arXiv Science⌕ Search
Home/Archive/arXiv

arXiv · 2610.32135

FacilityAudit-Agent: Beyond Mention-as-Usage for Fine-Grained Facility Contribution Attribution in Astrophysics Publications

Wujun Shao·Dongwei Fan·Daoqi Peng·Yunfei Xu·Chenzhou Cui·Xin Lyu·Wei Huang
View original source ↗Read source document ↗Export metadata ↓

Abstract

Observatory bibliographies increasingly need to determine not only whether a facility is named in a paper, but how its data support the scientific argument. We frame this need as semantic contribution attribution on a six-level scale (L0-L5), from non-use and mention-only cases to auxiliary, major, and exclusive data support. We present FacilityAudit-Agent, a framework that builds a page-bounded outline of each PDF, enriches sections with roles and facility mentions, and uses a tool-calling large language model (LLM) to fetch text on demand. Evidence from the sections is combined through an explicit decision tree and a coverage-calibrated confidence score, with uncertain cases routed to human review. The framework supports both single-facility and inventory audits and transfers to new facilities without retraining. On a test set of 39,358 human-annotated paper-facility pairs, FacilityAudit-Agent achieves 89.1% accuracy and 79.8% macro-F1 in single_audit mode, outperforming the other three baselines by 5.8, 8.3, and 12.1 accuracy points, respectively, with the clearest gains at L2 and L5. In inventory_audit mode, it also leads in facility listing and in level assignment among listed pairs. Ablations show multi-hop reading and tree indexing matter most for L3-L5. FacilityAudit-Agent has been deployed at the National Astronomical Data Center, where it is responsible for processing publications related to LAMOST, FAST, and EP.

Explore related subjects

astro-ph.IM

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Wujun Shao, Dongwei Fan, Daoqi Peng, Yunfei Xu, Chenzhou Cui, Xin Lyu, Wei Huang. 2026-10-23. FacilityAudit-Agent: Beyond Mention-as-Usage for Fine-Grained Facility Contribution Attribution in Astrophysics Publications. https://arxiv.org/abs/2610.32135

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

OmniSpectra: A Unified Foundation Model for Universal Spectra Representation Learning

Standard sequence models encode position by token rank rather than physical coordinate, typically forcing practitioners to destructively resample continuous physical signals such as spectra observed across heterogeneous instruments. We identify this coordinate mismatch as a structural limitation in cross-instrument representation learning, and propose OmniSpectra, a wavelength-native foundation model for spectral representation learning: a continuous-coordinate masked autoencoder that discards the integer pixel indexing of sequence modeling. Mapping physical coordinates to a logarithmic basis turns the multiplicative scaling induced by cosmological redshift into a uniform additive shift. In continuous rotary position embeddings, this shared shift cancels from relative token distances rather than producing grid-dependent positional distortion. We further counter the spectral bias inherent to implicit neural decoders with a physics-grounded, resolution-gated Fourier decoder that provides anti-aliasing. Pretrained on eight surveys and benchmarked against seven baselines, our model reaches state-of-the-art performance with only 12.2M parameters while outperforming substantially larger models, indicating that encoding physical structure can substitute for parameter scale.

astro-ph.IM↗
arXiv2026

What visibility-graph curvature measures in Forbush decreases: a calibration on the two largest events of solar cycle 25

Discrete curvatures of visibility graphs are increasingly used as descriptors of geophysical time series, but their response to a transient has rarely been calibrated against the simplest properties of that transient. We carry out this calibration for the nodal Forman--Ricci curvature $\mathcal{F}$ of natural visibility graphs applied to neutron-monitor records of the Gannon storm (2024 May 10--11) and the 2025 June 1 Forbush decrease (FD), the two largest FDs of solar cycle 25, using 30-min NMDB data from six NM64 stations ($R_c \simeq 0$--$7$~GV), i.e.\ twelve station--event series. Curvature collapses at FD onset in every series (Cliff's $δ= 0.63$--$1.0$), but a 6-h running standard deviation separates onset from quiet time more strongly (mean $δ= 0.96$ vs $0.89$; mean block-permutation displacement $z = -27$ vs $-14$). Injecting synthetic decreases into the real pre-event records of each station shows that the curvature extreme follows $|\mathcal{F}_{\min}| \propto (A/σ)^{0.50}$ ($R^2 = 0.88$), where $A$ is the FD amplitude and $σ$ the station's quiet-time noise, and is independent of the duration of the main phase (exponent $0.006$ over 3--36~h). With $A$, $σ$ and the main-phase duration of each series as the only inputs, the calibration predicts the twelve observed extremes with standardized residuals of mean $-0.26$ and standard deviation $0.94$, and the observed extremes correlate with $A/σ$ at Spearman $ρ= 0.95$. Within the precision allowed by these data, nodal Forman--Ricci curvature of the whole-series visibility graph carries no information about these FDs beyond their amplitude relative to station noise; in particular, it does not distinguish the single-step Gannon decrease from the two-step June 2025 decrease.

astro-ph.IM↗
arXiv2026

The MIRI/MRS Library. III. Detector-based optimal spectral extraction and high-contrast imaging applications

(Abridged) JWST has now been operational for five years, and the associated data reduction pipelines have reached maturity. For MIRI/MRS, one can generally get a good-quality spectrum by simply running the standard pipeline, especially for unresolved sources. However, users are starting to look deeper into the data, trying to detect the smallest features in the faintest objects, and find close companions. Following the corrections derived in Papers I and II in this series, we now introduce a novel way to extract MIRI/MRS spectra: from the detectors directly. Currently the most common way to analyse data from integral field spectrometers is to reconstruct a spectral cube from the detector images. Theoretically, detector-extracted spectra could reach higher S/N, and better handle bad pixels. We generate stable and well-sampled PSF models using the sub-pixel scan of calibration star 10 Lac observed in PID 3779. These PSFs were used in a detector-based optimal extraction algorithm building on past approaches, where the contributions per pixel were weighted by their noise properties and contribution to the total illumination. The PSF models were used to subtract the PSF of a bright central source from the detector, both by direct subtraction, and a PCA approach. We find that our optimal setup allows us to get a $>$1.5 times higher S/N compared to the standard pipeline cube extraction. However, we approximately matched more involved cube processing. We were able to subtract the PSF to sub-percent levels, allowing us to extract spectra from faint companions. The novel approach greatly improves on the current standard pipeline reduction. We have not yet reached the fundamental noise limits. There are clear points of improvement that can be addressed using additional data. We demonstrate that detector-based processing of MIRI/MRS data has great potential.

astro-ph.IM↗