CUPID: Uranian Moon & Multi-Domain Research

• CUPID is a multifaceted term referring to an inner Uranian moon, a next-generation neutrinoless double beta decay experiment, and several advanced computational systems.
• In astronomy, Cupid’s orbital dynamics exemplify resonance-mediated stability in a fragile, low-mass satellite system, offering insight into tidal and resonant effects.
• In particle physics and computing, CUPID leverages dual-readout bolometry and contextual selection techniques to enhance rare-event detection, matchmaking, and data curation.

CUPID designates both a specific astronomical body and a recurring research acronym across several domains. In astronomy, Cupid is a small inner moon of Uranus whose long-term survival is a sensitive test case for resonance-mediated stability in compact satellite systems. In experimental particle physics, CUPID is the “CUORE Upgrade with Particle IDentification,” a next-generation bolometric search for neutrinoless double beta decay at LNGS. The same name is also used for several recent computational systems and benchmarks, including frameworks for MOBA re-matchmaking, contextual preference inference in LLMs, visual analysis of prompt-conditioned image distributions, duplicate bug report detection, and robot imitation-learning data curation (Ćuk et al., 2022, Group, 2015, Fan et al., 2024, Kim et al., 3 Aug 2025, Zhao et al., 2024, Zhang et al., 2023, Agia et al., 23 Jun 2025).

1. Major referents and acronymic reuse

The term appears in the literature with distinct meanings rather than a single unified concept. In current arXiv usage, the most prominent referents are a Uranian inner moon and a bolometric rare-event experiment, alongside a cluster of computational systems that reuse the acronym for domain-specific purposes.

CUPID referent	Domain	Characterization
Cupid	Planetary science	Inner moon of Uranus (Ćuk et al., 2022)
CUORE Upgrade with Particle IDentification	Neutrino physics	Bolometric $0\nu\beta\beta$ experiment (Group, 2015)
Re-matchmaking system	HCI / games	MOBA fairness and position satisfaction (Fan et al., 2024)
Contextual User Preference Inference Dataset	LLM evaluation	Benchmark for contextual personalization (Kim et al., 3 Aug 2025)
Contextual Understanding of Prompt-conditioned Image Distributions	Visual analytics	Object-centric analysis of text-to-image outputs (Zhao et al., 2024)
ChatGPT-assisted duplicate bug report detection	Software engineering	Hybrid REP-plus-LLM ranking method (Zhang et al., 2023)
Robot data curation with influence functions	Robot learning	Demonstration ranking by policy-return influence (Agia et al., 23 Jun 2025)

This distribution of meanings suggests that “CUPID” functions less as a canonical term than as a reusable label for technically heterogeneous artifacts. The two historically deepest and most extensively developed usages in the provided literature are the Uranian moon and the CUORE successor experiment.

2. Cupid as a Uranian inner moon

Cupid is one of Uranus’s faint inner satellites. It was discovered in Hubble Space Telescope images rather than by Voyager 2, a consequence of its small size and low brightness. In updated dynamical work on Uranus’s 13 inner regular satellites, Cupid is treated as the innermost member of the Belinda group, orbiting inside Belinda and well inside Perdita. At epoch 2008-09-01 its quoted geometric orbital elements are $a = 74\,392.338\ \mathrm{km}$, $e = 0.00047$, and $i = 0.07028^\circ$; its mean radius is $R \approx 8.9 \pm 0.7\ \mathrm{km}$ (Ćuk et al., 2022).

Earlier numerical modeling emphasized Cupid’s extreme fragility. In the fiducial-density framework used by French and Showalter, it is the least massive known inner Uranian satellite in the model set, with $R_{\rm Cupid} \approx 9\ \mathrm{km}$ and $$GM_{\rm Cupid} = 2.038\times 10^{-4}\ \mathrm{km^3\,s^{-2}}$$. Its semi-major axis lies amid a dense set of low-order inner Lindblad resonances with Belinda, especially near the $58{:}57$ resonance, making it a natural locus for chaotic transport in the Uranian inner system (French et al., 2014).

The orbital architecture is nearly circular and nearly coplanar, but compact enough that small changes in fitted elements materially alter long-term forecasts. This is one reason the literature on Cupid’s fate split into an earlier “doomed” picture and a later resonance-stabilized one.

