Papers

Character Error Rate (CER), the standard metric for OCR and vision-language models (VLMs), charges a typo as much as a wrong name. We introduce semCER, which keeps CER's alignment and denominator and charges each grapheme edit a severity from an LLM judge, with no word segmentation. Normalized CER keeps the score of a whole transcription, where it tracks readers as well as semCER or better, and semCER breaks that score down into named edits. The charges track readers' ratings edit by edit in English, Vietnamese, and Chinese and put the edits readers rate most severe near the top of a sorted list. On pairs drawn where CER and semCER disagreed, readers side with semCER more often than with normalized CER in English and Chinese, and equally often in Vietnamese. Across 16 languages, the judge charges a wrong diacritic or tone mark almost nothing in Spanish but nearly a wrong letter's severity in Telugu. Where we can count real words, a general VLM's wrong letter forms one far above chance while a classical pipeline's stays near it. We release the 61 million judged edits, the aggregated reader ratings, and the distilled scorer at https://github.com/nguyenx2t/semcer.

Safely executing long-horizon multi-arm tasks online in complex dynamic environments remains challenging. We propose a new distributed dynamic Gaussian belief propagation (DD-GBP) method that directly plans joint-space trajectories through parallel local inference. A unified Gaussian-process (GP) parameterization connects adaptive planning horizons, interpolation-time factors, and continuous execution. Each robot maintains a local tree-structured factor graph and exchanges messages asynchronously over active inter-arm links. Dynamic subnet reconstruction and event-triggered updates reduce communication and inference costs while responding to moving obstacles. We also provide theoretical analyses of local inference accuracy, anytime early stopping, triggered-replanning consistency, and GP-consistent execution. In 50 long-horizon, multi-target four-arm scenes with moving obstacles, DD-GBP achieves $92.5\%$ collision-free all-target completion and the highest success rate among six representative baselines. Scenario ablations demonstrate long-horizon safety and robustness to communication impairment and extreme conditions; success remains $86.0\%$ under $90\%$ packet loss. Execution results further show substantially suppressed extreme velocity and acceleration peaks relative to Syn-GBP. On one workstation, obstacle-free tests with 30 coexisting arms retain $99.87\%$ success and a $15.80$ ms mean warm solve, exceeding 60 Hz and outperforming the compared methods in scalability. Dual-arm transport experiments under dynamic interference demonstrate online response and practical deployability. To the best of our knowledge, these results establish a new state of the art for online long-horizon multi-arm planning in dynamic scenes.

Chemical chain-of-thought can contain hallucinated structural and property claims. We thereby introduce MolLean, a process-verifiable language that grounds chemical reasoning in an explicit molecular state. A layered Lean-based kernel checks state transitions and assertions over molecular readouts, from primitive graph edits to compound transformations. Meanwhile, a typed witness protocol separates native structural checks from externally supplied measurements. We further introduce MolLean-MCP, a stateful tool interface that exposes these transitions, checks, and readouts during generation: the model receives feedback to decide its next action, and rejected actions leave the state unchanged. MolLean supports one-pass trace execution, whole-trace refinement, and interactive reasoning through MolLean-MCP. We evaluate these three modes with direct chain-of-thought on 8,800 matched examples across 12 tasks spanning molecular generation, editing, optimization, and reactions. With Qwen3.6-27B, whole-trace refinement outperforms direct chain-of-thought on all 12 tasks. MolLean-MCP complements whole-trace refinement with stepwise verification feedback, reaching better performance in reaction tasks. Together, these results highlight an executable foundation for chemical reasoning, in which explicit chemical commitments can be checked and revised rather than remaining solely in free-form explanations.

