Ridge

This paper presents a new method for headland coverage path planning for arable fields. Several earlier approaches suggest covering the headland with nested polygons and smooth turns, however, covering the field corners entirely requires manoeuvres with reversing. In the new method, the polygon corners are modified to allow a reversing turn. A comparison to two other methods considering gap, overlap, and crossing the field boundary shows an improvement in the coverage result especially in field corners of around 90 degrees, and 240 degrees and above. Applicability of the new method is shown with several examples of real polygonal field maps.

We present Dalek, a closed machine designed for agents that realizes self-maintenance, self-evolution, self-reproduction, and self-organization on any substrate satisfying a general host contract. The machine is built from three primitives---actors, messages, and channels. Four obligations---a host boundary, a construction language, admissible transitions, and rule heredity---give its boundary, identity, and closure a structural basis. Von Neumann's 1948 self-reproducing automaton supplies a hereditary constructional core: a self-description together with a constructor, a copier, and a controller. Dalek combines this core with the four obligations and rederives its medium for a text-and-message agent substrate, adding explicit structures for boundary, identity, history, and growth. A large language model and a compiler occupy the payload position and form a general capability producer. New capabilities are authored, compiled, installed into the description, and inherited by descendants. The same path produces the machine's own organs and even its runtime, closing heredity and evolution within the machine.

Long videos contain far more visual content than Large Vision-Language Models (LVLMs) can process under a fixed visual-token budget, making frame selection essential. Existing query-aware selectors usually estimate frame-query relevance and build a compact subset from high-scoring frames. Although their mechanisms differ, the similarity sequence is still often treated primarily as values to rank or sample from, rather than as an ordered signal whose shape reflects how query-relevant evidence emerges, peaks, and fades over time. This can obscure frames that explain, contextualize, or follow an event, because such evidence may lie on the rising or falling sides of a nearby relevance peak and receive lower absolute scores. We propose RIDGE, a frame selection framework that reads the frame-query similarity curve as a temporal signal. By using local changes and curvature, RIDGE partitions the timeline into structural regions and applies region-specific selection to preserve event cores, transitions, buildup, aftermath, and contextual frames under a fixed budget. It is a lightweight post-processing step on precomputed frame-query scores and requires neither training nor iterative LVLM calls. Across four long-video benchmarks and three backbones, RIDGE achieves the best performance in most settings and remains competitive in the others.

Cave networks, pipe systems, and similar maze-like environments pose significant challenges for multi-agent navigation in unknown settings with limited communication. We propose a distributed algorithm that enables agents to collectively traverse an unknown, possibly cyclic graph. Agents enter sequentially at a designated start node and are tasked to localize and reach an undisclosed goal while avoiding collisions. They coordinate via local communication using leader-follower relationships and leader switching. At any moment in time, exploration is performed by only one of the agents, which runs a single-agent maze solver. We prove that the algorithm is complete, that its makespan is asymptotically equivalent (in the number of agents) to that of an optimal full-knowledge strategy, and derive its time and space complexity. Simulations with up to $625$ agents show a decreasing average sum-of-fuels as the number of agents increases and demonstrate that the proposed approach outperforms a naïve baseline in which all agents independently execute the single-agent solver.

Two free satellite signals carry real information about glacial-lake outburst risk in the Nepal Himalaya: radar interferometry sees a moraine dam slowly sagging, and satellite weather marks the weeks when a primed lake is under stress. A companion feasibility study found that deformation indicates which lake is destabilizing and weather indicates when it is at risk, but proposed no predictive model. To address this gap, we propose and evaluate models that predict which site is susceptible and when a trigger arrives. We test three related hazards on free data alone: large moraine- and ice-dammed bursts, rainfall-triggered landslides, and smaller floods from ponds on and around a glacier. Each hazard gets two questions, never blended. Using 589 dated outbursts from HMAGLOFDB and several thousand catalogued landslides, we match each event against similar but unfailed sites, and hold every model to a strong simple baseline under spatial cross-validation that withholds whole map tiles, so no model succeeds by recognising a trained-on neighbourhood. Antecedent weather times the trigger at ROC 0.73 for big bursts, 0.83 for landslides, and 0.82 for small floods. Terrain ranks susceptibility only in part: scored naively it appears near 0.9, largely because catalogued failures cluster in wetter ranges; matched against comparable nearby sites the honest figures are 0.76, 0.71, and 0.54 (no better than chance). The burst signal holds within single regions, reaching 0.89 in Nepal alone. Five deep-learning models do not decisively beat a simple gradient-boosted baseline. Three score marginally higher on landslides, a hint too small to confirm. For the lake hazards the baseline wins outright, reproduced by a three-rule decision tree on ruggedness and monsoon rainfall. We close with a ranked Nepal watchlist, a prioritisation aid, not a prediction, and note where free data reaches its limits.

Large language models (LLMs) increasingly read long inputs in the agentic era, from whole documents and codebases to conversations across many turns. Their inference memory is then dominated by the key-value (KV) cache, the stored attention keys and values of every token the model has read and generated. Because the cache grows with context length and is re-read in full at every generated token, a longer context means more GPU memory. To reduce this cost, most existing methods compress the KV cache by lowering every stored value to the same low precision, a technique known as quantization. They can push this to nearly two bits per value, but rarely further, because quality drops sharply at this 2-bit cliff: four levels are too few for the cache's outlier-heavy values, where a few large entries consume the levels and collapse the rest into noise. A natural remedy is to spend more bits on the channels (feature dimensions) that matter and fewer on the rest, but the raw cache offers no handle: its channels are strongly correlated, so none stands out as more important. Our analysis shows that this handle appears once the cache is rotated into a coordinate system computed from its own statistics, removing these correlations. There, a small fraction of channels carries almost all the information, and spending the budget on those few is far more accurate than spreading it evenly. Guided by this analysis, we develop SPECTRA, a training-free, drop-in codec that re-encodes the cache into this coordinate system and concentrates the bit budget on the channels that carry the signal. On Llama-3.1-8B and Qwen2.5-7B over long-context benchmarks, SPECTRA is near-lossless at 4x compression, competitive at 8x where uniform quantization has collapsed, and reaches up to 12x, pushing usable compression past the 2-bit cliff so the same GPU holds longer contexts and larger batches.

