arXiv:2602.05235v2 Announce Type: replace
Abstract: Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by grounding generation in external knowledge to improve factuality and reduce hallucinations. Yet most deployments assume a centralized corpus, which is infeasible in privacy aware domains where knowledge remains siloed. This motivates federated RAG (FedRAG), where a central LLM server collaborates with distributed silos without sharing raw documents. In context RAG violates this requirement by transmitting verbatim documents, whereas parametric RAG encodes documents into lightweight adapters that merge with a frozen LLM at inference, avoiding raw-text exchange. We adopt the parametric approach but face two unique challenges induced by FedRAG: high storage and communication from per-document adapters, and destructive aggregation caused by indiscriminately merging multiple adapters. We present FedMosaic, the first federated RAG framework built on parametric adapters. FedMosaic clusters semantically related documents into multi-document adapters with document-specific masks to reduce overhead while preserving specificity, and performs selective adapter aggregation to combine only relevance-aligned, nonconflicting adapters. Experiments show that FedMosaic achieves an average 10.9% higher accuracy than state-of-the-art methods in four categories, while lowering storage costs by 78.8% to 86.3% and communication costs by 91.4%, and never sharing raw documents.
Science Journals
arXiv:2607.11015v1 Announce Type: cross
Abstract: Holographic multiple-input multiple-output (HMIMO) has emerged as a promising technology for future wireless systems by enabling continuous electromagnetic (EM) field control over large apertures. However, user-side EM exposure has become an increasingly important concern in large-scale array systems. This paper addresses this issue by developing a multiuser uplink HMIMO model, where a physically consistent specific absorption rate (SAR) model is established to quantify the EM exposure. On this basis, a spectral efficiency (SE) maximization problem is addressed by developing a modified iterative water-filling algorithm. Simulation results demonstrate that the proposed algorithm effectively improves the system SE while satisfying the SAR constraints.
arXiv:2607.11407v1 Announce Type: new
Abstract: Precise alignment of the undulator trajectory onto a straight line is a per-requisite for high fidelity SASE operation, which in turn also enables non-standard operation modes like self-seeding or two-color operation. At European XFEL electron and photon beam-based method are combined with a step-by-step performance-based optimization to ensure that the trajectory is on a straight line over most of the length of the about 250 m long undulator system.
arXiv:2607.11385v1 Announce Type: cross
Abstract: Diffusion-weighted imaging (DWI) is acquired as part of bi-parametric prostate MRI, but suffers from artifacts that degrade downstream quantitative and diagnostic performance. While DWI preprocessing is standard in brain imaging, its adoption in prostate imaging remains limited and lacks standardized pipelines. This study investigated the effect of different DWI preprocessing strategies on apparent diffusion coefficient (ADC) estimation and automatic Prostate Imaging Reporting and Data System (PI-RADS) classification. 268 cases were derived from the fastMRI prostate cohort by sequentially applying denoising, Gibbs-ringing correction, and diffeomorphic registration for susceptibility distortion correction. ADC maps were compared using linear least squares (LLS) and iteratively-weighted LLS (IWLLS). A 3-class DenseNet classifier was trained to predict PI-RADS scores from multi-channel MRI inputs. ADC analysis revealed statistically significant differences across preprocessing pipelines, with LLS and IWLLS producing numerically equivalent maps. Linear relationships between ADC values were preserved across most datasets (PCC ~0.99), while distortion correction realigned DWI to T2w anatomy and altered ADC values accordingly (PCC ~0.90). Classification showed the best AUROC and sensitivity for high-risk PI-RADS classes in the fully processed dataset. False-negative analysis revealed this dataset produced the least overconfident incorrect predictions on high-risk classes, which is a desirable property for clinical triage. DWI preprocessing, particularly distortion correction, enhances both ADC map quality and the predictive power of deep learning models for PI-RADS classification, supporting the need for optimized preprocessing pipelines in prostate MRI.