3. Dynamical stability and revised forecasts

The 2014 instability analysis concluded that the inner Uranian satellite system is generically unstable across a wide range of mass assumptions, with Cupid–Belinda usually the first pair to cross orbits. Depending on density assumptions, the quoted crossing times for Cupid–Belinda span $10^3\text{–}10^7$ years; in the unit-density 13-satellite models the first crossing is typically on the order of $10^4\text{–}5\times 10^5$ years. The proposed mechanism is resonance hopping: Cupid spends long intervals near specific Belinda inner Lindblad resonances, is kicked out, and diffuses chaotically in semi-major axis and eccentricity until $a = 74\,392.338\ \mathrm{km}$0 (French et al., 2014).

A 2022 reanalysis substantially altered that picture by using updated orbital fits from French (2017), explicit secular precession, and tidal considerations. In that treatment, the decisive structural element is the Belinda–Perdita $a = 74\,392.338\ \mathrm{km}$1 mean-motion resonance. Long-term integrations of the three-body Belinda group with nominal masses produced no collisions in 50 Myr, and the resonant argument librated for the full run. Seven out of nine simulations in a $a = 74\,392.338\ \mathrm{km}$2 grid of Belinda and Perdita masses remained stable for 50 Myr, whereas a configuration with all three moons at double the nominal mass became unstable in less than 3 Myr. The revised conclusion is that the Belinda group can plausibly remain stable on $a = 74\,392.338\ \mathrm{km}$3-year timescales provided the moons are relatively low-density bodies and the Belinda–Perdita resonance remains deep (Ćuk et al., 2022).

The same study also links stability, tides, and system age. Because Belinda, Cupid, and Perdita all lie inside Uranus’s synchronous orbit, standard Uranian tides drive inward migration and therefore divergent evolution for Belinda and Perdita; this cannot establish the observed $a = 74\,392.338\ \mathrm{km}$4 resonance. The proposed alternative is outward migration of Belinda through ring–moon torques during a past ring–moon cycle. In that framework, Cupid is probably not primordial: its maximum dynamical age is inferred to be of order $a = 74\,392.338\ \mathrm{km}$5, and the current resonant configuration may date from a disruption and re-accretion episode “possibly hundreds of Myr ago” (Ćuk et al., 2022).

A common misconception in the older literature is therefore that Cupid is necessarily lost on $a = 74\,392.338\ \mathrm{km}$6 year timescales. The later result does not claim indefinite stability, but it does move the dominant forecast from imminent collision to resonance-mediated survival over at least tens of Myr and plausibly $a = 74\,392.338\ \mathrm{km}$7 years.

4. CUPID as the CUORE successor experiment

In particle physics, CUPID denotes the “CUORE Upgrade with Particle IDentification,” a next-generation, tonne-scale bolometric neutrinoless double beta decay experiment designed to reuse the CUORE cryogenic infrastructure at LNGS. Its central scientific objective is to test the Majorana nature of neutrinos and discover lepton-number violation through $a = 74\,392.338\ \mathrm{km}$8, with sensitivity that fully covers the inverted-ordering region of the neutrino-mass parameter space (Group, 2015).

The original design study framed CUPID as a CUORE-scale apparatus with enriched isotopes, upgraded purification and crystallization, and active particle identification. Candidate absorbers included TeO$a = 74\,392.338\ \mathrm{km}$9, ZnMoO$e = 0.00047$0, ZnSe, and CdWO$e = 0.00047$1, with total fiducial masses of 750, 540, 670, and 980 kg respectively; the corresponding isotope fiducial masses were 543 kg for $e = 0.00047$2Te, 212 kg for $e = 0.00047$3Mo, 335 kg for $e = 0.00047$4Se, and 283 kg for $e = 0.00047$5Cd. The stated performance targets were $e = 0.00047$6 keV at the endpoint, event-selection efficiency of $e = 0.00047$7, and background within one FWHM at the endpoint below $e = 0.00047$8 (Group, 2015).