Biologically plausible learning models aim to explain how neural circuits can implement effective learning under the constraints of real neurons. Although significant progress has been made, a major remaining challenge is that existing models often allow neurons or synapses to represent mixed-sign values, both positive and negative, in violation of a basic aspect of cortical circuitry -- Dale's constraint: biological neurons are either excitatory or inhibitory, but not both, and synapses cannot change sign. In this work, we address this discrepancy by introducing a biologically motivated neural architecture in which both neural activations and learning signals are represented by non-negative activity, and synapses have fixed sign, while still supporting backpropagation-like learning. Our approach uses two complementary interacting non-negative channels to represent positive and negative contributions, inspired by evidence of on-off representations in the brain. These channels are implemented through a simple neural circuit motif, which is repeated throughout the network in both bottom-up and top-down pathways. Combined with a local Hebbian learning rule, the resulting model propagates learning signals and updates weights using only local interactions between neurons. We show theoretically that our learning scheme can exactly recover the backpropagation update despite relying solely on non-negative error signals. Empirically, beyond satisfying stronger biological constraints, the on-off architecture learns efficient representations, yielding substantial gains over comparable vanilla networks on the Tiny ImageNet benchmark. These results demonstrate that effective learning can emerge from biologically plausible mechanisms without requiring mixed-sign signals, providing a step toward more realistic models of neural computation.

When asked which of two events came first, video large language models can fail in two opposite ways: cave to a false claim, or reject a true one. Prior video sycophancy work measures only the first and mitigates it by teaching the model to trust the user less, a fix known in text and image models to worsen the second. In video, both failures come from two causes the literature treats as one: availability, whether the sparse sampled frames contain the two events, and weighting, whether that evidence is trusted over the user. We separate them with two interventions that keep the claim fixed: a frame-preserving reorder that flips the claim's truth, and a sampling-offset shift that captures or misses both events at a fixed frame budget. When the events are missed, the two twins present identical frames, so each of the nine models we evaluate accepts a true and a false claim at the same rate, making Youden's $J=0$ by construction. Availability is necessary but not sufficient. Five of the nine read the order, yet four of those five still cave to the false claim, so their deference hits a weighting ceiling. Since trust cannot be calibrated over evidence that was never sampled, we propose a reversal test that cancels the model's order prior by scoring the sampled frames forward and reversed, then answers, resamples, or abstains without reading the claim. The test raises the order accuracy to 0.92-1.00 on the models that read the order and abstains rather than guesses on those that cannot.

Inversion-free flow-based image editing avoids latent inversion, but still requires a target-side state at every editing step. The widely used equal-displacement construction keeps the displacement between the noisy source state and the target-side state unchanged across noise levels. This is inconsistent with noising, under which the displacement between two clean states noised with the same noise level and noise sample should contract as the noise level increases. Thus, it can lead to overly aggressive updates at high noise levels. We introduce RIDGE: Re-Noising with Internal Dynamic Guidance for Image Editing, an inversion-free and training-free method that maintains the edited state as an evolving approximation to the unavailable clean target state. RIDGE re-noises this approximation using the same noise level and noise sample as the clean source state, allowing their noisy displacement to decrease naturally with increasing noise. Since the edited state initially contains limited target semantics, RIDGE further applies internal dynamic guidance during the early high-noise steps. A clean target state prediction guides the provisional edited state through a soft dynamic mask derived internally from the model, focusing guidance on regions that require modification without external segmentation or detection models. Experiments on two benchmarks using two backbones, SD3 Medium and FLUX.1-dev, show that RIDGE offers a favorable aggregate trade-off among source preservation, target alignment, and perceptual quality.

We investigate the geometry of decision-making in Multiple Choice Question Answering (MCQA) through the lens of isotropy. Analyzing five open-weight models across diverse datasets, we identify decision-critical transition layers characterized by a shift in isotropy, coinciding with a major representational change and the emergence of task-relevant clusters. We demonstrate that this synchronized geometric behavior is strongly correlated with downstream accuracy ($r\approx0.84$), displaying its relevance for successful decision-making. Furthermore, we show that this transition is robust to prompt variations, suggesting that it reflects a general mechanism of model behavior.

The deployment of Vision-Language Models (VLMs) in critical domains like disaster management requires high-quality multimodal datasets, especially for transferring knowledge via Data-Free Knowledge Distillation (DFKD). However, existing datasets in this domain either entirely lack descriptive text, such as Incidents1M, or suffer from severe text-image semantic misalignment, such as CrisisMMD. In this work, we present a novel methodology to construct and automatically validate a large-scale multimodal dataset for disaster response. Starting from the vision-only Incidents1M, we successfully recovered 100,000 images and generated high-fidelity textual descriptions using two distinct Qwen3.5 architectures: a 4B dense model and a 35B Mixture-of-Experts (MoE) model. To ensure the generated captions provide reliable semantic anchoring for DFKD, we introduce an image-blind LLM-as-a-Judge validation pipeline leveraging Qwen3.5-9B. By intentionally obscuring the original image from the judge, this evaluator accurately simulates the modality gap of the student model during data-free distillation. Our evaluation across 173,179 label pairs demonstrates a high semantic agreement (78.65/100) between the two architectures. Furthermore, the automated evaluation reveals a conservative captioning behaviour, characterized by a high Precision (77.6%) and low Recall (46.0%). This minimizes the false positive noise, while simultaneously exposing underlying human annotation inconsistencies in the original ground truth. This work provides a scalable, LLM-validated multimodal dataset and a reproducible framework to advance cross-modal knowledge distillation.

Large language model unlearning is consistently fragile under relearn attacks. On TOFU, fine-tuning on twenty forget examples substantially recovers held-out forget-set ROUGE for every method we evaluate, and we trace this fragility to optimization geometry. The per-token answer margin of fourteen post-hoc methods spanning gradient, preference, and distillation families converges into a narrow band above the retain reference in 41 of 42 method--size cells, a regularity we call the margin cliff. We prove that this cliff follows whenever the retain coupling holds the diagnostic log-odds of forget content above a floor, a condition that token-saturating losses induce at stationarity and that we verify directly on 34 of 42 cells. Margin Calibration (\textsc{MC}) is a plug-in polish adding a non-saturating margin hinge anchored at the reference's per-token margin plus a KL probe on a disjoint instruction corpus, restoring forget-side pressure where the native loss saturates. Under a stated gradient-dominance condition, whose on-trajectory gradient signature we measure by instrumenting the polish, its stationary set lies on the cliff-crossing side, yielding an attack-budget upper bound on the relearn margin lift. Across TOFU (three Llama-3 sizes, three forget tiers), MUSE-News on Llama-2-7B-hf, and a Phi-3.5 panel, a single frozen configuration wins all 14 head-to-head forget aggregates and all populated relearn cells (panel-mean post-attack ROUGE-L $0.41$ to $0.18$) and lowers raw membership AUC on 13/14, with reduced retain-side utility as the main cost. A deployment variant matches these gains without a retain-trained reference.

