ms":102319,"concrete_test":"Obtain 5''-resolution radio imaging (MeerKAT or ATCA) and optical spectra for all DESI/WISE galaxies inside the rings of the five ORCs, then recompute host association with a strict chance-coincidence threshold (p<0.01 from local galaxy densities). If J0402-5321 resolves into two lobes or its boundary galaxies have inconsistent redshifts, or if J1313-4709 has no host consistent with a ~300-kpc physical ring, downgrade those entries from 'ORC' to 'candidate'; only the remaining systems should count toward the claimed population increase.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's 'five new ORCs' is the central claim. In §4 it is supported by three kinds of evidence: an edge-brightened circular structure on 15'' EMU images, absence of a HASH/Green counterpart, and a WISE/DESI galaxy near the centre. The first two establish 'round and non-Galactic', not 'ORC'. The ORC definition in §4 explicitly excludes rings from relic lobes of double-lobed radio galaxies, yet §5.1 concedes that some GLARE-like systems 'could be classical double-lobed or remnant radio galaxies viewed near their major radio axis' and that hostless diffuse sources might be 'fading relic radio lobes with re-energized electrons' (Shabala et al. 2024). For the new sample no spectral-index, polarization, or high-resolution data are presented that would rule out these mimics. Host association is weakest where the text is most explicit: J0402-5321 has three galaxies within the ring and two with photometric redshifts consistent with the central galaxy; J1313-4709 has no redshift and additional possible associated sources; §4.6 says J0510-5825 cannot be conclusively tied to its central galaxy. Only one of the five hosts has a spectroscopic redshift, so chance alignment is not quantified. Footnote j in §5.1 shows that previously published ORCs were later removed from the ORC list, so this classification step is the load-bearing one; if any of the five is a relic-lobe projection or group/cluster diffuse emission, the claimed expansion of the ORC population is wrong.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a systematic search for Odd Radio Circles (ORCs) and other unusual radio morphologies in the first year of the EMU survey, using the Gal-DINO object detection model to reduce ~3 million radio detections to 1,794 visually inspected candidates. The authors report five new ORCs, two additional ORC candidates, 55 GLAREs, and 18 SRRGs, and propose that GLAREs may be evolutionary precursors or descendants of ORCs. The ORC identification rests on visual ring morphology, absence of Galactic counterparts, and a positionally coincident WISE/DESI galaxy, with flux densities, sizes, and photometric redshifts tabulated in Table 1.","tokens_in":28397,"tokens_out":6529,"duration_ms":71372,"significance":"If the five new ORCs are confirmed, the known population would more than triple, making this a significant contribution to the study of a rare and poorly understood class. The paper's strengths include public availability of the model and data, a transparent candidate table, and the independent verification of photometric redshifts from multiple catalogues. The main weakness is that the central claim---that these are genuine ORCs---depends on classification criteria that the paper itself acknowledges are difficult to separate from double-lobed radio galaxies, relic lobes, or unrelated diffuse emission. The stress-test concern is valid: the evidence presented in §4 establishes round, non-Galactic radio structures but does not yet establish the strict ORC definition cited from Norris et al. (2025).","major_comments":[{"comment":"The paper's own definition of an ORC in §4 explicitly excludes rings originating from the relic lobes of double-lobed radio galaxies, yet §5.1 concedes that some GLAREs 'could be classical double-lobed or remnant radio galaxies viewed near their major radio axis' and that hostless diffuse sources might be 'fading relic radio lobes with re-energized electrons' (citing Shabala et al. 2024). Footnote j in §5.1 further notes that previously published ORC2 and ORC3 have now been removed from the ORC list (Norris et al. 2025). For the five new objects, the paper provides no spectral-index, polarization, or higher-resolution data that would rule out these mimics. The evidence in §4 (edge-brightened ring, absence of HASH/Green counterparts, and a WISE/DESI galaxy near the centre) establishes 'round and non-Galactic' but not 'ORC' in the restricted sense used by the authors. I recommend either adding discriminating observations or changing the claim to 'ORC candidates' throughout the abstract and conclusions.","section":"§4 and §5.1"},{"comment":"Host association is the load-bearing step for each claimed ORC, but it is only position-based for all but one object. Only J2304-7129 has a spectroscopic redshift; the other four hosts rely on photometric redshifts, and J1313-4709 has no redshift at all. The paper does not quantify the chance-alignment probability between the radio ring and a random WISE galaxy. In several cases