{"id":"3a55d67c-7a2c-4605-80ca-41972ce57de5","arxiv_id":"2505.07541","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"The SUPAX design report introduces a tunable multi-cavity haloscope and reports new dark photon exclusion limits around 35 µeV.","lead":"SUPAX is a proposed haloscope experiment that uses a 12 T magnet and tunable superconducting cavities to search for axion and dark photon dark matter. Its prototype has already placed new exclusion limits on dark photons around 35 µeV, and the paper details the design and projected sensitivity.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 95% dark-photon exclusion rests on a re-scan threshold set to the largest fluctuation in the same dataset, so the quoted confidence level is not statistically justified.","rationale":"The paper is primarily a conceptual design report plus a prototype measurement, and the engineering and R&D content is credible and useful. However, the strongest claim advertised in the abstract is the exclusion of dark photons, and an exclusion lives or dies on the statistical procedure. The reader identified the data-dependent re-scan threshold as the weakest assumption; I agree and would make it the single load-bearing concern. The Savitzky-Golay filtering issue is also real—if the filter window is wider than the dark-photon line, a genuine signal could be absorbed into the baseline—but the paper does not state the filter width, so that concern is less directly evidenced. The threshold problem is explicit in the text: no re-scans were performed, and Θ was set to the largest observed fluctuation in the same data used for the limit. That makes the threshold stochastic and data-dependent, so the 95% coverage claim is not established by the paper. The central result might survive a corrected analysis, and the limit may even be slightly conservative if a fixed threshold were used, but as written the statistical confidence is not supported. This does not change the reader’s CONDITIONAL verdict: the prototype result is a useful first step, but the exclusion claim needs a pre-specified or calibrated threshold before it can be taken at the quoted confidence level.","tokens_in":10930,"tokens_out":7325,"duration_ms":80567,"concrete_test":"Run a Monte Carlo coverage study with the prototype’s actual bin count, bandwidth, and noise level. Generate 10⁴ noise-only spectra; for each, apply the Section 5.3 pipeline exactly, including setting Θ to the maximum observed fluctuation and computing the Eq. (5) 95% upper limit. Then inject simulated dark-photon signals at the claimed boundary, e.g. χ = 5×10⁻¹⁴, in another 10⁴ trials and record the exclusion rate. If the fraction of trials excluding a true boundary signal is below 95%, or if the limit distribution shifts materially when Θ is replaced by a fixed pre-specified threshold (e.g. 5σ plus a trials factor), the data-dependent threshold is confirmed as a bias. A simpler variant is to recompute the published limit with Θ fixed to 5σ and compare the resulting curve to the quoted one.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Section 5.3, after stating that no re-scans were performed, the authors set the re-scan threshold Θ to “the largest observed fluctuation at 4.1σ” and compute the 95% CL sensitivity with R_T = Θ + Φ⁻¹(0.95) = 5.7 via Eq. (5). This makes the threshold a random variable drawn from the same noise realization that is being used to claim exclusion. A fixed 95% confidence procedure requires the threshold (or the false-positive rate) to be chosen before inspecting the data, or the trials factor to be incorporated into the limit. Here the maximum observed fluctuation is exactly what defines “no candidate,” so the resulting interval is a post-hoc description of what this particular noise realization would have allowed, not a pre-specified test with the stated coverage. The headline exclusion χ > 5×10⁻¹⁴ at 95% CL is therefore not statistically certified. The numerical bias may be moderate—a 1σ shift in the observed maximum changes R_T by about 1, roughly 18% in the limit—but the paper provides no coverage or calibration check. Because the central claim is an exclusion limit, the confidence level is the load-bearing element.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper is a conceptual design report for SUPax, a haloscope experiment combining a 12 T magnet, tunable superconducting cavities, and quantum-limited readout to search for axion-like particles in the 8–30 µeV mass range. It also reports on a prototype dark-photon search performed with an 8.47 GHz cavity at 2 K, from which the authors derive upper limits on the kinetic mixing parameter around mass 35 µeV, with a headline value of about 5×10^-14, and project future axion sensitivity reaching the QCD axion band. The manuscript includes technical design details, prototype hardware characterization, data-acquisition and analysis descriptions, and a comparison to existing limits.","tokens_in":11160,"tokens_out":7301,"duration_ms":74909,"significance":"If the dark-photon exclusion is statistically sound, the prototype result is a useful proof of principle: it covers a previously unexplored mass/coupling region, and the full design provides a concrete, falsifiable projection for a next-generation haloscope. The paper