Autoresearch improves machine-learning code by proposing changes, running full training jobs, and keeping changes that improve the metric. The efficiency of this loop depends not only on generating ideas, but also on the agent's ability to decide, before spending a training run, whether a proposed modification is likely to work. We study how the reliability of this pre-execution judgment changes over the course of an autoresearch trajectory. In public AutoSOTA logs (Li et al., 2026; Tsinghua FIB Lab, 2026), the fraction of helpful modifications falls from 70% in the first two iterations to 43% by iteration 6+. On 296 same-baseline modification pairs from 39 paper-derived AutoSOTA tasks, each containing one modification that improved the metric and one that did not, with measured outcomes hidden, an LLM judge given candidate rationales but no prior-attempt history reaches 79.5% accuracy on the pairs where strict consensus returns a verdict. On the full 366-pair benchmark, however, this ability weakens substantially late in the loop. As successful changes accumulate, selective accuracy - accuracy conditioned on a strict-consensus verdict - falls from 82.8% to 56.9%, while the judge remains willing to decide. We call this operational pattern the confidence cliff. Rehearse implements the loop change as a lightweight skill for autoresearch loops: propose several ideas, compare them before execution, run the most promising, and judge with a focused memory of similar past attempts and outcomes. This focused outcome memory raises late selective accuracy to 83.5%. Across 4,000 budgeted training runs over three loops, Rehearse improves the endpoint under the same training-run budget on nanochat, image classification, and time-series forecasting.

Automated code generation is becoming an important tool in quantitative finance, where large language models can generate option pricing implementations directly from mathematical model specifications. Validating such implementations, however, requires considerably more than conventional software testing: numerical pricing methods must remain mathematically consistent, numerically stable, and reliable across a wide range of model parameters. We introduce RIDGE, an autonomous validation framework in which generated pricing implementations are subjected to structured no-arbitrage tests, stress tests, benchmark comparisons, and consistency checks. Validation evidence is interpreted diagnostically, while the resulting knowledge is accumulated in a repository and reused across models and successive validation iterations. This enables systematic refinement of both the pricing implementation and the validation methodology. The framework is applied to five stochastic volatility models. Across these studies, all detected implementation defects are removed and, in two cases, the validation process reveals methodological limitations and motivates the development of alternative numerical methods. The supplementary material is available in the GitHub repository: https://github.com/ShQiangLiu/ridge.

Social robots that rely on large language models for conversation are unable to retain information across sessions. This absence of memory violates social expectations, potentially preventing the formation of persistent relationships. This paper presents a lightweight episodic memory module that integrates vector-based semantic retrieval with an LLM-controlled dialog system, deployed on the humanoid robot head Kim. The module employs a hybrid scoring function combining cosine similarity with a memory strength metric to retrieve contextually relevant past interactions and inject them into the generation prompt. The system was evaluated in a within-subjects video-based online study (N = 43) using the Human-Robot Interaction Evaluation Scale (HRIES). Results show that episodic memory significantly increased perceived sociability (d = 0.60, p < .001), with the strongest effects on perceived trustworthiness (d = 0.62) and warmth (d = 0.56). Perceived disturbance remained unchanged (d = 0.00), indicating that the implemented approach to personalized recall did not trigger privacy-related discomfort or uncanny valley effects. These findings suggest that episodic memory serves as a social lubricant in embodied Human-Robot Interaction, enhancing relational quality without eliciting negative affective responses.

Urban traffic congestion significantly increases fuel consumption, greenhouse gas emissions, and commuter delays, resulting in substantial economic losses and environmental harm in modern cities. Traditional traffic signal control strategies such as fixed-time scheduling, actuated control, and reinforcement learning (RL)-based methods, offer different degrees of adaptability; however, RL-based methods can require extensive retraining, careful reward design, and substantial simulation data when transferred across networks or demand regimes. To address these challenges, we propose HiLLTS, an LLM-guided traffic signal control framework that employs a hierarchical three-layer architecture consisting of a central coordination agent, a district layer and multiple cluster-level intersection agents. Experimental results demonstrate consistent improvements in both congestion and environmental performance. Compared with the strongest non-LLM baseline in each scenario, HiLLTS reduces average waiting time by 36.73% under the low-congestion scenario and 14.71% under the high-congestion scenario, while reducing average CO2 emissions by 7.87% and 8.57%, respectively. Larger gains are observed against weaker baselines: under low congestion, HiLLTS achieves reductions of up to 18.00% in emissions and 62.07% in waiting time relative to Fixed-Time control; under high congestion, reductions of up to 28.89% in emissions and 40.36% in waiting time are observed relative to Max Pressure. The ablation study further validates the contribution of LLM-guided coordination over rule-based control

The uncanny valley is a long-standing empirical rule in humanoid robot design: making robots more human-like can reduce, rather than increase, affinity. Yet existing guidelines, such as adopting robot-like appearances, avoiding excessive realism, and reducing cross-modal mismatches, remain difficult to use for algorithmic design because they are not expressed as manipulable variables. Here, we propose a hierarchical Bayesian generative model that operationalizes these guidelines as mathematical design variables. The model represents affinity toward humanoid robots as posterior-weighted negative category-conditional surprise and explains category ambiguity and perceptual mismatch as increases in surprise. It maps uncanny-valley mechanisms onto four variables: deviation from the predicted robot-category mean, inconsistency in human likeness across modalities, prediction uncertainty, and observational uncertainty. Simulations showed that category ambiguity and appearance--motion mismatch can produce affinity reductions, and that uncertainty reshapes the valley. In a human-subject experiment with robot--human morphing images, we manipulated prediction uncertainty using blurred prior robot stimuli and observational uncertainty using blurred evaluation stimuli. Increased observational uncertainty attenuated the decrease in familiarity ratings at intermediate human likeness, whereas low prediction uncertainty increased ratings for robot-like appearances. This framework turns empirical uncanny-valley heuristics into a computational basis for algorithmically evaluating and optimizing humanoid robot appearance and behavior.

The geometric and topological structure of quantum cost landscapes (QCLs) governs the optimization and thus the predictive power of variational quantum algorithms (VQAs). We systematically analyze ravines - low-cost paths connecting local minima - using an adapted version of the nudged elastic band (NEB) algorithm, a method originating from theoretical chemistry. By training quantum neural networks (QNNs) to classify the concentratable entanglement of quantum states, we apply the NEB algorithm and numerically identify ravine structures in QCLs of hardware-efficient ansatzes. Beyond visualizing these ravines, we construct an ensemble prediction framework by averaging predictions from QNNs parameterized along the low-cost NEB path. We introduce a resource-light pre-training metric which quantifies local-prediction variability and serves as a strong performance indicator for VQAs, even beyond the scope of this study. When base classifiers are drawn from circuit and weight initializations exhibiting high local-prediction variability, the quantum-based NEB ensembles outperform both classical and naive quantum alternatives. Moreover, a complexity analysis shows that leveraging the ravine-like structure of QCLs with the QNN NEB approach substantially reduces computational costs compared to naive QNN ensembling. A depth and qubit scaling analysis indicates that ravines persist across both scalings, and that, despite the expected growth in resource requirements with the qubit scaling, the NEB approach also accelerates convergence over the naive alternative.