arXiv:2606.29116v2 Announce Type: replace
Abstract: Large Language Models (LLMs) are rapidly being adopted in low-code and no-code automation platforms, where non-expert users design workflows that combine natural language understanding with external services and APIs. LLM agents are LLM systems that use LLMs as a core "brain" to reason, plan, and autonomously execute complex, multi-step tasks. In this paper, we present the first large-scale empirical study of LLM agentic workflows in low-code automation platforms. We analyze more than 6,000 publicly available n8n workflows and examine four aspects of their design: task distribution, structural and tool use patterns, reliability mechanisms, and autonomy levels. Our analysis shows that LLM workflows are not merely prompt response pipelines. Instead, LLMs are commonly embedded within broader automation structures involving control logic, external tools, communication services, storage systems, and human review points. We further find that while many workflows include lightweight post-processing or routing logic after LLM execution, explicit reliability mechanisms such as structured fallback paths, repair loops, failure-specific alerts, and human approval gates remain relatively uncommon. These results reveal a gap between the increasing deployment of LLM agents in practical automation ecosystems and the limited engineering support for reliability, safety, and governance. Overall, our study provides ten empirical findings and five research takeaways for researchers, platform developers, and practitioners seeking to understand and improve real-world LLM agentic workflows.
arXiv:2607.11112v1 Announce Type: new
Abstract: Dynamic graph continual learning (DGCL) is an effective manner for handling catastrophic forgetting in dynamic graphs. However, existing DGCL methods underutilize temporal information across graph snapshots. To address this critical issue, we propose a novel framework for Dynamic Graph Continual Learning via Condensation and Attachment (CA-DGCL). Specifically, CA-DGCL first condenses historical graph snapshots into compact semantic representations efficiently. Further, a cross-timestamp node chains is built to construct a third-order tensor and Tucker decomposition is applied to this tensor for obtaining stable node features, which encapsulate historical knowledge. Finally, these node features are used to generate new nodes and attached to the current graph for replaying of past information without compromising the new patterns. In addtion, a refined forgetting measure is introduced to make it more suitable for dynamic graph settings. Extensive experiments demonstrate that CA-DGCL outperforms baselines in forgetting suppression as well as maintain competitive accuracy, proving its efficacy for dynamic graph continual learning.
arXiv:2607.11418v1 Announce Type: new
Abstract: It is shown that an efficient, direct, and fully quantum mechanical calculation of thermal reaction rate coefficients requires a new, partially rearranged form of the numerically-constructed exact kinetic energy part of the rovibrational Hamiltonian expressed in internal coordinates. Using this Hamiltonian and an accurate, full-dimensional potential energy surface characterizing the H$_2$ + H exchange reaction, developed by Mielke, Garrett, and Peterson (J. Chem. Phys. 2002, 116, 4142), reaction rate coefficients in the temperature range of $75-800$ K have been computed for the H$_2$ + D $\rightarrow$ HD + H reaction. The paper puts particular emphasis on the exact treatment of overall molecular rotation and on nuclear spin symmetry.
arXiv:2607.11548v1 Announce Type: new
Abstract: Spatial football metrics such as pitch control assume access to the positions of all 22 players, yet the most widely available source of positional data -- the broadcast main camera -- shows only 10-16 of them at any moment. We quantify the resulting distortion with an open, reproducible benchmark: a simulated broadcast viewport applied to open full-pitch tracking data (Metrica Sports; three matches, one held out from method development). Ignoring off-screen players -- the visible-only baseline implied whenever a video-based game-state-reconstruction (GSR) pipeline adds no imputation layer -- inflates hidden-zone pitch-control error to 25.1-26.9 percentage points and a mean absolute control-share error of 11.1-13.4 points across the three matches. We then evaluate a ladder of training-free, online imputation baselines that use only observations from the match being analysed. The best overall on these decision-relevant metrics, role-anchored centroid voting (each visible player votes for the full-team centroid by subtracting its running role offset, attenuating the viewport-induced subset bias), roughly halves hidden-zone error (to 12.2-13.8 points) and cuts control-share error to 28-48% of the ignore policy at every viewport width from 36 m to 60 m in all three matches. For occlusions <=9.6 s -- the regime of the closest learned prior work -- it reaches binwise median position errors of 3.3-8.9 m; but 50-57% of hidden-player observations lie beyond that regime. Integrated end-to-end into a broadcast-video GSR pipeline, imputation moves a downstream possession-quality score (Space-Creation Index) by 15.6 and 17.2 points on two real World Cup broadcast windows, flipping the verdict class in one.
arXiv:2607.11404v1 Announce Type: cross
Abstract: Links between physicals systems and computability properties have been an active field of investigation in recent years. Inspired by a previous work in the context of positive temperature Gibbs measures, we prove here that in the context of zero-temperature ergodic optimisation, for a computable potential and provided with several reasonable assumptions, the maximum ergodic average is a computable real number, and the set of maximising measures is a $\Pi_1$-computable compact set.