also gives useful detail on multi-cavity tuning, superconducting coatings, and an end-to-end analysis chain. However, the central quantitative claim depends on a data-dependent re-scan threshold, so the 95% CL statement is not currently certified. The projected sensitivities rest on unverified assumptions about quantum-limited readout and high-Q cavities, which is acceptable for a design report only if clearly labeled.","major_comments":[{"comment":"The 95% CL dark-photon exclusion shown in Fig. 13 is not statistically justified because the re-scan threshold is taken from the same dataset used to derive the limit. After stating that no re-scans were performed, the authors set Θ to the largest observed fluctuation (4.1σ) and define R_T = Θ + Φ^{-1}(0.95) = 5.7; since Θ is a random draw from the noise realization, the quoted confidence level does not correspond to the coverage of the resulting interval. The paper should either use a pre-specified threshold, incorporate a trials factor for the scanned frequency range, and calibrate coverage with Monte Carlo or injected-signal studies, or present Fig. 13 as an expected/median sensitivity rather than an observed 95% CL exclusion. The numerical bias may be moderate, but no calibration check is provided.","section":"Section 5.3, Eq. (5)"},{"comment":"The analysis is not self-contained: the 'dark photon conversion power profile' used to produce the rescaled spectra is never defined, and Eq. (5) introduces quantities |χ0| and R̃_g^l without giving the expected signal SNR as a function of the kinetic mixing parameter and cavity parameters such as volume, quality factor, mode overlap, noise temperature, and polarization averaging. As a result, the mapping from the measured spectra to the limits in Fig. 13 cannot be reproduced or checked. Please include the full signal model and all numerical inputs.","section":"Section 5.3"}],"minor_comments":[{"comment":"The text says each cavity has a dedicated readout chain and that the system is implemented two times in parallel, while Section 2.2 describes three cavities and Fig. 2 indicates C1–C3; the number of cavities and readout chains should be reconciled.","section":"Section 2.2/2.3"},{"comment":"The design description mentions 'three triple-cavities' in Section 2.2 and 'both cavities' in the Conclusion; please clarify the intended number of cavities and scan configurations.","section":"Section 2.2 and Section 6"},{"comment":"The text quotes a scanned frequency band of 150 kHz around f0, but the horizontal range shown in Fig. 13 spans about 120 kHz; please reconcile the stated and plotted scan range.","section":"Section 5.3 / Fig. 13"},{"comment":"The statement that the normalized-power distribution 'follows nicely a gaussian' would be more informative with a fit result, a χ²/dof value, and a statement of the number of independent frequency bins, since adjacent bins in the grand unified spectrum are not independent.","section":"Fig. 12"},{"comment":"The projected sensitivity curves rest on assumptions such as quantum-limited readout and ReBCO-coated cavities with Q0 ≈ 4.5×10^5; these are appropriate as design goals, but the text should state explicitly which curves are demonstrated by the prototype and which are assumed for the final experiment.","section":"Section 3, Fig. 3"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the statistical calibration issue in Section 5.3 is the main obstacle to acceptance; the hardware and design work are otherwise a reasonable fit for the journal. If the authors supply a calibrated threshold or relabel the result as an expected sensitivity, and provide the missing signal model, the paper could become acceptable after revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nHere's my read on SUPAX. The main thing you should know: it's a serious conceptual design for a multi-cavity haloscope, and the prototype produces a dark photon exclusion at ~35 µeV that looks genuinely new. The soft spot is the statistics behind that limit—the re-scan threshold is set from the same data that defines the exclusion, so the 95% CL label is not as clean as it should be.\n\nWhat's actually new: the three-cavity-in-one-body design with off-center dielectric tuning and helium pressure fine-tuning is a sensible combination I haven't seen in exactly this form. The prototype measurement at 8.471 GHz using a NbN-coated cavity is real work—they measure Q ≈ 3×10^5, scan 300 kHz, and get a limit χ > 5×10^-14 in a window that appears unexplored. The technical sections on the cryostat, magnet, and readout are detailed enough to be useful to someone planning a haloscope. The projected sensitivity for the full experiment follows standard formulas and reaches the QCD axion band if the ReBCO cavities and JPA readout perform as claimed; those are explicitly labeled as targets, not yet demonstrated, so I don't hold that against them.