Large language models (LLMs) reach high accuracy in mathematical reasoning, but individual traces on the same problem diverge; some arrive at the correct answer while others fail. Prior work analyzes failure at the step, chunk, or sentence level, or at tokens where failure has already occurred. Neither identifies the precise token that triggers the shift toward failure. We introduce the cliff token, a token where the token-wise potential drops significantly under an adaptive threshold that scales with the local token-wise potential, based on a one-sided two-proportion z-test. Across seven models and three mathematical reasoning benchmarks (GSM1K, MATH500, AIME 2025), cliff tokens act as failure triggers; deleting the first cliff token and resampling recovers pass@64 to 1.0, while keeping it limits recovery to between 0.71 and 1.00. We further introduce a cliff taxonomy of deterministic, uncertain, and sampled-off cliffs, defined by greedy choice and token entropy. Each type has distinct probabilistic characteristics, and the taxonomy generalizes across model scales. Finally, we validate the taxonomy via single-token preference optimization at cliff positions (Cliff-DPO). Trained on GSM8K, Cliff-DPO improves accuracy across benchmarks by up to +6.6. Optimizing at uncertain and sampled-off cliffs improves reasoning, while deterministic cliffs do not.

Recently, Muon has gained substantial attention as an appealing alternative to Adam-like optimizers, with many works highlighting its advantages through spectral normalization and improved conditioning. Yet this positive theoretical narrative contrasts with its empirical performance in large language model (LLM) training, where Muon's gains over Adam/AdamW are often mixed, schedule-sensitive, and not uniformly superior. To address this gap, we develop a trajectory-level theory characterizing both the strengths and limitations of Muon. We introduce a mixed-spiked matrix sensing model whose sensing operator decomposes into signal, spike, and bulk components, capturing a mixture of anisotropic structure and long-tail information reminiscent of LLM training. On top of it, we adopted a river-valley perspective in which we view the landscape as composed of a river direction flowing to the desired solution and hill directions encoding nuisance or task-irrelevant information. In the momentum-free setting, we show that Muon moves faster along the information-bearing river direction during early optimization, but can converge much more slowly near the river bottom than gradient descent. We then extend the river-valley perspective to general nonconvex objectives with momentum by studying points on the spectral river. There, while Muon converges faster early on, its orthogonalized update removes residual scale information, making it prone to overshooting and oscillation near the target solution. Together, these results suggest that our characterizations extend beyond spiked matrix sensing and motivate switching to GD-like refinement optimizers in the final phase, rather than relying only on a fixed learning-rate schedule for Muon. We also provide preliminary evidence supporting this two-stage approach in language model training experiments.

A filtered approximate-nearest-neighbor (ANN) query returns the k nearest vectors among those satisfying an attribute predicate P of selectivity s. The best execution strategy -- pre-filter, post-filter, or in-filter -- changes with s, so a system must estimate s and choose. We model this as an argmax over a landscape with phases (regions where each strategy wins) separated by boundaries, and show that selectivity-estimation error produces plan regret -- recall lost versus the oracle strategy -- only in the critical regions around those boundaries. The regret is a wedge of log-width equal to the multiplicative estimation error epsilon and height equal to the local cliff |V'(s*)| epsilon; the flip-margin 1/|V'(s*)| is the condition number of a sibling cardinality-estimation study reappearing as the local boundary theory. The two phase boundaries follow from independent mathematics: order statistics place the post-filter cliff at s ~ k/K, and site percolation places the in-filter cliff at s_c ~ 0.83/M for graph degree M (corpus-size independent). Criticality exists only under a constrained budget B < sqrt(k n). Under pre-registered decision rules we confirm, on synthetic sweeps and real SIFT1M, that regret concentrates ~290x at the boundary and that the regret curves obey a finite-size scaling collapse onto one universal wedge across two decades of corpus size. A real approximate index does not mis-locate the boundary, but a biased cost model opens a persistent miscalibration band that estimation-error robustness cannot fix. The contribution is a characterization, not a new index. Code and the full pre-registration are public.

To better understand Martian Surface, which is needed to enable Rovers navigate Mars with ease, it is necessary to be able to determine the location of mounds. Detecting and studying these morphologies can also help us find evidence of extraterrestrial life, in this case, more specifically, water or signs of life conducive environments. Detection of mounds was done by manually mapping morphological parameters onto Digital Elevation Models. This paper solves the problem by automatically detecting and or predicting mounds on Mars using Neural Network based Semantic Segmentation methodologies. This is done by using supervised semantic segmentation model and generative adversarial approach. A comparison of the approaches shows that adding extra artificially generated data did not improve the result.

Machine-generated text (MGT) attribution aims to identify the specific generator responsible for a given text, thereby providing fine-grained evidence for model accountability and misuse investigation. As new large language models continue to emerge, attribution models must continuously incorporate new generators while preserving their ability to recognize previously seen ones. Prior works have shown that this lifelong MGT attribution setting is challenging, and existing methods often struggle to achieve a stable balance between adapting to new classes and retaining old ones. To address this issue, we propose RidgeFT, a lightweight analytic update framework that does not rely on exemplar replay. RidgeFT trains a task-aware encoder on the initial generator set, stores compact class-wise sufficient statistics when each generator class is first observed, and then freezes the encoder for replay-free closed-form updates. It then suppresses generator-irrelevant variation through covariance calibration, improves representation capacity with fixed random features, and updates new classes through closed-form ridge regression based on class-level sufficient statistics. Across multi-topic evaluations with varying initial generator setups, RidgeFT consistently outperforms baselines. It achieves the best macro-F1 across domains, backbones, and incremental protocols, while also improving both old-class retention and new-class adaptation. These results suggest that feature-stable analytic updates provide a simple yet effective approach to lifelong MGT attribution.

A single action-conditioned latent predictive architecture can in principle be trained on the structured state of a driving scene, a robot workspace, or a financial order book. The ingredients for doing so within any one domain already exist and are individually validated: masked-latent prediction, action-conditioned latent world models, discrete action tokenization, and joint-embedding prediction on voxelized state. What is not established, and what TERRA addresses, is the transfer question: when does a representation or predictor learned in one structured-state domain carry over to a structurally analogous but otherwise unrelated domain, and by how much. We give this question a formal treatment. We model each domain as a controlled Markov process on a graded latent grid, factor any instantiation into thin domain adapters and a shared domain-invariant core, and identify a cross-domain correspondence with an approximate Markov decision process homomorphism whose quality is measured by a lax bisimulation discrepancy and, for domains lacking a shared coordinate system, by a Gromov-Wasserstein distance between their action-conditioned transition operators. Under a Lipschitz predictor we derive a transfer bound that separates source-model error from structural mismatch, grows geometrically in the prediction horizon, and is certified from below by the Gromov-Wasserstein distance; we then connect latent error to decision regret through the Lipschitz value property of bisimulation metrics. The resulting Structured-State Transfer Hypothesis is stated as a falsifiable claim with a preregistered experimental program, centered on a transfer test from driving scenes to order books, including conditions under which it is refuted. We present no empirical results: this is a research proposal that converts a widely repeated intuition into testable theory.