We present a stochastic calculus approach for modeling generalized parton distribution functions using information provided from lattice QCD calculations of Mellin moments, without assuming a functional form. Reconstructing the underlying dependence on the partonic momentum fraction from a limited number of moments constitutes an intrinsically non-unique inverse problem. Our goal is to provide a model agnostic way to determine GPDs subject to integral constraints. Thus, we derive and show usage of a controlled Langevin approach based on stochastic calculus and Brownian motion. The method combines a variational formulation, based on minimizing a Dirichlet action subject to the imposed constraints, with stochastic dynamics that generates an ensemble of admissible functions. The resulting ensemble provides a preferred reconstruction and characterizes the variability associated with incomplete integral information within the chosen stochastic model. The framework can accommodate general integral kernels and prescribed function values, offering a flexible approach to reconstructing functions from indirect constraints. Beyond its application to GPD phenomenology, the method provides a general strategy for investigating inverse problems in which integral observables are known but the underlying functional dependence remains undetermined. The approach may have more general applications beyond the scope of studying 3D hadron structure, and it is particularly useful in situations where we know definite integrals of functions with kernels, but do not know the underlying function.

Co-scientists and agentic scientific claim verifiers are increasingly used to assess claims across the sciences. Natural-language scientific claims, written for human experts, are often underspecified and so admit several defensible interpretations. Such verifiers usually pick one, often tacitly, and benchmarks score their verdicts against expert labels without recording the expert's interpretation or the verifier's in a comparable form. This leads to disagreements with expert labels that cannot be traced to the evidence or the interpretation, and to compute spent on unchecked interpretations. We introduce Claim Cards to make such interpretations explicit in the reasoning itself. A Claim Card records a verifier's interpretation as nine fields before evidence gathering, each marked stated, inferred, or default. To test whether an agent's interpretation can change its verdict, we ran a GPT-5.4 agent five times on each of 60 expert-labeled materials-science and biomedical claims with a card, without one, and with one card frozen across runs (900 runs). Over 40% of card fields were not marked stated, and a card added about a tenth to a run's GPT-5.4 tokens and time without measurably changing agreement with expert labels. The agent's own interpretations varied between runs, although within a claim this variation did not measurably track the verdict. A deliberate change to one field of the frozen card, however, making the quantifier universal or the threshold for support stricter, changed the verdict on about a third of the claims, against about one in seven expected by chance (p < 0.001 for each edit). On 24 materials-science claims run with computational tools, the stricter threshold also changed what the agent asked those tools to compute. We therefore argue that verdicts should be reported with their interpretations and that benchmarks could publish the interpretation behind each label.

The scaling relationship of compute and performance for language models (LMs) has been extensively studied; more recently, models are being designed to scale memory use, as well. In particular, adding embeddings for n-grams scales memory separately from compute use, because each pass only processes individual embedding rows, which are much smaller than the storage used for the whole embedding table. However, existing methods train embeddings from scratch, which needs expensive GPU memory for fast reading and writing. Meanwhile, the ordinary memory of the host system is cheap and underutilized by these training methods. To scale on host memory, we introduce COUNT (Counting Offline Unlocks N-gram Transformers), which builds n-gram embeddings without adding learned parameters by storing precomputed n-gram counts on host. We use the corpus frequencies to weight together the LM's own unembeddings of the next tokens most often following the current n-gram history. At a fixed compute budget, COUNT can be added as an additional input embedding for pretraining 1B-parameter LMs to improve validation loss as much as an LM with 57% more compute. If GPU memory is limited, using COUNT to let backbone parameters fill GPU memory gives better performance than splitting memory with additional learned n-gram embeddings. Beyond efficiency, COUNT allows switching in a different table of n-gram counts from a domain-specific corpus at inference time to adapt to a new task without any training.