\n\nThe soft spot is the analysis in Section 5.3. They state that no re-scans were performed, then set the threshold Θ to the largest observed fluctuation at 4.1σ in the same dataset used to compute the limit. That makes Θ a random variable from the noise realization under test. The quoted 95% CL is therefore not a pre-specified coverage; it's a post-hoc description of what this particular noise run would have allowed. The numerical bias may be moderate—a 1σ move in the maximum changes R_T by about 1, roughly 18% on the limit—but there is no coverage check. This is fixable: use a fixed threshold chosen before the scan, apply a trials factor for the number of independent frequency bins, or at least calibrate with Monte Carlo noise-only simulations. As written, the central claim is not statistically certified. Minor issues: Section 2.3 says the readout is implemented \"two times in parallel\" while there are three cavities, and the introduction misnumbers the sections. Small stuff, but symptomatic of a rushed draft.\n\nWho gets value from this: experimentalists working on haloscope design and cavity R&D, and anyone tracking dark photon limits in the 30–40 µeV range. The sensitivity projections are useful as a benchmark even if the experiment never reaches them.\n\nMy recommendation: send it to peer review, but with a request that the dark photon limit be re-derived using a statistically sound procedure. The design content deserves a referee; the current limit claim needs more work.\n\nBest,\n[You]","headline":"A credible haloscope design report with a genuinely new dark-photon exclusion, but the limit's confidence level is compromised by a data-dependent re-scan threshold.","tokens_in":11648,"tokens_out":2670,"would_cite":true,"duration_ms":25255,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A prototype haloscope using a superconducting cavity excludes previously unexplored dark photons around 35 µeV with kinetic mixing down to χ = 5×10⁻¹⁴, and the full SUPAX design projects sensitivity to the QCD axion band.","keywords":["axion-like particles","dark photons","haloscope","superconducting cavities","kinetic mixing","tunable microwave cavity","Josephson parametric amplifier","QCD axion"],"falsifier":"The limit could be checked by injecting a calibrated, narrow-band tone through the weakly coupled cavity port at a level corresponding to $\\chi \\approx 5\\times10^{-14}$ and verifying that the full analysis chain recovers it at the expected signal-to-noise ratio; a second check is to recompute the limit with the re-scan threshold fixed before looking at the data (for instance, at 5σ) rather than at the largest observed 4.1σ fluctuation.","tokens_in":10762,"feed_emoji":"🧲","tokens_out":9596,"duration_ms":79560,"temperature":0.7,"pith_summary":"The SUPAX project sets out to search for axion-like particles in the mass range 8 to 30 µeV by converting them into photons in tunable superconducting cavities inside a 12 T magnetic field. The paper's prototype, a superconducting cavity cooled to 2 K, has already produced the project's first physics result: a scan of 300 kHz around 8.471 GHz finds no dark-photon signal and excludes kinetic mixing parameters χ above 5×10⁻¹⁴ at masses near 35 µeV, a part of parameter space not reached by previous experiments. The design report argues that the full experiment, with three simultaneously scanned cavities and a projected quality factor of about 4.5×10⁵, would reach axion-photon couplings below 2×10⁻¹⁵ GeV⁻¹ and enter the QCD axion band. If correct, this would turn a compact haloscope into a competitive probe of both axion dark matter and dark photons with one instrument.","feed_headline":"Prototype search excludes dark photons near 35 µeV down to χ=5×10⁻¹⁴","feed_subtitle":"The prototype probes a 300 kHz band near 8.471 GHz, reaching mixing limits as low as 5×10⁻¹⁴.","key_machinery":"The load-bearing object is the tunable superconducting cavity: a copper body shaped as a tetracontagon (a 40-sided polygon), coated with superconducting tape, whose $\\mathrm{TM}_{010}$ mode is read out through a weakly coupled port. Coarse tuning comes from a piezo-actuated off-center dielectric rod; fine tuning comes from changing helium gas pressure inside the cryostat, which shifts the dielectric constant and moves the resonance frequency by $-27.45\\,\\mathrm{kHz/mbar}$. The signal-search chain, following the analysis procedure of reference [23], removes the cavity response with a Savitzky-Golay filter, rescales spectra by the expected dark-photon conversion power, combines frequency bins into a grand unified spectrum, and converts a null result into a 95% CL limit using a re-scan threshold set to the largest observed fluctuation.","core_discovery":"The paper claims that a haloscope using superconducting cavities can simultaneously reach previously unexplored dark-photon parameter space in prototype form and project sensitivity to the QCD axion band in the final design. The prototype, a copper cavity coated with NbN operated in a 14 T field at 2 K, achieved a quality factor of $Q \\approx 3\\times10^5$ and scanned a 300 kHz band around 8.471 GHz by varying helium pressure. From a null result, the analysis sets 95% CL upper limits on the kinetic mixing parameter in the range $\\chi$ between $1\\times10^{-14}$ and $5\\times10^{-14}$ for dark photon masses near 35 µeV, assuming random polarization. The full experiment is designed around three combined cylindrical cavities with tetracontagon cross-sections, ReBCO-coated superconducting surfaces, and simultaneous multi-frequency scanning between 2 and 7.2 GHz.","pith_inferences":["The quoted limit would be more reliable if the re-scan threshold