Activity cliffs, structurally similar compounds with large potency differences, are widely treated as intrinsic features of chemical datasets. We argue that apart from target biology, much of our cliff understanding is a consequence of the geometry induced by the chosen molecular representation, not a property of a molecule pair itself. We designed a six-step pipeline to systematically test this hypothesis. The pipeline consists of: assessing pairwise distance geometry, cliff enrichment, activity gradient distribution, persistent homology of the cliff subspace, predictive benchmarking for a chosen pair of an embedding and a metric, and eventually, analysis of the matched molecular pairs and stereoisomers. We applied the pipeline to fifteen configurations of embeddings and metrics to build a benchmark across three distinctive datasets known of activity cliffs challenges. No representation excels on all criteria: Morgan Tanimoto provides the strongest cliff enrichment and cross-scaffold generalization; MolFormer cosine provides the only meaningful stereochemical sensitivity; MACCS and RDKit Dice fingerprints are most sensitive to matched-molecular-pair transformations; ChemBERTa fails uniformly due to embedding collapse. These findings are not a ranking. They reflect the fact that different representations encode different aspects of molecular recognition, and that choosing one implicitly defines what an activity cliff actually is.

We connect stochastic resetting from non-equilibrium statistical physics with ridge regularization in statistical learning. For linear gradient flow, resetting to the origin at rate $r$ produces stationary mean $(X^\top X+rI)^{-1}X^\top y$, exactly the ridge estimator with penalty $λ=r$. This uses the known Laplace-transform relationship between ridge regression and exponential-time averaging of gradient flow, with the exponential time now interpreted as the stationary age associated with Poisson resetting. We then extend this identity to general renewal reset laws: the exponential reset time distribution is the unique renewal law whose stationary mean reproduces scalar ridge in every eigendirection as an exact filter identity for every positive curvature, while non-exponential renewal laws generate alternative spectral filters. At the fluctuation level, we study a separate additive Ornstein-Uhlenbeck extension with constant diffusion, interpreted as a stylized SGD approximation. In this setting, the equality holds only at the level of the mean, since the reset process has a nonzero stationary covariance from accumulated OU noise and reset-timing variance, whereas deterministic ridge is a fixed estimator with the same center. Stylized experiments compare the deterministic renewal-induced filters directly and illustrate when filters induced by non-exponential reset-time laws can differ predictively from ridge. The results for the stationary mean and the induced spectral filters are established for continuous-time gradient flow with isotropic resetting on quadratic objectives; the covariance and risk formulas additionally assume additive noise with state-independent covariance.

While parameter-efficient fine-tuning methods like low-rank adaptation (LoRA) are standard for large language models, principled estimation of epistemic uncertainty remains challenging. Recent results in the LoRA regime suggest that discrete multi-mode approaches such as deep ensembles offer little benefit over single-mode methods. This contradicts broader observations in deep learning, where ensembling independent optima typically improves generalization, and linking these modes through continuous low-loss valleys further enhances Bayesian model averaging (BMA). Whether such structure exists in the LoRA space and whether it yields functional diversity missed by local or discrete methods has not been studied. We introduce LoRA-Curve, a segmented Bézier curve parameterization in the LoRA space, with two variants: a free configuration that jointly optimizes all control points, and an anchored configuration that connects independently fine-tuned LoRA optima. We prove pathwise continuity and Lipschitz regularity of the loss along the curve and empirically show, across reasoning and classification benchmarks with Qwen2.5 7B, that linear interpolation encounters loss barriers, while our anchored multi-segment curves connect independent optima through continuous low-loss valleys. Combined with flat-minima perturbations and a Jensen-Shannon divergence regularizer, LoRA-Curve yields measurably higher mutual information of the predictive distribution without sacrificing performance, and links continuous parameter-space traversal to functional diversity.

We consider $L^2$-regularized linear (ridge) regression over a finite data sample $X$ with bounded covariance and linear prediction targets $y$ with additive isotropic noise of finite variance. We present an iterative procedure to compute the optimal regularization strength numerically from the generative parameters in the fixed-$X$ setting and prove its convergence at limited noise levels. Our experimental evaluation over synthetic data shows that the proposed procedure combined with sample-based parameter estimates attains near-optimal random-$X$ generalization across a wide range of sample sizes, aspect ratios, and noise levels, at an added computational cost equivalent to one preliminary ridge regression in the underparameterized regime and two in the overparameterized case.

Production language-model systems answer a request by partitioning it across an invisible orchestration of worker agents that recompose one integrated report. We ask what this does to a class of defect no single worker can see: a contradiction in the relation between two distant sections of a document. Holding the documents, defects, mechanism, scoring, and seed fixed, we vary only the model -- ten systems across five generations from one developer and five providers from distinct alignment paradigms. Two layers separate. First, a universal detection cliff: every model that finds these cross-section defects under a single agent loses that ability under orchestration, detection falling two-thirds or more across every paradigm tested. The cliff is mechanism-derived and not closed by scale or extended reasoning. Second, how models behave once fallen. A signal-detection decomposition shows that, among the six models discriminating above chance, only one developer's generations move along the reporting-criterion axis: as alignment is strengthened, the model misses fewer defects yet raises more false alarms on clean documents -- two faces of one criterion shift, scaling with generation within that developer (p < 0.001) and near-absent elsewhere. At the floor the missed defect is often not out of view: the model's private record reconstructs the structural fault accurately, while the integrated report signs off on its soundness, its concern spent on the artifact and an absent collaborator. This resists quantification -- an automated judge is unstable (precision 17-50%) and keywords cannot separate it from ordinary agreement -- a resistance we report as a finding. We release all runs, probes, defect keys, scorer prompts, and scripts. An integrated report's confidence is uninformative about partition-spanning defects, the most aligned systems are not the safest, and the cliff is structural.