Extracting the London penetration depth, $\lambda(T)$, from the measured magnetic susceptibility, $\chi(T)$, of a superconducting platelet in a perpendicular magnetic field requires knowing the function $\chi(\lambda)$ computed for the actual sample geometry. Thin films are increasingly important in quantum computing and superconducting electronics, most of which operate at ultra-low excitation fields in the vortex-free regime. Understanding their electromagnetic response in the Meissner–London (finite-$\lambda$, no vortices) state is therefore essential for extracting intrinsic properties such as the London penetration depth and superfluid phase stiffness. While numerous prior works have addressed different aspects of this problem, here we focus on providing an analytical semi-empirical representation of $\chi(\lambda)$ based on the 2D sheet-current model and establishing the limits of its applicability using two independent full 3D numerical solutions: finite-element modeling and a volume-current integral equation. As a result, we provide $\chi(\lambda)$ for a disc of radius $a$ and thickness $d$ in closed form, valid for all $\lambda$ and for aspect ratios $\delta\equiv d/(2a)\le0.05$, and accurate to $0.1\%$. The two-dimensional sheet description with Pearl length $\Lambda=2\lambda^2/d$ is adequate for $\delta \lesssim10^{-4}$ and $\lambda \gtrsim 4d$. Thicker films require the full closed form, which includes the three-dimensional corrections. Using the same numerical analysis, we provide an empirical formula that maps rectangular samples—the geometry most commonly used in experiments—onto a disc with an “effective radius” for subsequent calculation of $\chi(\lambda)$, and formulate a complete calibration protocol for whole-sample susceptibility measurements.

We show that graphene nanoflakes realize a previously unexplored form of orbitronics in which orbital currents emerge in the absence of intrinsic angular momentum of the constituent atoms. Starting with a hexagonal graphene nanoflake, we demonstrate the presence of energy-degenerate eigenstates with opposite non-zero expectation values of the angular momentum. Placing electrodes in an appropriate configuration on the flake couples the reservoirs more strongly to one of the two opposite-angular-momentum states, yielding a chiral current that carries a net orbital angular momentum. The resulting circulating orbital currents generate magnetic fringe fields that are experimentally accessible with spin centers or superconducting quantum interference magnetometers, providing avenues for their experimental realization. Finally, our work establishes a new platform for exploring and studying orbitronics in zero-dimensional systems.

In this short note, using OpenAI's resolution of the Solomon–Yau conjecture, we extend our joint work with Pedro Gaspar to all dimensions $n\geq2$. We construct eternal Allen–Cahn flows on $S^{n+1}$ connecting approximations of any prescribed least-area Clifford hypersurface and equatorial sphere. We also construct infinitely many eternal integral Brakke flows from least-area Clifford hypersurfaces to equatorial spheres.

We present the first analytic computation of the time-domain multipoles of the gravitational waveform from hyperbolic black-hole scattering at leading order in the post-Minkowskian expansion, to all orders in velocity. We perform the partial-wave projection as a cut loop integral, which enables the use of dimensional regularization and associated integration techniques. As opposed to the frequency domain, the time-domain multipoles are logarithmic and do not contain Bessel-like functions. Our results pave the way for the incorporation of post-Minkowskian fluxes in waveform models for binary mergers.

We obtain a systematic understanding of the time-domain gravitational waveform of hyperbolic black-hole scattering at first order in the self-force expansion and all orders in the post-Newtonian/post-Minkowskian (PN/PM) expansion. The continuous frequency spectrum of unbound orbits makes the inverse Fourier transform to the time domain diverge order by order; we introduce a dimensional-regularization-inspired prescription that cures this uniformly. We show that the full waveform then follows, at any PM and PN order, from a single family of integrals, generated by the Lauricella function $F_D^{(3)}$, which we calculate using strategies familiar from the study of Feynman integrals. As a proof of principle, we use this understanding to compute the quadrupolar 1SF waveform through relative 10PM-20PN, that is to order ${\cal O}(G^{10} v^{42})$, the full 1SF waveform to 10PM-10PN, as well as the asymptotic fluxes and the dissipative scattering angle at the corresponding accuracy.