Then, in the more specific context of symbolic dynamics, with finite-range interactions on subshifts of finite type, we provide an explicit algorithm to compute both the maximum ergodic average and the set of maximising measures in finite time, with a matching code repository.
arXiv:2607.11218v1 Announce Type: new
Abstract: Scholarly infrastructures increasingly expose their data through REST APIs that follow shared specifications, such as the Scientific Knowledge Graph - Interoperability Framework (SKG-IF), which defines a common data model, exchange format, and REST API for research information. Implementing such specifications over existing data sources, however, requires a development effort that many open infrastructures cannot afford. RAMOSE, the RESTful API Manager Over SPARQL Endpoints, is an open-source Python framework that reduces this effort by turning a declarative configuration file into a documented REST API over RDF triplestores. This article presents its second major version, which extends the tool with nine new requirements. The new features include query orchestration across multiple SPARQL endpoints and non-RDF sources, with joins across their results; pluggable output formats and request parameters; pagination and caching; OpenAPI export; and write operations, with authentication both of API consumers and towards protected endpoints. A built-in module packages the format and filters that SKG-IF prescribes, letting a provider expose a compliant endpoint only through configuration. A functional comparison with nine similar tools, grounded in reproducible tests, shows that only RAMOSE joins RDF and non-RDF results on arbitrary keys within one API operation. RAMOSE serves the OpenCitations REST APIs, which peaked at almost 38 million monthly requests between May 2025 and May 2026, and has been adopted by the GRAPHIA project to onboard data sources into its SKG-IF-based federation.
arXiv:2607.11701v1 Announce Type: cross
Abstract: Quantitative Structure-Activity Relationship ($\mathtt{QSAR}$) modeling is a foundational computational methodology in early-stage drug discovery, heavily relied upon for predicting compound toxicity, bioavailability, and therapeutic potential. However, classical methods often struggle to effectively map the highly complex, non-linear, and high-dimensional interactions inherent in molecular data, leading to reduced predictive accuracy and costly late-stage clinical failures. In this paper, we present a Quantum Multiple Kernel Learning ($\mathtt{QMKL}$) framework, dubbed Next-Gen $\mathtt{Q^2SAR}$, that leverages Quantum Support Vector Machines ($\mathtt{QSVMs}$) to overcome these classical limitations. By encoding molecular descriptors into exponentially large quantum Hilbert spaces, our approach substantially enhances the expressiveness of non-linear modeling. Benchmarking our quantum-enhanced framework on a dataset targeting the $\mathtt{DYRK1A}$ kinase (a critical target for Alzheimer's disease), the $\mathtt{QMKL}$-$\mathtt{SVM}$ achieves an impressive Area Under the Curve ($\mathtt{AUC}$) score of $0.8750$, significantly outperforming classical state-of-the-art Gradient Boosting models ($\mathtt{AUC} = 0.8037$). Furthermore, we establish a theoretical and empirical pathway toward resolving classical data bottlenecks through projected quantum kernels ($\mathtt{PQK}$) and measurement accelerators. As quantum computing architecture matures, this framework paves the way for autonomous cognitive architectures and self-improving drug discovery pipelines, promising to unlock deeper insights across vast chemical spaces and to accelerate the development of life-saving therapeutics.
arXiv:2607.10907v1 Announce Type: cross
Abstract: In this work, Ge-doped $\beta$-Ga$_2$O$_3$ homoepitaxial films were grown on native (010) $\beta$-Ga$_2$O$_3$ substrates using low-pressure chemical vapor deposition (LPCVD). Controlled $n$-type doping was achieved with room-temperature carrier concentrations ranging from $7.4\times10^{17}$ to $2.57\times10^{18}\ \mathrm{cm}^{-3}$ and corresponding electron mobilities of 105-62 cm$^2$/V$\cdot$s. The films exhibited smooth surface morphology with RMS roughness values of 2.94-3.97 nm, while X-ray diffraction, Raman spectroscopy, and X-ray photoelectron spectroscopy confirmed phase-pure $\beta$-Ga$_2$O$_3$ with excellent crystalline quality and near-stoichiometric composition. Temperature-dependent Hall measurements on the film with a room-temperature carrier concentration of $7.4\times10^{17}\ \mathrm{cm}^{-3}$ and mobility of 105 cm$^2$/V$\cdot$s yielded a peak electron mobility of 234 cm$^2$/V$\cdot$s at 116 K, while charge-neutrality and transport modeling revealed a dominant shallow donor level with an activation energy of 14 meV, confirming efficient electrical activation of Ge donors. Vertical Ni/$\beta$-Ga$_2$O$_3$ Schottky barrier diodes fabricated using the Ge-doped drift layer exhibited good rectifying behavior with a turn-on voltage of 0.74 V, an ideality factor of 1.32, a Schottky barrier height of 1.02 eV, and a specific on-resistance of 2.49 m$\Omega\cdot$cm$^2$. Capacitance-voltage measurements yielded a net donor concentration of $7.7\times10^{17}\ \mathrm{cm}^{-3}$ and a Schottky barrier height of 1.13 eV, in good agreement with Hall and current-voltage measurements. These results demonstrate that LPCVD enables controllable Ge doping while maintaining high structural and electronic quality, establishing LPCVD-grown Ge-doped $\beta$-Ga$_2$O$_3$ as a promising platform for future high-voltage power electronic devices.