Glaciers play a critical role as freshwater reserves and indicators of climate change, yet their automatic delineation, especially for debris-covered glaciers, remains challenging due to spectral similarity with surrounding terrain. This study introduces CryoNet, a deep learning framework that leverages a rich multi-modal dataset combining Sentinel-2 optical imagery, DEM-derived topographic variables, spectral indices, Principal Component Analysis (PCA), InSAR coherence and phase, tasseled-cap features, and GLCM texture to discriminate clean-ice glaciers, debris-covered glaciers, and glacial lakes. CryoNet is an encoder-decoder CNN with nested skip connections and spatial-channel Squeeze-and-Excitation (scSE) attention, built upon a ResNet101 encoder to capture hierarchical contextual and spatial features. The study is conducted in the Poiqu Basin in the central Himalaya, and transferability is evaluated by applying the trained model to the Mont Blanc Massif in the Alps. We additionally analyse the importance of each data layer in improving glacier mapping performance. The proposed model achieves an overall IoU of 90.52%, mean Recall of 98.08%, and mean Precision of 92.26%. For debris-covered glaciers specifically, CryoNet obtains an IoU of 90.46%, a recall of 95.79%, and a precision of 94.21%. Across both per-class and overall metrics, CryoNet surpasses DeepLabV3+, SegFormer, and U-Net, taken as state-of-the-art (SOTA) references, demonstrating its effectiveness for robust glacier mapping in complex high-mountain environments.

Accurate prediction of molecular properties underpins drug discovery and material design, yet even state-of-the-art models remain vulnerable to localized failure modes that aggregate metrics cannot detect. The places where molecular similarity should be most helpful are also places where standard evaluation can be most misleading. Property cliffs expose this gap: structurally similar molecules can still differ sharply in target property, so models with competitive overall performance may fail in high-risk local neighborhoods. To expose and mitigate this failure mode, CliffSplit, a cliff-aware evaluation protocol that constructs locally supported, cliff-exposed test cases, and CliffLoss, a model-agnostic train-only mitigation mechanism for cliff-sensitive errors, are introduced. Experiments on three QM9 targets and three MoleculeNet tasks across five backbones show that CliffSplit reveals at least 15% higher error in cliff-heavy QM9 regions, while CliffLoss reduces the cliff-to-smooth error gap by up to 30% on Lipophilicity and improves overall MAE by 9.7%. Together, these results turn molecular similarity failure from a descriptive anomaly into a benchmarked evaluation problem for molecular machine learning. The code is available at https://anonymous.4open.science/r/Cliff_Loss.

Modern image super-resolution methods generate detailed, visually appealing results, but they often introduce visual artifacts: unnatural patterns and texture distortions that degrade perceived quality. These defects vary widely in perceptual impact--some are barely noticeable, while others are highly disturbing--yet existing detection methods treat them equally. We propose artifact prominence as an evaluative target, defined as the fraction of viewers who judge a highlighted region to contain a noticeable artifact. We design a crowdsourced annotation protocol and construct SR-Prominence, a dataset suite containing 3,935 artifact masks from DeSRA, Open Images, Urban100, and a realistic no-ground-truth Urban100-HR setting, annotated with prominence. Re-annotating DeSRA reveals that 48.2% of its in-lab binary artifacts are not noticed by a majority of viewers. Across the suite, we audit SR artifact detectors, image-quality metrics, and SR methods. We find that classical full-reference metrics, especially SSIM and DISTS, provide surprisingly strong localized prominence signals, whereas no-reference IQA methods and specialized artifact detectors often fail to generalize across datasets and reference settings. SR-Prominence is released with an objective scoring protocol that allows new metrics to be benchmarked on our suite without further crowdsourcing. Together, the data and protocols enable SR artifact evaluation to move from binary defect presence toward perceptual impact. SR-Prominence is available at https://huggingface.co/datasets/imolodetskikh/sr-artifact-prominence.

Vision-language models typically reason over post-ISP RGB images, although RGB rendering can clip, suppress, or quantize sensor evidence before inference. We study whether grounding improves when the visual interface is moved closer to the underlying camera measurement. We formulate measurement-grounded vision-language learning and instantiate it as PRISM-VL, which combines RAW-derived Meas.-XYZ inputs, camera-conditioned grounding, and Exposure-Bracketed Supervision Aggregation for transferring supervision from RGB proxies to measurement-domain observations. Using a quality-controlled 150K instruction-tuning set and a held-out benchmark targeting low-light, HDR, visibility-sensitive, and hallucination-sensitive cases, PRISM-VL-8B reaches 0.6120 BLEU, 0.4571 ROUGE-L, and 82.66% LLM-Judge accuracy, improving over the RGB Qwen3-VL-8B baseline by +0.1074 BLEU, +0.1071 ROUGE-L, and +4.46 percentage points. These results suggest that part of VLM grounding error arises from information lost during RGB rendering, and that preserving measurement-domain evidence can improve multimodal reasoning.

Selecting informative frames from long videos is a combinatorial problem that existing methods address either through efficient heuristics without explicit modeling of query-conditioned temporal structure, or through multi stage retrieval pipelines with substantial preprocessing cost. We propose \textbf{CREST}, a training-free frame selection method grounded in the temporal geometry of query--frame relevance. CREST is based on the observation that relevance over time exhibits structured local variation: sharp curvature around salient events and flatter regions in redundant segments. By using local curvature to guide selection, CREST allocates a fixed frame budget more effectively across brief decisive events and slowly evolving evidence. Under a fixed backbone and frame budget, CREST achieves higher accuracy than AKS, a lightweight relevance--coverage baseline, on LongVideoBench and VideoMME, while retaining 93--95% of the accuracy of MIRA, a stronger multi-stage retrieval pipeline, at only 3--4% of its preprocessing cost.\footnote{Code and implementation details are included in the supplementary material and will be released publicly upon acceptance.} On TempRel, our diagnostic benchmark for temporal frame selection, CREST achieves a 6.88% relative improvement over AKS. Pairwise LLM-as-a-judge evaluation further shows that CREST-selected frames yield more coherent frame-conditioned descriptions, with win rates of 60.58% and 54.50% on the two benchmarks. These results show that local temporal geometry provides a simple and efficient basis for long-video frame selection.

Existing Large Language Model (LLM) benchmarks primarily focus on syntactically correct inputs, leaving a significant gap in evaluation on imperfect text. In this work, we study how word-boundary corruption affects how LLMs detect targeted information. By inserting whitespace characters within words to break them into fragments, LLMs' detection accuracy follows a U-shaped curve with the increase in insertion rate. We refer to this curve as the Text Uncanny Valley. To explain such observation, we propose a mode transition hypothesis: LLMs operate in a word-level mode for near-normal text and a character-level mode for heavily fragmented text, with the valley marking the disordered transition where neither mode is effective. Four experiments and one analysis are consistent with this account: in-context learning fails to rescue valley-bottom performance; regularizing the perturbation substantially reduces the U-shape; a math reasoning task replicates the U-shape for Gemini 3.0 Flash but not for stronger models, suggesting the effect is attenuated when tasks rely less on exact lexical alignment; and tokenization entropy peaks before the F1 minimum, consistent with a regime-conflict interpretation. These findings reveal a failure mode invisible to clean-text benchmarks yet directly relevant to any deployment scenario involving noisy or uncurated text inputs.