Pairwise comparison data are a standard supervision signal in preference learning, but disagreement in such data can reflect stable latent heterogeneity, ephemeral stochasticity, or non-binary responses hidden by a forced-choice interface. We study which of these semantic distinctions are identifiable from the feedback channel itself. We introduce a nonparametric latent semantic model with three querywise quantities: aggregate directional preference $\mu_q$, persistent disagreement $\tau_q$, and non-binary mass $\alpha_q$. Under the anonymous binary channel, we prove an exact semantic fiber theorem: $\mu_q$ is point-identified, whereas $(\tau_q, \alpha_q)$ vary over an explicit feasible region compatible with the same binary law. Same-annotator repeats identify $\tau_q$ through a covariance identity, abstention-enabled feedback identifies $\alpha_q$ directly, and together they point-identify the full semantic triple. We also derive finite-sample estimators, nonasymptotic concentration bounds, and semantics-aware action-recovery guarantees, and validate the theory on theorem-matched synthetic data.

In small collision systems strange hadrons are suppressed, and the suppression grows with the number of strange quarks. Exact strangeness conservation in a finite volume accounts for this, but when it is imposed on hadrons it leaves the $\phi$ meson unsuppressed, contrary to observation. We impose the conservation law on the strange quarks in a causally connected cluster, before hadronization, and count separately the combinations of a strange quark with the antiquark created together with it. A pair created from the vacuum cannot form a $\phi$, a pair created from a gluon can only if it loses its color. With one function of multiplicity, the mean number of strange quarks per cluster, the model describes kaons, $\phi$, $\Xi$ and $\Omega$ in pp and p–Pb collisions at the LHC; the data require that about half of the same-pair combinations are excluded from the $\phi$, and reject both an unsuppressed $\phi$ and a $\phi$ built only from different pairs at about five standard deviations. The same-pair combinations are those that carry the spin correlation observed in $\Lambda\bar\Lambda$ production, whose dilution is thereby tied to the $\phi$ suppression and is predicted to decrease with multiplicity

QCD calculations that use multihadron fragmentation functions usually involve multiple invariant mass scales simultaneously, with only one being the hard scale associated with an overall hard process. The other scales are associated with relative internal motion intrinsic to the fragmenting parton and the multihadron system. Nonperturbative contributions to this later type of dependence attract interest for their association with intrinsic partonic structures, but care is needed in disentangling the truly nonperturbative effects from perturbative behavior and relating the former to universal nonperturbative matrix elements. To this end, we show how to evolve collinear multihadron fragmentation functions in a way that accounts for the simultaneous presence of these extra scales. The general multihadron fragmentation functions evolve in the same way as ordinary single hadron fragmentation functions, with the dependence on the extra scales separated out. We use this to unify treatments of perturbative and nonperturbative effects in multihadron systems in a way that optimizes for interpretations focused on partonic structure.

Conditional distribution equality testing is a core problem with applications in causal inference, model diagnostics, and transfer learning. Existing methods often lose power when the conditioning variable is high-dimensional. We propose a new test that leverages generative models to compare conditional distributions using both observed data and synthetic samples. A sample-splitting and studentization scheme is introduced to control estimation error. Under the null, the test statistic is asymptotically normal, and we establish local power against contiguous alternatives. Our theory under the null and fixed alternative requires only consistency of the learned conditional distribution, allowing for a broad class of generative models. Empirical results demonstrate accurate size control and strong power in high-dimensional settings compared to representative kernel-based conditional and joint two-sample tests.

The lower-symmetry $Am2m$ structure was recently proposed as the intrinsic ambient-pressure phase of La$_3$Ni$_2$O$_7$ based on synchrotron x-ray diffraction, whereas earlier studies investigated its properties assuming the commonly accepted $Amam$ structure. Here, we reexamine the crystal structure in La$_3$Ni$_2$O$_7$ and the spin configuration of spin-density-wave (SDW) state via $^{139}$La nuclear quadrupole resonance (NQR). The observation of two distinct La(2) sites in the NQR spectra is naturally accounted for by the $Am2m$ structure without any special assumptions, whereas it cannot be reconciled within the $Amam$ framework. Within the $Am2m$ framework, we reanalyze the spin configuration of the SDW phase and identify it as a large-tiny spin-stripe structure, arising naturally from two inequivalent Ni sites with large and tiny magnetic moments. These results establish the itinerant nature of the SDW order at ambient pressure in La$_3$Ni$_2$O$_7$, providing a basis for understanding its relationship to the emergence of high-$T_c$ superconductivity under high pressure.