arXiv:2603.16728v2 Announce Type: replace
Abstract: Vision-language models (VLMs) are increasingly deployed in high-stakes settings where reliable uncertainty quantification (UQ) is as important as predictive accuracy. Extended reasoning via chain-of-thought (CoT) prompting or reasoning-trained models has become ubiquitous in modern VLM pipelines, yet its effect on UQ reliability remains poorly understood. Our results show that reasoning tends to degrade the quality of many uncertainty estimates, even when it improves task accuracy. We identify implicit answer conditioning as the primary mechanism: as reasoning traces converge on a conclusion before the final answer is generated, token probabilities increasingly reflect consistency with the model's own reasoning trace rather than uncertainty about correctness. In effect, the model becomes overconfident in its answer. In contrast, agreement-based consistency remains robust and often improves under reasoning, making it a practical choice for uncertainty estimation in reasoning-enabled VLMs.
arXiv:2607.07135v2 Announce Type: replace
Abstract: Contrastive Language-Image Pre-training (CLIP) relies on softmax-based self-attention, a strictly positive distribution that assigns probability mass to every pair of tokens-even semantically irrelevant ones. While these dense softmax weights are effective for gathering broad context during pre-training, they spread attention across many low-salience tokens, producing noise that obscures the fine-grained, spatially localized cues required for dense, open-vocabulary prediction. We study an inference-time substitution of the row-wise softmax in the final visual self-attention layers with the $\alpha$-entmax transform, applied across both the standard query-key attention and self-correlation variants. Because entmax applies a data-dependent threshold that maps low scores exactly to zero, it acts as an implicit denoiser, zeroing contextually irrelevant dependencies while redistributing mass onto the most relevant tokens. We evaluate on open-vocabulary tasks-dense semantic segmentation (Pascal VOC, Pascal Context, ADE20K) and fine-grained retrieval (FG-OVD)-and find the gain from attention sparsification is proportional to how much the baseline attention spreads off the target class.
arXiv:2607.11632v1 Announce Type: new
Abstract: Human choice behavior, including route choice, exhibits systematic behavioral biases that deviate from the assumptions of full rationality. Cumulative prospect theory (CPT) has been widely recognized as an effective framework for characterizing such behavioral patterns. However, its large-scale application, particularly in simulation and agent-based modeling, critically depends on specifying individual-level CPT parameters, which remain a major bottleneck. Conventional approaches typically rely on surveys and controlled experiments to calibrate CPT parameters, yet these methods are difficult to generalize and often fail to capture the full diversity of human decision-making. To address this challenge, this paper investigates whether large language models (LLMs) can reproduce human behavioral biases in choice-making without explicit specification of prospect-theoretic parameters. Using route choice as a representative scenario, we design a behavioral evaluation framework and systematically compare LLM-generated decisions with established human behavioral patterns predicted by CPT. Experimental results demonstrate that LLMs are capable of reproducing non-rational human choice biases and can exhibit decision behaviors consistent with prospect-theoretic effects under uncertainty. These findings suggest that generative AI models may provide a scalable alternative for modeling human decision processes and offer a promising foundation for next-generation large-scale agent-based simulation and AI-driven behavioral research.
arXiv:2607.10022v1 Announce Type: cross
Abstract: We prove that, given any contact $3$-manifold and any computable function $f: \mathbb{N} \dashrightarrow \mathbb{N}$, there exists a defining contact form and a Poincar\'e section of its Reeb flow whose partially defined return map computes $f$.