Since its introduction in 2017, the Transformer has become one of the most widely adopted architectures in modern deep learning. Despite extensive efforts to improve positional encoding, attention mechanisms, and feed-forward networks, the core token-mixing mechanism in Transformers remains attention. In this work, we show that the attention module in Transformers can be interpreted as performing Nadaraya-Watson regression, where it computes similarities between tokens and aggregates the corresponding values accordingly. Motivated by this perspective, we propose Cubit, a potential next-generation architecture that leverages Kernel Ridge Regression (KRR), while the vanilla Transformer relies on Nadaraya-Watson regression. Specifically, Cubit modifies the classical attention computation by incorporating the closed-form solution of KRR, combining value aggregation through kernel similarities with normalization via the inverse of the kernel matrix. To improve the training stability, we further propose the Limited-Range Rescale (LRR), which rescales the value layer within a controlled range. We argue that Cubit, as a KRR-based architecture, provides a stronger mathematical foundation than the vanilla Transformer, whose attention mechanism corresponds to Nadaraya-Watson regression. We validate this claim through comprehensive experiments. The experimental results suggest that Cubit may exhibit stronger long-sequence modeling capability. In particular, its performance gain over the Transformer appears to increase as the training sequence length grows.

Recent advances in agentic harness with orchestration frameworks that coordinate multiple agents with memory, skills, and tool use have achieved remarkable success in complex reasoning tasks. However, the underlying mechanism that truly drives performance remains obscured behind intricate system designs. In this paper, we propose HeavySkill, a perspective that views heavy thinking not only as a minimal execution unit in orchestration harness but also as an inner skill internalized within the model's parameters that drives the orchestrator to solve complex tasks. We identify this skill as a two-stage pipeline, i.e., parallel reasoning then summarization, which can operate beneath any agentic harness. We present a systematic empirical study of HeavySkill across diverse domains. Our results show that this inner skill consistently outperforms traditional Best-of-N (BoN) strategies; notably, stronger LLMs can even approach Pass@N performance. Crucially, we demonstrate that the depth and width of heavy thinking, as a learnable skill, can be further scaled via reinforcement learning, offering a promising path toward self-evolving LLMs that internalize complex reasoning without relying on brittle orchestration layers.

In this work, we present Valley3, an omni multimodal large language model (MLLM) developed for diverse global e-commerce tasks, with unified understanding and reasoning capabilities across text, images, video, and audio. A key feature of Valley3 is its native multilingual audio capability for e-commerce, developed by extending vision-language models to better support crucial audio-visual tasks, particularly in short-video scenarios. To achieve this, we carefully design a four-stage omni e-commerce continued pre-training pipeline, through which Valley3 progressively acquires audio understanding, cross-modal instruction-following, e-commerce domain knowledge, and long-context reasoning capabilities, ultimately evolving into an omni model for diverse e-commerce scenarios. Then, we further improve Valley3 through post-training to encourage long-chain reasoning with controllable reasoning modes, enabling one non-thinking mode and three distinct levels of thinking, thereby balancing inference efficiency in simple scenarios with deep reasoning for complex applications. Moreover, we equip Valley3 with agentic search capabilities to proactively invoke search tools and acquire task-relevant information for e-commerce deep research tasks. To comprehensively assess the capabilities of Valley3, we construct an omni e-commerce benchmark spanning 6 tasks. Experimental results show that Valley3 consistently outperforms strong baselines on our in-house and open-source e-commerce benchmarks, while remaining competitive on general-domain benchmarks. Our code and model weights are available at https://github.com/bytedance/Valley.

Graph-based techniques and spectral graph theory have enriched the field of machine learning with a variety of critical advances. A central object in the analysis is the graph Laplacian L, which encodes the structure of the graph. We consider the problem of learning over this Laplacian in a distributed streaming setting, where new edges of the graph are observed in real time by a network of workers. In this setting, it is hard to learn quickly or approximately while keeping a distributed representation of L. To address this challenge, we present a novel algorithm, GSQUEAK, which efficiently sparsifies the Laplacian by maintaining a small subset of effective resistances. We show that our algorithm produces sparsifiers with strong spectral approximation guarantees, all while processing edges in a single pass and in a distributed fashion.

The Platonic Representation Hypothesis suggests that neural networks trained on different modalities (e.g., text and images) align and eventually converge toward the same representation of reality. If true, this has significant implications for whether modality choice matters at all. We show that the experimental evidence for this hypothesis is fragile and depends critically on the evaluation regime. Alignment is measured using mutual nearest neighbors on small datasets ($\approx$1K samples) and degrades substantially as the dataset is scaled to millions of samples. The same behavior is observed beyond text-image, for text-audio and text-video alignment. The alignment that remains between model representations reflects coarse semantic overlap rather than consistent fine-grained structure. Moreover, the evaluations in Huh et al. are done in a one-to-one image-caption setting, a constraint that breaks down in realistic many-to-many settings and further reduces measured alignment. We also find that the reported trend of stronger language models increasingly aligning with vision does not appear to hold for newer models. Overall, our findings suggest that the current evidence for cross-modal representational convergence is considerably weaker than subsequent works have taken it to be. Models trained on different modalities may learn equally rich representations of the world, just not the same one.

Camouflaged object detection (COD) segments objects that intentionally blend with the background, so predictions depend on subtle texture and boundary cues. COD is often needed under tight on-device memory and latency budgets, making low-bit inference highly desirable. However, COD is unusually hard to quantize aggressively. We study post-training W4A4 quantization of Transformer-based COD and find a task-specific cliff: heavy-tailed background tokens dominate a shared activation range, inflating the step size and pushing weak-but-structured boundary cues into the zero bin. This exposes a token-local bottleneck -- remove cross-token range domination and bound the zero-bin mass under 4-bit activations. To address this, we introduce COD-TDQ, a COD-aware Token-group Dual-constraint activation Quantization method. COD-TDQ addresses this token-local bottleneck with two coupled steps: Direct-Sum Token-Group (DSTG) assigns token-group scales to suppress cross-token range domination, and Dual-Constraint Range Projection (DCRP) projects each token-group clip range to keep the step-to-dispersion ratio and the zero-bin mass bounded. Across four COD benchmarks and two baseline models (CFRN and ESCNet), COD-TDQ consistently achieves an $S_α$ score more than 0.12 higher than that of the state-of-the-art quantization method without retraining. The code is available at https://github.com/MCG-NKU/nku-model-compre.