Subsequent design consolidation selected $e = 0.00047$9Mo in enriched $i = 0.07028^\circ$0 as the baseline technology. The 2025 design reports specify 1596 crystals in 57 towers, a total crystal mass of 450 kg, isotopic enrichment $i = 0.07028^\circ$1, and a $i = 0.07028^\circ$2Mo isotope mass of approximately 240 kg. Under the baseline assumptions of 5 keV FWHM energy resolution at $i = 0.07028^\circ$3, a background index of $i = 0.07028^\circ$4, and 10 live-years, CUPID reaches a $i = 0.07028^\circ$5 C.L. half-life exclusion sensitivity of $i = 0.07028^\circ$6 yr and a $i = 0.07028^\circ$7 discovery sensitivity of $i = 0.07028^\circ$8 yr. These map to $i = 0.07028^\circ$9 sensitivities of $R \approx 8.9 \pm 0.7\ \mathrm{km}$0 meV for exclusion and $R \approx 8.9 \pm 0.7\ \mathrm{km}$1 meV for discovery, thereby covering the full inverted-ordering region and a portion of normal ordering with lightest neutrino mass larger than 10 meV (Collaboration et al., 1 Mar 2025).

The experimental strategy is defined by the same tension that shaped CUORE: preserve bolometric energy resolution while aggressively suppressing backgrounds. CUPID’s distinctive answer is simultaneous heat and light readout, high-$R \approx 8.9 \pm 0.7\ \mathrm{km}$2 isotope choice, and the reuse of a proven cryogenic system rather than construction of a new large underground platform.

5. Detector architecture, light readout, and prototyping

CUPID’s baseline module is a scintillating bolometer based on cubic $R \approx 8.9 \pm 0.7\ \mathrm{km}$3 crystals coupled to cryogenic light detectors. A first full test of the baseline module reported an energy resolution of $R \approx 8.9 \pm 0.7\ \mathrm{km}$4 keV FWHM at the $R \approx 8.9 \pm 0.7\ \mathrm{km}$5Mo $R \approx 8.9 \pm 0.7\ \mathrm{km}$6-value of about 3034 keV and validated particle identification with $R \approx 8.9 \pm 0.7\ \mathrm{km}$7-particle rejection higher than $R \approx 8.9 \pm 0.7\ \mathrm{km}$8. In the adopted baseline geometry, the bottom light detector is 0.5 mm from the lower crystal face and the top light detector is 4 mm from the upper face; an alternative “gravity-assisted” configuration improved total light yield only modestly, from about $R \approx 8.9 \pm 0.7\ \mathrm{km}$9 to $R_{\rm Cupid} \approx 9\ \mathrm{km}$0 keV/MeV for $R_{\rm Cupid} \approx 9\ \mathrm{km}$1, and was therefore not selected as the baseline (collaboration et al., 2022).

A major R&D theme is light-detector performance in the mechanically noisy environment of a pulse-tube cryostat. A first test of CUPID prototypal Ge-wafer light detectors with NTD-Ge sensors achieved intrinsic baseline resolutions of 70–90 eV RMS for four devices, all within the CUPID goal of 100 eV RMS. One detector reached $R_{\rm Cupid} \approx 9\ \mathrm{km}$2 keV FWHM at the 356 keV line from a $R_{\rm Cupid} \approx 9\ \mathrm{km}$3Ba source, while pulse rise times were about 1.7–2.0 ms. The same paper emphasizes that low-frequency vibration noise is visible in the power spectra, but does not prevent compliance with CUPID’s baseline-noise specification (collaboration et al., 2023).