arXiv:2503.11891v2 Announce Type: replace
Abstract: We analyze the landscape and training dynamics of diagonal linear networks in a linear regression task, with the network parameters being perturbed by isotropic normal noise during training. The addition of such noise may be interpreted as a stochastic form of sharpness-aware minimization (SAM) and we prove several results that relate its action on the underlying landscape and training dynamics to the sharpness of the loss. In particular, the noise induces a weighted mixture of fractional norm penalties on the network parameters, which forces the individual layers to balance at a fast rate and changes the underlying landscape to favor solutions that result from a shrinkage-thresholding operator applied to the true parameter. We show that balancing the layers equates to minimizing the average sharpness, as well as the trace of the Hessian matrix, among all possible factorizations of the same linear predictor. Further, we characterize how the noise level of the normal perturbations acts as a regularization parameter, with exact descriptions of its effect on shrinkage, thresholding, and balancing speed.
arXiv:2607.10990v1 Announce Type: new
Abstract: Automated understanding of complex soccer scenarios from video remains a significant challenge for contemporary vision-language models (VLMs), which suffer from shallow cross-modal alignment and exhibit fundamental limitations in multi-step reasoning and coordinated tool integration. We present TreeSoc, a structured reasoning framework that reformulates soccer video question answering as a hierarchical search problem rather than a single-pass prediction. Specifically, TreeSoc employs a dynamic depth-first search (DFS) mechanism that decomposes complex queries into sequentially ordered sub-tasks, enabling iterative reasoning refinement through explicit intermediate states. This tree-structured decomposition naturally supports adaptive tool routing, wherein domain-specific modules are selectively activated and their outputs incorporated at each reasoning node to produce contextually grounded predictions. On SoccerBench, TreeSoc achieves state-of-the-art performance, with accuracies of 85.2%, 87.4%, and 82.2% on TextQA, ImageQA, and VideoQA, respectively. Additionally, TreeSoc further demonstrates strong cross-domain generalization, attaining 74.16% accuracy on NExT-QA. These results establish structured, tool-augmented tree reasoning as an effective paradigm for robust video understanding. Code is available at: https://github.com/thanhnhan29/TreeSoc.
arXiv:2607.11295v1 Announce Type: new
Abstract: Magnetic resonance imaging exhibits substantial acquisition variability, where identical anatomy can appear markedly different across scanners and imaging protocols. Consequently, learned representations entangle biological structure with acquisition-dependent appearance, limiting interpretability, generalisation, and clinical deployment. We show that these sources of variation can be separated by jointly modelling MRI images and DICOM metadata. Using large-scale clinical brain MRI data, we learn representations that separate anatomical structure from contrast-dependent appearance. Resulting contrast representations organise heterogeneous acquisitions, support sequence understanding, and detect image--metadata inconsistencies, whereas anatomical representations suppress acquisition-specific variation while preserving biologically relevant information. Building on these disentangled representations, we introduce a unified anatomy-preserving harmonisation model for cross-modality and cross-site adaptation, conditioned on image or acquisition metadata. Our findings suggest that acquisition variability is a structured component of the imaging process that can be modelled, audited, and controlled, providing a foundation for acquisition-aware representation learning in large-scale medical imaging.
arXiv:2607.11210v1 Announce Type: new
Abstract: We study robust repeated contextual pricing, where valuations depends linearly on the features. At each round $t\in[T]$, a seller observes a context, posts a price, and receives only a possibly corrupted binary sale feedback. The seller knows an upper bound $C$ on the number of corrupted rounds. We design an algorithm with regret $\mathcal O(Cd+d^2\log T)$, where $d$ is the context dimension. This is the first guarantee for robust contextual pricing that separates the dependence on the corruption budget $C$ from the horizon $T$, closing the problem left open by Gupta, Guruganesh, Paes Leme, and Schneider (2025).
arXiv:2607.11333v1 Announce Type: new
Abstract: Inspired by recent advances in notions of spectral approximation of digraphs [Ahm+20], we study spectral algorithms for finding periodic structures in digraphs via the spectrum of a class of rotated Laplacian matrices. This class of Laplacian matrices was previously studied by Lange, Liu, Peyerimhoff, and Post [Lan+15]. We consider a notion of periodicity ratio that generalizes the bipartiteness ratio of Trevisan [Tre09], and show that it is closely related to the spectrum of rotated Laplacian matrices. In particular, if the digraph is strongly connected and represents a Markov chain, this periodicity ratio for a given $p \in \mathbb{N}$ is a quantitative measure of how close this Markov chain is to having periodicity $p$.