were fixed before the data were examined; a blind analysis would test whether the 4.1σ threshold contributes to the result.","The pressure-tuning mechanism is generic and could be adopted by other cryogenic cavity searches, since helium gas tuning avoids moving parts in the high-field region.","A simple extension of the prototype scan beyond 300 kHz, using the same linear pressure response, could cover a substantially wider dark-photon mass window at similar sensitivity before the full experiment turns on.","The quality factor of superconducting cavities at these frequencies may be limited by surface defects and coating inhomogeneities; direct comparisons of NbN and ReBCO coatings in the same cavity geometry would test whether the design $Q_0$ is achievable."],"forward_implications":["If the full experiment performs as designed, SUPAX will scan axion masses from 8 to 30 µeV over roughly 80 weeks of data taking starting in 2026, with sensitivity reaching the QCD axion band.","The verified pressure-tuning mechanism, with a measured response of −27.45 kHz/mbar, gives a 550 kHz scan range per 20 mbar pressure change, which allows the mechanical tuning to use coarse steps.","The prototype's dark-photon limit already establishes the readout chain (cavity, cryogenic amplifier, spectrum analyzer, and analysis software) as ready for the final experiment's axion search.","Because the dark-photon search needs no magnetic field, the same cavity technology can run as a standalone dark-photon experiment in the gaps between axion scans.","With superconducting coatings, the design target $Q_0 \\approx 4.5\\times10^5$ would improve the scanning speed by roughly an order of magnitude over the normal-conducting copper configuration."],"supporting_citations":[{"why":"Supplies the scanning-speed formula used to project the final experiment's reach toward the QCD axion band.","marker":"[16]"},{"why":"Characterizes the superconducting cavity used in the prototype, including the quality factor that sets the dark-photon search sensitivity.","marker":"[22]"},{"why":"Provides the four-step analysis procedure that converts recorded spectra into the grand unified spectrum and the exclusion limits.","marker":"[23]"},{"why":"The Savitzky-Golay filter that removes the cavity response and readout artifacts from the raw spectra.","marker":"[24]"},{"why":"The compilation of existing limits against which the new constraint is compared, establishing that the 35 µeV region was previously unexplored.","marker":"[25]"},{"why":"The source of the prototype cavity design, including the rounded-corner half-shell structure.","marker":"[17]"},{"why":"Reports the NbN coating deposited on the copper cavity that gives the high superconducting quality factor.","marker":"[18, 19]"}],"fun_headline_variants":["Prototype haloscope excludes dark photons at 35 µeV","Superconducting cavity probes dark photon parameter space","New limits on dark photons from SUPAX prototype","Haloscope prototype opens new window to dark matter","Simultaneous scanning boosts axion search in prototype"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The result depends on the rule used to decide when a bump is a candidate: that rule was set to the largest fluctuation seen in the same data that produced the limit, and the smoothing step is assumed not to remove a real signal.","fun_headline_variants_meta":{"raw":{"variants":["Prototype haloscope excludes dark photons at 35 µeV","Superconducting cavity probes dark photon parameter space","New limits on dark photons from SUPAX prototype","Haloscope prototype opens new window to dark matter","Simultaneous scanning boosts axion search in prototype"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000748,"raw_usage":{"total_tokens":3327,"prompt_tokens":938,"completion_tokens":2389,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":554,"completion_tokens_details":{"reasoning_tokens":2313}},"tokens_in":554,"tokens_out":2389,"duration_ms":18786,"temperature":1.0,"reasoning_tokens":2313,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:13:33.526483+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"The limit could be checked by injecting a calibrated, narrow-band tone through the weakly coupled cavity port at a level corresponding to $\\chi \\approx 5\\times10^{-14}$ and verifying that the full analysis chain recovers it at the expected signal-to-noise ratio; a second check is to recompute the limit with the re-scan threshold fixed before looking at the data (for instance, at 5σ) rather than at the largest observed 4.1σ fluctuation.","supporting_citations":[{"cited_title":"Completion of Phase I and Preparation for Phase II of the HAYSTAC Experiment","cited_arxiv_id":"1809.05913","evidence_quote":"Supplies the scanning-speed formula used to project the final experiment's reach toward the QCD axion band."},{"cited_title":"Savitzky and M","cited_arxiv_id":null,"evidence_quote":"The Savitzky-Golay filter that removes the cavity response and readout artifacts from the raw spectra."},{"cited_title":"cajohare/axionlimits: Axionlimits,","cited_arxiv_id":null,"evidence_quote":"The compilation of existing limits against which the new constraint is compared, establishing that the 35 µeV region was previously unexplored."}],"review_version":1}