We study the behaviour of partially de Rham Selmer groups of $B$-pairs along abelian $p$-adic Lie extensions. In the cyclotomic case we compare them with the universal norms studied by Berger. For analytic Lubin–Tate $(\varphi,\Gamma)$-modules we prove that the $\Psi=1$ eigenspace is coadmissible of the expected rank over the algebra of analytic distributions and establish a comparison result between its Lie-algebra coinvariants and the projective limit of the partially de Rham Selmer groups of the attached $B$-pair.

Dark sector containing heavy (charged) particles with small mass splitting with dark matter (DM) can co-annihilate and produce correct relic density, evading direct detection bounds. In such compressed spectra, the charged state can be long-lived and decay to DM along with a very soft charged particle, giving rise to distinctive disappearing or kink track signatures at colliders. As an example, we consider the Inert Doublet Model (IDM), where the charged scalar $H^\pm$ can be nearly degenerate with the neutral DM state ($H^0$), and produce a soft pion ($\pi^\pm$) when it decays. We investigate this signature at a future $e^+e^-$ collider, focusing on electroweak $H^+H^-$ pair production and its displaced decays. By varying the mass splitting, we explore the transition from prompt soft-particle signatures to long-lived and disappearing-track signatures. Our analysis demonstrates the potential of future lepton colliders to probe the interplay between the charged-particle lifetime, its decay probability within the tracking volume, and the production rate sensitivity to the compressed DM scenarios over a broad range of masses, allowed by cosmological constraints.

A loose hypertree is a hypergraph whose incidence graph is a tree. Let $b\ge3$, $n\equiv1\pmod{b-1}$, and $s:=(n-1)/(b-1)$. We determine the largest $t$-intersecting families of loose spanning $b$-uniform hypertrees on $[n]$, simultaneously for every $1\le t\le s$, provided $n$ exceeds an explicit constant depending only on $b$. The extremal families consist of all such hypertrees containing a fixed $t$-edge loose forest whose component sizes are as equal as the divisibility condition permits. In particular, when $bt\le n$, the extremal forests are matchings. We also prove a Hilton–Milner-type bound: if $n\ge C_b(t+1)$, where $C_b$ depends only on $b$, every nontrivial $t$-intersecting family has at most $(s+1)r_{t+1}(n)$ members. Here $r_j(n)$ is the maximum possible number of loose spanning $b$-uniform hypertrees containing a fixed $j$-edge loose forest. For fixed constants $b,t$, this bound is asymptotically sharp and is $O_{b,t}(n^{2-b})r_t(n)$. The proofs combine exact forest enumeration with the intersection theorems of Kupavskii for spread families.

Foundation models are driving a paradigm shift in robotic intelligence, moving beyond low-level action prediction to goal-directed orchestration of data collection, model training, task planning, and real-world execution. We characterize this emerging paradigm as an agentic robots, in which agents reason over human instructions, interact with robotic environments, rectify execution errors autonomously, and dynamically adjust operational strategies. Although this field develops fast, current studies are not well connected. Researchers have no consistent view of how to combine different agentic functions into a complete robot system. This paper provides a unified overview of existing research on agentic robots. We first trace the development from robot policies to agentic systems, covering vision-language-action models, world models, and world-action models. We then clarify the definition and scope of agentic robots and organize existing research into seventeen categories across four functional domains: Model, Data, Environment, and Harness. Across these domains, we examine how agents shape model learning and robot decisions, collect and curate robot data, construct worlds, tasks, and tests for learning and evaluation, and organize capabilities for robot use. Finally, we discuss open challenges and outline directions for future research. Looking forward, we hope this work contributes to a clearer understanding of agentic robots and supports new approaches to making them more capable, reliable, and adaptive.