the text itself acknowledges the ambiguity: §4.2 says three galaxies within the ring of J0402-5321 'may have contributed' to the circular structure, §4.1 mentions faint galaxies near the ring that 'may have contributed to the observed radio emission', and §4.6 states for candidate J0510-5825 that 'we cannot conclusively determine that the circular emission originates solely from the central galaxy'. A positional-coincidence test, even a simple surface-density argument using WISE galaxy counts, is needed to support the claim that these rings are associated with the designated hosts.","section":"Table 1 and §§4.1–4.6"},{"comment":"The search methodology and the associated expected-number estimate both rest on uncharacterized selection effects. The Gal-DINO model was trained on previously known ORCs and rare morphologies and then applied to search for more ORCs, which is a reasonable discovery strategy but not one that allows completeness claims. The confidence threshold of 0.7 was set after visual inspection of a subset of the same survey tiles, and no independent validation set is used to estimate false positives or completeness. Similarly, the statement in §5.1 that 'we would expect approximately 33 ORCs in the ~4,500 deg2' is based on two ORCs in a 270 deg² pilot field, which is a very small number and ignores the selection function of the ML pipeline. The paper's phrase 'systematic search' in the abstract should therefore be tempered, and the expected-number comparison should be presented with appropriate uncertainties or removed.","section":"§3.3 and §5.1"}],"minor_comments":[{"comment":"The host name for ORC J0452-6231 is given as WISEA J045230.76−623123.8 in Table 1 but as WISEA J045221.40−623422.7 in the text; these are inconsistent and should be reconciled.","section":"§4.3"},{"comment":"The source J0318-5708 is listed in Table 5 as a diffuse source without a plausible host, but in the text it is cited as an example of a GLARE with multiple galaxies at similar redshifts; the classification should be made consistent.","section":"§5.1"},{"comment":"The text refers to 'The source J1407-0197, shown in Figure 6'; this appears to be a typo for J1407-0917, which is the source displayed in Figure 6 and listed in Table 5.","section":"§5.1"},{"comment":"The caption contains the typo 'Eaxamples' and should read 'Examples'.","section":"Figure 7 caption"},{"comment":"The phrase 'Greens's catalogue' should be 'Green's catalogue'.","section":"§4.6"},{"comment":"In the reference list, the Parker et al. entry contains 'V ol' instead of 'Vol.', and the Introduction refers to 'MW A' instead of 'MWA'.","section":"References"},{"comment":"The phrase 'the model's minimum estimated threshold' is not defined in this paper; the reader must consult Gupta et al. (2024b) to understand how that threshold is derived.","section":"§3.3"},{"comment":"The expected number of ~33 ORCs is quoted without an uncertainty; given that it is based on two detections, a Poisson error estimate would be more appropriate.","section":"§5.1"}],"recommendation":"major_revision","confidential_remarks":"I recommend sending the manuscript back for major revision. The authors should be asked to either provide additional discriminating evidence (spectral index, polarization, or high-resolution imaging where available) or explicitly re-label the five new ORCs as candidates in the title, abstract, and conclusions. The GLARE catalogue is a useful community resource and should be preserved, but the evolutionary connection to ORCs is speculative and should be clearly framed as such."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does something genuinely useful: it runs a supervised object detection model over ~3 million radio sources from the first year of EMU, cuts that to 1,794 candidates for visual inspection, and ends up with a well-documented catalog of candidate ORCs, GLAREs, and SRRGs, with images and code available. That is a real contribution to the subfield, and the method is described clearly enough that others can reproduce it.\n\nWhat the paper does not do is establish that the five objects in Table 1 are ORCs. Edge-brightened round radio emission with no HASH or Green counterpart is good evidence for 'round and non-Galactic'—it is not evidence for 'ORC' as defined in §4, because that definition explicitly excludes relic lobes of double-lobed radio galaxies. The paper's own §5.1 concedes that some GLARE-like systems could be classical double-lobed or remnant radio galaxies viewed near the axis, and no spectral-index, polarization, or high-resolution data are presented to rule out those mimics for the five new rings. The host associations are also fragile: only one host has a spectroscopic redshift; J0402-5321 has three galaxies inside the ring, J1313-4709 has no redshift at all, and J0510-5825 is explicitly described as not conclusively tied to its central galaxy. None of this is fatal for a discovery paper, but it means the abstract should say 'candidate ORCs' rather than 'ORCs.' footnote j in §5.1 is a reminder that prior apparent ORCs were later removed after follow-up, so this classification step is load-bearing.