We propose and analyze a periodicity-ratio variant of the spectral algorithm by Louis, Raghavendra, Tetali and Vempala [Lou+12]. We show that the algorithm runs in randomized polynomial time and can find many nearly periodic components (i.e, components with small periodicity ratio). This also implies a new higher-order Cheeger-type inequality for periodicity in the spirit of that in [Lou+12; LOT14].
As part of our analysis, we prove a new theorem that upper bounds the probability that the largest magnitudes of two sequences of coordinate-wise correlated complex Gaussian random variables occur at different indices, which may be of independent interest. Previously, an analogous result was known only for real Gaussian random variables.
arXiv:2602.05867v2 Announce Type: replace
Abstract: Mysterious citations are routinely appearing in peer-reviewed publications throughout the scientific community. In this paper, we developed an automated pipeline and examine the proceedings of four major high-performance computing conferences, comparing the accuracy of citations between the 2021 and 2025 proceedings. While none of the 2021 papers contained mysterious citations, every 2025 proceeding did, impacting 2-6% of published papers. In addition, we observe a sharp rise in paper title and authorship errors, motivating the need for stronger citation-verification practice. No author within our dataset acknowledged using AI to generate citations even though all four conference policies required it, indicating current policies are insufficient.
arXiv:2406.14211v2 Announce Type: replace-cross
Abstract: Convergence guarantees for optimization over bounded-rank matrices are delicate to obtain because the feasible set is a nonsmooth and nonconvex algebraic variety. Existing techniques include direct optimization over bounded-rank matrices (e.g., projected gradient descent), fixed-rank optimization (over the maximal-rank stratum), and the LR parameterization. They all lack either global guarantees (the ability to accumulate only at stationary points) or fast local convergence (e.g., if the limit has non-maximal rank). We study a lifted geometry that allows algorithms to enjoy both.
Khrulkov and Oseledets [2018] parameterize the bounded-rank variety via a desingularization to recast the optimization problem onto a smooth manifold. Building on their ideas, we develop a Riemannian geometry for this desingularization, also with care for numerical considerations. We use it to ensure conditions that, for many standard algorithms, yield global convergence to stationary points with fast local rates. On matrix completion tasks, we find that this approach is comparable to others.
arXiv:2607.08857v2 Announce Type: replace
Abstract: Human egocentric video is a scalable supervision source for humanoid policy learning, but current pipelines struggle with hand-object occlusion, oversimplified motion, or specialized capture hardware. We introduce AgenticFocus, a Mixed Reality synthesis pipeline that converts ordinary first-person-view human videos into robot-trainable demonstrations by restoring occluded object geometry, reconstructing full-hand motion, and retargeting it to a humanoid embodiment through camera-relative alignment and layered compositing. The resulting dataset pairs focused visual observations with synchronized robot actions and states. AgenticFocus achieves lower trajectory error and smoother wrist motion than cross-embodiment baselines, with SPARC scores of -5.18 versus -5.56 and -6.05.
arXiv:2607.11231v1 Announce Type: new
Abstract: Cloud gaming streams cloud-rendered frames under strict motion-to-photon latency, yet its at-scale viability is increasingly constrained by bandwidth cost: in our study of the T cloud gaming platform, bandwidth accounts for 30-60% of total operating expense. This high bandwidth consumption stems from a fidelity-first objective of making the stream perceptually indistinguishable from local gameplay. It drives production systems toward best-effort bitrate allocation that pushes the encoder to the highest rate allowed by congestion control. However, the bitrate-perception relationship saturates: beyond a frame-dependent perceptually lossless threshold, additional bits yield negligible perceptual improvement, creating systematic redundant quality that wastes bandwidth.
We present SAIL, a production quality-aware rate control system with the goal of achieving perceptually lossless quality while avoiding unnecessary bandwidth waste. SAIL adopts a post-encoding architecture to enable millisecond-scale feedback at near-zero overhead. It comprises three key designs: (i) an encoder-driven quality assessment model that leverages zero-cost encoder outputs for real-time quality estimation; (ii) a hybrid rate control mechanism that balances steady-state adaptation with dynamic spike absorption; and (iii) a network-aware strategy that coordinates with congestion control to prevent capacity underestimation. SAIL has been fully deployed on the T cloud gaming platform and reduces bandwidth consumption by 44.27% and end-to-end latency by 8.37% without degrading perceived quality, serving tens of millions of users and accumulating billions of hours of total gameplay.