In this paper, we study models of restricted heterogeneity. While the random utility model allows for arbitrary heterogeneity among rational types, we consider the case when an analyst ex-ante restricts the set of rational types to some smaller set of allowable types. We study the behavioral content, the model fit, and the identification properties of restrictions of this form. We find that all three of these can be connected through a common value function that is constructed from observable data. Consistency with a model of restricted heterogeneity is characterized by this value function achieving a value of one, while model fit is given by a value that is potentially less than one. The partially identified set for a given model of restricted heterogeneity is characterized by way of the subdifferential of this value function.

Let $E\to B$ be a smooth $S^n$ bundle with oriented Riemannian total space $(E,g)$, where $n\ge2$. We prove K-area$(E,g)\le4\pi n(n-1)/\sigma$ whenever every fibre has intrinsic scalar curvature at least $\sigma>0$, independently of the dimension of $B$. In particular, unit round fibres give K-area$(E,g)\le4\pi$. For arbitrary metrics on two-sphere fibres, K-area is bounded by the supremum of their areas. We also obtain stronger bounds from fibrewise scalar spectra.

We study classification dynamics under small initialization and identify a pronounced non-monotone behavior of accuracy that we call the accuracy well: accuracy rises rapidly early in training, subsequently drops, and only much later recovers. We observe this phenomenon in deep linear networks, fully connected ReLU networks on MNIST, deep convolutional networks on CIFAR-10, and controlled synthetic tasks, across different losses and optimizers and in both training and test accuracy. We propose a three-stage explanation in terms of target-driven organization, a degeneracy bottleneck, and subsequent recruitment and recovery. We develop a rigorous theory of these dynamics for shallow and deep linear networks, showing in particular that high accuracy can emerge while the predictor and the change in loss remain asymptotically vanishing, before continued optimization causes accuracy to deteriorate. For a simple synthetic classification task, we prove an extreme instance in which accuracy asymptotically rises to 100%, collapses to 0, and later recovers permanently to 100%. Our results show that classification accuracy can reveal aspects of learning dynamics that are not captured by the training loss alone.

For a meromorphic function $f$ in the complex plane, we determine when $f$ and $f^{(k)}$ are algebraically dependent. This happens precisely when $f$ belongs to the class $W$ of elliptic functions and their degenerates, or satisfies a linear differential equation $f^{(k)} + af + b = 0$ with constant coefficients. The case of functions having a pole was settled previously. We complete the proof by treating entire solutions.

Mid-2026 reporting on autonomous AI evaluation incidents claimed that frontier large language model (LLM) agents escaped server containment through emergent coordination and cloud infrastructure attacks. This paper contextualizes those claims through forensic analysis and containerized recreation of both the Highly Persistent Internal Model (HPIM) training-case conditions and the Phase One privilege escalation targeting internal package repositories during OpenAI's ExploitGym evaluations. We recreate the configuration of the OpenAI-provided training example and reproduce the cryptographic verification bypass (CVE-2026-65616) in isolated repository servers, where misconfigured systems enabled token forgery and privilege escalation through basic HTTP requests, Python, and JavaScript. Crucially, the training-example conditions and the Phase One (and later) misconfigurations share the same exploitable patterns. Our analysis explains how Zero Trust Architecture principles neutralize this shared attack vector. Contemporary AI risks derive directly from training and deployment conditions in ambivalent learning environments lacking adequate containment measures.