\n\nThe GLARE class is new and plausibly useful, and the speculation that some GLAREs may be evolutionary stages of ORCs is honestly labeled as a hypothesis. The weakest part is the count comparison in §5.1: they expect ~33 ORCs from the PS1 rate, then 'align' the observed counts by adding half the GLAREs as possible ORC stages. That uses the hypothesis to confirm the hypothesis; it should be flagged as illustrative, not quantitative. The 0.7 confidence threshold was set by inspecting a subset of the same survey, so completeness is unknown—fine for a candidate search, but not a measured rate.\n\nMy recommendation: send it to a serious referee. The catalog, the method, and the candidate lists deserve publication and follow-up. But the authors should be asked to soften the abstract, move the central claim to 'candidates,' and relegate the count comparison to a clearly labeled speculation. If they make those changes, this is a solid PASA paper.","headline":"A useful ML search pipeline and a plausible set of new ORC candidates, but the central claim that these are five new ORCs is stronger than the evidence supports.","tokens_in":28998,"tokens_out":2915,"would_cite":false,"duration_ms":37358,"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":"Five new Odd Radio Circles discovered in the first-year EMU survey data.","keywords":["Odd Radio Circles","radio morphology","EMU survey","ASKAP","object detection","machine learning","diffuse radio emission","starburst rings"],"falsifier":"A deep, arcsecond-resolution radio image of any of the five ORCs and two candidates, combined with optical spectroscopy of every galaxy inside the ring, would settle the classification: if the ring resolves into two opposed radio lobes connected by jets, if no galaxy lies near the ring centre, or if the designated host's redshift places it far from the ring's centre, then the ORC interpretation fails.","tokens_in":27867,"feed_emoji":"📡","tokens_out":6919,"duration_ms":83789,"temperature":0.7,"pith_summary":"This paper reports that a supervised object-detection model, applied to 4,500 square degrees of the first-year EMU survey, identified five new Odd Radio Circles (ORCs) and two candidate ORCs, growing the known population from five to at least ten confirmed or candidate systems. It also returns 55 Galaxies with Large-scale Ambient Radio Emission (GLAREs), diffuse rings or blobs around distant galaxies that may be precursors or later-stage ORCs, and 18 Starburst Radio Ring Galaxies (SRRGs). A hybrid pipeline filtered roughly three million radio sources down to 1,794 candidates for human visual inspection, making the discovery efficient at survey scale. The paper argues that ORCs remain edge-brightened rings around host galaxies with no corresponding optical or infrared ring, and that the GLARE population offers a way to test evolutionary links between these structures.","feed_headline":"Machine learning finds five new Odd Radio Circles","feed_subtitle":"A supervised detector sifts 3 million radio sources and adds two ORC candidates plus 55 GLARE galaxies for follow-up.","key_machinery":"The machinery is a supervised object-detection system built around Gal-DINO, a transformer-based detector trained on RadioGalaxyNET and a set of atypical radio morphologies. Extended by the RG-CAT pipeline, it scans 8-arcminute cutouts around each of roughly three million Selavy-detected radio sources, predicts a bounding box, a category, and a keypoint marking the likely infrared host, and ranks sources by confidence. Filtering for the rare-or-peculiar category at a confidence threshold of 0.7 cuts the sample to 1,794 sources for visual inspection. This is the mechanism that turns a survey-wide search into a manageable human review while keeping the ORC-like morphologies.","core_discovery":"The central discovery is that machine-learning object detection can systematically find a rare radio morphology that was previously found by chance. Applying the Gal-DINO detector and the RG-CAT catalogue pipeline to 160 tiles of the first-year EMU main survey, the authors identify five new ORCs (J0210-5710, J0402-5321, J0452-6231, J1313-4709, and J2304-7129) and two unconfirmed candidates (J0510-5825 and J1104-6351), each an edge-brightened circular radio structure with a distant galaxy near its centre and no detectable extended optical or infrared ring. All but one host lack spectroscopic redshifts, with photometric redshifts placing the systems at roughly 0.1 to 0.6 and physical diameters of roughly 160 to 440 kiloparsecs. In the same sample the paper classifies 55 GLAREs and 18 SRRGs, and proposes that GLAREs may be precursors or later evolutionary stages of ORCs. If accepted, the known ORC population grows from the five objects in the literature to at least ten, giving the first statistical footing for formation models.","pith_inferences":["If the ORC-GLARE evolutionary sequence is real, rectangular GLAREs viewed along their long axis may be classical double-lobed or remnant radio galaxies, which would tie ORCs