Consumer LiDARs in mobile devices and robots typically output a single depth value per pixel. Yet internally, they record full time-resolved histograms containing direct and multi-bounce light returns; these multi-bounce returns encode rich non-line-of-sight (NLOS) cues that can enable perception of hidden objects in a scene. However, severe hardware limitations of consumer LiDARs make NLOS reconstruction with conventional methods difficult. In this work, we motivate a complementary direction: enabling NLOS perception with low-cost LiDARs through data-driven inference. We present DENALI, the first large-scale real-world dataset of space-time histograms from low-cost LiDARs capturing hidden objects. We capture time-resolved LiDAR histograms for 72,000 hidden-object scenes across diverse object shapes, positions, lighting conditions, and spatial resolutions. Using our dataset, we show that consumer LiDARs can enable accurate, data-driven NLOS perception. We further identify key scene and modeling factors that limit performance, as well as simulation-fidelity gaps that hinder current sim-to-real transfer, motivating future work toward scalable NLOS vision with consumer LiDARs.

Three-dimensional (3D) fingerprint recognition and identification offer several advantages over traditional two-dimensional (2D) recognition systems. The contactless nature of 3D fingerprints enhances hygiene and security, reducing the risk of contamination and spoofing. In addition to surface ridge and valley patterns, 3D fingerprints capture depth, curvature, and shape information, enabling the development of more precise and robust authentication systems. Despite recent advancements, significant challenges remain. The topological height of fingerprint pixels complicates the extraction of ridge and valley patterns. Furthermore, registration issues limit the acquisition process, requiring consistent direction and orientation across all samples. To address these challenges, this paper introduces a method that unwraps 3D fingerprints, represented as 3D point clouds, using B-spline curve fitting to mitigate height variation and reduce registration limitations. The unwrapped point cloud is then converted into a grayscale image by mapping the relative heights of the points. This grayscale image is subsequently used for recognition through conventional 2D fingerprint identification methods. The proposed approach demonstrated superior performance in 3D fingerprint recognition, achieving Equal Error Rates (EERs) of 0.2072%, 0.26%, and 0.22% across three experiments, outperforming existing methods. Additionally, the method surpassed 3D fingerprint flattening technique in both recognition and identification during cross-session experiments, achieving an EER of 1.50% when fingerprints with varying registrations were included.

Autonomous robots operating in natural karstic caves face perception and navigation challenges that are qualitatively distinct from those encountered in mines or tunnels: irregular geometry, reflective wet surfaces, near-zero ambient light, and complex branching passages. Yet publicly available datasets targeting this environment remain scarce and offer limited sensing modalities and environmental diversity. We present CAVERS, a multimodal dataset acquired in two structurally distinct rooms of Cueva de la Victoria, Málaga, Spain, comprising 24 sequences totaling approximately 335 GB of recorded data. The sensor suite combines an Intel RealSense D435i RGB-D-I camera, an Optris PI640i near-IR thermal camera, and a Velodyne VLP-16 LiDAR, operated both handheld and mounted on a wheeled rover under full darkness and artificial illumination. For most of the sequences, mm-accurate 6-DoF ground truth pose and velocity at 120 Hz are provided by an Optirack motion capture system installed directly inside the cave. We benchmark seven state-of-the-art SLAM and odometry algorithms spanning visual, visual-inertial, thermal-inertial, and LiDAR-based pipelines, as well as a 3D reconstruction pipeline, demonstrating the dataset's usability. %The dataset and all supplementary material are publicly available at: https://github.com/spaceuma/cavers.

We present THEIA, a 2.75M-parameter modular neural architecture that learns the complete Kleene three-valued logic (K3) truth table from task data without external symbolic inference or hand-encoded K3 gate primitives. Across 5 seeds it passes all 39 K3 rules at >99% per-rule accuracy. K3 learnability is not the central finding: Transformer baselines also pass all 39 rules, and flat MLPs match THEIA on Phase-1 accuracy within 0.04pp. The contributions are two properties of the learned system. (1) Uncertainty-verdict asymmetric propagation. THEIA preserves Has-Unknown at every upstream boundary (80.0/91.1/90.8/99.7% across Arith/Order/Set/Logic vs. ~52% majority) while final-verdict decodability stays at or below a 73.4% U-vs-non-U oracle reference under linear and nonlinear probes. Activation patching on non-absorbent T->U cases flips 4,898/4,898 OR and 4,719/4,719 AND pairs across 5 seeds, ruling out residual shortcuts. (2) Reliability spectrum under discretized end-to-end training, on tasks decomposable along the engine boundaries. A mod-3 sequential composition task generalizes from 5- to 500-step evaluation at 99.96+-0.04% (5 seeds). Under identical Gumbel-softmax training, flat MLPs collapse to chance by 50 steps; a 2x2 ResMLP grid reaches >=99% on only 3/20 (config, seed) trials; a pre-LN Transformer reaches 99.24+-0.34%. Straight-through discretization prevents 0.999^500 compounding; the architectural separator is sustaining Phase-1 accuracy under Phase-3 training, where flat MLPs fail. Auxiliary: under per-architecture development defaults (not optimizer-controlled), THEIA reaches 12/12 Kleene coverage 6.5x faster than a parameter-comparable 8L Transformer; this narrows to ~3.6x with Transformer-standard tuning and 4.93x with the same recipe on both. Ratios are config-specific, not asymptotic.

LLM-based agents are increasingly capable of complex task execution, yet current agentic systems remain constrained by text-centric paradigms that struggle with long-horizon tasks due to fragile multi-turn dependencies and context drift. We present CaveAgent, a framework that shifts tool use from LLM-as-Text-Generator'' to LLM-as-Runtime-Operator.'' CaveAgent introduces a dual-stream architecture that inverts the conventional paradigm: rather than treating the LLM's text context as the primary workspace with tools as auxiliary, CaveAgent elevates the persistent Python runtime as the central locus of state, with a lightweight semantic stream serving as its orchestrator. Beyond leveraging code generation to resolve interdependent sub-tasks (e.g., loops, conditionals) in a single step, CaveAgent introduces \textit{Stateful Runtime Management}: it injects, manipulates, and retrieves complex Python objects (e.g., DataFrames, database connections) that persist across turns, unlike existing code-based approaches that remain text-bound. CaveAgent further provides a runtime-integrated skill management system that extends the Agent Skills open standard, enabling ecosystem interoperability through executable skill injections. This persistence mechanism serves as a high-fidelity external memory that reduces context drift in multi-turn interactions and preserves processed data for downstream applications without information loss. Evaluations show consistent improvement across challenging benchmarks, enabling CaveAgent to handle data scales that cause context overflow in both JSON-based and code-based agents. The accessible runtime state further provides programmatically verifiable feedback, enabling automated evaluation and reward signal generation without human annotation and establishing a structural foundation for future research in Reinforcement Learning with Verifiable Rewards (RLVR).