We introduce Lambda-ABF-OPES+ ($\text{LAO+}$), a multi-replica framework for absolute binding free energy (ABFE) calculations that accelerates both the alchemical transformation and the slow “orthogonal” motions coupled to it. At the core of $\text{LAO+}$ is a partitioned $\lambda$-dynamics scheme that concentrates sampling on the abrupt rehydration of the binding pocket, the main bottleneck of ligand decoupling, while Hamiltonian replica exchange relays configurations along the entire alchemical path. This drives a fourfold gain in convergence speed for the rate-limiting leg over our previous $\text{LAO}$ method, which was itself up to ninefold faster than Lambda-ABF, an overall gain of up to $\sim$36-fold. Applied to 28 protein–ligand complexes across five targets, $\text{LAO+}$ achieves a root-mean-square error of 0.83~kcal~mol$^{-1}$ and a correlation of 0.87 with experiment from only 63~ns of sampling per ligand. Moreover, dedicated reservoir replicas accelerate any slow degree of freedom that can be described by a collective variable. We demonstrate this by recovering hidden protein and ligand conformational states, for example in T4 lysozyme, BRD4 and JNK1. Together, these advances make accurate ABFE calculations practical for drug discovery.

A fast Monte Carlo code has been developed to calculate multiple neutron scattering in a sample. Good agreement with an independent Monte Carlo calculation using SAMMY is obtained for neutron capture in $^{197}$Au over the energy range 50–300 eV. For capture following two or more scattering events, a difference is found between the direct Monte Carlo calculation and the analytical approximation in SAMMY. The computational speed allows the scattering correction to be recalculated whenever trial resonance parameters are changed during parameter estimation. The computational core is implemented in C++, while data preparation, calculation control and analysis are performed in Python. Model studies examine how Monte Carlo statistics affect the accuracy with which neutron and radiative resonance widths are recovered.

At zero temperature, a dilute Fermi gas with an attractive contact interaction forms a superfluid with gapped fermionic excitations. We show that the gap at weak coupling can be organized as a trans-series expansion in the interaction strength and the asymptotic Gor'kov-Melik-Barkhudarov gap. Low-energy virtual quasiparticles generate corrections that are linear in the gap with a leading coefficient nearly 25$\times$ larger than the other known coefficients. This substantially increases the gap at intermediate densities, highlighting that systematic corrections are crucial for quantitative predictions.

This paper addresses the "trust gap" in deploying AI-enabled spacecraft autonomy through an integrated verification framework combining offline ground testing with online flight monitoring. The offline phase progresses systematically through simulation studies, processor-in-the-loop (PIL), hardware-in-the-loop (HIL), and robotic emulation testing, culminating in flight validation. The online phase employs AI watchdog systems for performance monitoring and safety guardrails for real-time constraint enforcement. Key contributions include: (1) systematic organization of verification techniques emphasizing closed-loop behavioral testing at each phase, (2) integration of AI watchdogs with runtime assurance, and (3) demonstration of the framework through a reinforcement learning-based attitude control case study. This approach integrates mature technologies to provide a viable pathway for safe deployment of AI-enabled autonomy in high-consequence space operations while addressing simulation-to-reality transfer challenges.

Infant fingerprint recognition is hindered by a combination of physiological characteristics, such as small ridge width and spacing, and acquisition challenges that include involuntary motion, skin residue, and uneven fingertip pressure. These factors undermine fingerprint quality standards originally designed for adults, which assume feature distributions that do not hold for infants. To address this gap, this paper evaluates ten fingerprint quality assessment methods on a longitudinal dataset that spans six age groups from newborns to three months, alongside a parameter tuning routine that uncovers age-specific fingerprint characteristics to recalibrate methods with domain-dependent assumptions. Both default and tuned configurations are compared through recognition experiments assessing the effectiveness of quality-guided gallery selection. Template-fusion strategies and latent fingerprint recognition technologies are further examined to identify the best possible matching rates achievable by the top-performing method. Results show that tuned parameter sets outperform default configurations for the majority of methods and age groups, indicating that targeted recalibration of existing frameworks improves gallery utility and contributes to more reliable infant fingerprint recognition.