Alternative light-sensor technologies are being developed in parallel. Large-area photon calorimeters based on Ir–Pt bilayer transition-edge sensors met the CUPID baseline noise requirement of $R_{\rm Cupid} \approx 9\ \mathrm{km}$4 eV RMS, exhibited risetimes of $R_{\rm Cupid} \approx 9\ \mathrm{km}$5s, and achieved measured timing jitter of $R_{\rm Cupid} \approx 9\ \mathrm{km}$6s for the expected signal-to-noise at the decay $R_{\rm Cupid} \approx 9\ \mathrm{km}$7-value, satisfying CUPID’s criterion for rejecting $R_{\rm Cupid} \approx 9\ \mathrm{km}$8 pileup events (Singh et al., 2022). A complementary prototype using a cubic $R_{\rm Cupid} \approx 9\ \mathrm{km}$9-side $$GM_{\rm Cupid} = 2.038\times 10^{-4}\ \mathrm{km^3\,s^{-2}}$$0 crystal in the CROSS facility achieved 6 keV FWHM at the 2615 keV $$GM_{\rm Cupid} = 2.038\times 10^{-4}\ \mathrm{km^3\,s^{-2}}$$1 line, full $$GM_{\rm Cupid} = 2.038\times 10^{-4}\ \mathrm{km^3\,s^{-2}}$$2 separation between $$GM_{\rm Cupid} = 2.038\times 10^{-4}\ \mathrm{km^3\,s^{-2}}$$3 and $$GM_{\rm Cupid} = 2.038\times 10^{-4}\ \mathrm{km^3\,s^{-2}}$$4 events above 2 MeV, and internal radiopurity limits of less than 3 and 8 $$GM_{\rm Cupid} = 2.038\times 10^{-4}\ \mathrm{km^3\,s^{-2}}$$5Bq/kg for $$GM_{\rm Cupid} = 2.038\times 10^{-4}\ \mathrm{km^3\,s^{-2}}$$6Th and $$GM_{\rm Cupid} = 2.038\times 10^{-4}\ \mathrm{km^3\,s^{-2}}$$7Ra, respectively (Group et al., 2020).

These developments converge on a single detector logic: the heat channel preserves CUORE-class calorimetry, while the light channel provides the event-by-event discrimination needed to suppress the $$GM_{\rm Cupid} = 2.038\times 10^{-4}\ \mathrm{km^3\,s^{-2}}$$8-dominated background that limited CUORE. A plausible implication is that CUPID’s decisive technical hurdle is no longer whether dual-readout bolometry works in principle, but whether it can be industrialized across a full-scale, low-background, mechanically stable array.

6. CUPID in computing, AI, and data systems

Outside astronomy and rare-event physics, CUPID has become a productive acronym for computational systems that optimize context-sensitive decisions. In online MOBA games, CUPID is a two-stage re-matchmaking framework that first forms a 10-player lobby with a standard MMR system and then reassigns teams and positions by combining a position-satisfaction filter with a pre-match win-rate prediction model. On two large-scale, real-world datasets it surpassed all existing baselines, with an average relative improvement of $$GM_{\rm Cupid} = 2.038\times 10^{-4}\ \mathrm{km^3\,s^{-2}}$$9 in win prediction accuracy, and its deployment in a popular mobile MOBA improved match fairness and player satisfaction in A/B testing (Fan et al., 2024).

In LLM evaluation, CUPID is a benchmark of 756 human-curated interaction session histories designed to test whether models can infer contextual rather than global user preferences from prior dialogues. Across 10 open and proprietary LLMs, the benchmark reports that state-of-the-art models struggle with this capability, remaining under $58{:}57$0 precision and $58{:}57$1 recall in preference inference from multi-turn histories (Kim et al., 3 Aug 2025). In visual analytics, CUPID denotes “Contextual Understanding of Prompt-conditioned Image Distributions,” a system that analyzes distributions of text-to-image outputs by constructing density-based embeddings of contextualized object representations and extending them to conditional density embeddings for object-dependency analysis (Zhao et al., 2024).

In software engineering, CUPID combines the REP duplicate-bug-report detector with ChatGPT-based keyword extraction. On Spark, Hadoop, and Kibana, it achieved Recall Rate@10 values ranging from 0.602 to 0.654, improving over the previous state of the art by $58{:}57$2 and surpassing the strongest deep-learning baseline by up to $58{:}57$3 (Zhang et al., 2023). In robot imitation learning, CUPID is an influence-function-based data-curation method that ranks demonstrations by their estimated effect on a policy’s expected return; training with less than $58{:}57$4 of curated data can yield state-of-the-art diffusion policies on the simulated RoboMimic benchmark, with analogous gains in hardware (Agia et al., 23 Jun 2025).

Across these usages, the recurring conceptual pattern is contextual selection under uncertainty: resonance selection in the Uranian inner system, event selection in cryogenic rare-event searches, team and role assignment in online games, history selection for contextual LLM alignment, object-conditioning in image-distribution analysis, report selection in software repositories, and trajectory selection in robot learning. This suggests that the acronym’s repeated reuse is not entirely arbitrary; many CUPID systems are explicitly about identifying the small subset of structure that determines downstream behavior.