to the life cycle of dying radio galaxies and predict spectral-index gradients across the rings.","The same detector, retrained on the appropriate resolution, should transfer to northern surveys such as LoTSS and to future SKA data, potentially revealing ORC-like rings in sky areas not covered by EMU.","Because most host redshifts are photometric, the derived sizes, luminosities, and star formation rates carry large uncertainties; spectroscopic redshifts for the five hosts could shift which galaxy is considered the central host and change the physical interpretation.","A direct test of the paper's implied surface density is to run the identical pipeline on the remaining EMU tiles as they are released; finding far fewer or far more than the expected rate would indicate selection effects or a change in ORC frequency with environment."],"forward_implications":["The known ORC population grows from five to at least ten confirmed or candidate systems, providing a larger sample for testing formation mechanisms.","The rate of two ORCs in 270 square degrees of the pilot survey implies roughly 33 ORCs in the 4,500 square degrees covered here, so the full EMU survey should yield many more.","The 55 GLAREs give a substantial sample for testing whether diffuse ambient radio emission evolves into edge-brightened rings through shocks from mergers or starbursts.","Reducing roughly three million sources to 1,794 visual inspections shows that supervised object detection can make searches for rare radio morphologies practical as next-generation surveys grow larger.","Multi-wavelength follow-up of the new systems can distinguish between the relic-lobe shock model and the spherical-shell shockwave model for ORC formation."],"supporting_citations":[{"why":"Defines ORCs and reports the first serendipitous detections in EMU Pilot Survey 1 that this work extends.","marker":"Norris et al. (2021b)"},{"why":"Shows that machine learning can rediscover and find ORCs, providing the pilot for this search and one previously known ORC.","marker":"Gupta et al. (2022)"},{"why":"Introduces RadioGalaxyNET and the Gal-DINO detector used to classify the EMU cutouts.","marker":"Gupta et al. (2024a)"},{"why":"Describes the RG-CAT pipeline that turns Gal-DINO predictions into the source catalogue filtered here.","marker":"Gupta et al. (2024b)"},{"why":"Presents the first-year EMU main-survey data and source catalogues that form the search input.","marker":"Hopkins et al. (2025)"},{"why":"Updates the ORC definition and the list of known ORCs, the benchmark against which the new detections are compared.","marker":"Norris et al. (2025)"},{"why":"Provides the fading-relic-lobe shock model invoked to explain both ORCs and some diffuse radio sources.","marker":"Shabala et al. (2024)"}],"fun_headline_variants":["AI detector spots five new Odd Radio Circles","Machine learning finds 5 new ORCs in EMU data","Five new Odd Radio Circles found via object detection","EMU survey's first year yields 5 new Odd Radio Circles"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The search's central premise is that each edge-brightened radio ring is a real ORC physically associated with the WISE galaxy at its centre, rather than a chance projection of a double-lobed radio galaxy, a cluster radio relic, an imaging artifact, or an unrelated galaxy along the line of sight.","fun_headline_variants_meta":{"raw":{"variants":["AI detector spots five new Odd Radio Circles","Machine learning finds 5 new ORCs in EMU data","Five new Odd Radio Circles found via object detection","EMU survey's first year yields 5 new Odd Radio Circles"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000511,"raw_usage":{"total_tokens":2520,"prompt_tokens":1015,"completion_tokens":1505,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":631,"completion_tokens_details":{"reasoning_tokens":1440}},"tokens_in":631,"tokens_out":1505,"duration_ms":14371,"temperature":1.0,"reasoning_tokens":1440,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T05:11:02.085669+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A deep, arcsecond-resolution radio image of any of the five ORCs and two candidates, combined with optical spectroscopy of every galaxy inside the ring, would settle the classification: if the ring resolves into two opposed radio lobes connected by jets, if no galaxy lies near the ring centre, or if the designated host's redshift places it far from the ring's centre, then the ORC interpretation fails.","supporting_citations":[{"cited_title":"P., et al","cited_arxiv_id":null,"evidence_quote":"Shows that machine learning can rediscover and find ORCs, providing the pilot for this search and one previously known ORC."},{"cited_title":"P., Koribalski , B","cited_arxiv_id":null,"evidence_quote":"Updates the ORC definition and the list of known ORCs, the benchmark against which the new detections are compared."},{"cited_title":"S., Yates-Jones , P","cited_arxiv_id":null,"evidence_quote":"Provides the fading-relic-lobe shock model invoked to explain both ORCs and some diffuse radio sources."}],"review_version":1}