Daily Papers

Language model agents are increasingly effective in solving realistic tasks through multi-turn tool use. However, training reliable tool-using agents remains challenging in practice. While reinforcement learning provides an on-policy paradigm for improving agents from their own environment interactions, its effectiveness depends heavily on the training task distribution. When tasks are fixed before training, the task distribution can become increasingly mismatched with the policy's evolving capabilities, causing many rollouts to be spent on uninformative tasks. We propose SENTINEL, a failure-driven reinforcement learning framework that turns the Solver's rollout failures into targeted training tasks. SENTINEL follows a Controller--Proposer--Solver loop: the Controller analyzes failed trajectories and summarizes recurring error patterns, the Proposer generates executable tasks that stress these weaknesses, and the Solver is trained on the targeted tasks. On Tau2-Bench Retail with Qwen3-4B-Thinking-2507, SENTINEL improves Pass\^{}1 from 66.4 to 74.9 and outperforms RL on general synthetic tasks across Pass\^{}k metrics. These results demonstrate that model failures provide an effective and scalable source of targeted training signal for improving tool-using language model agents.

Reinforcement Learning from Verifiable Rewards (RLVR) has emerged as a powerful paradigm for enhancing the complex reasoning capabilities of Large Reasoning Models. However, standard outcome-based supervision suffers from a critical limitation that penalizes trajectories that are largely correct but fail due to several missteps as heavily as completely erroneous ones. This coarse feedback signal causes the model to discard valuable largely correct rollouts, leading to a degradation in rollout diversity that prematurely narrows the exploration space. Process Reward Models have demonstrated efficacy in providing reliable step-wise verification for test-time scaling, naively integrating these signals into RLVR as dense rewards proves ineffective.Prior methods attempt to introduce off-policy guided whole-trajectory replacement that often outside the policy model's distribution, but still fail to utilize the largely correct rollouts generated by the model itself and thus do not effectively mitigate the narrowing of the exploration space. To address these issues, we propose SCOPE (Step-wise Correction for On-Policy Exploration), a novel framework that utilizes Process Reward Models to pinpoint the first erroneous step in suboptimal rollouts and applies fine-grained, step-wise off-policy rectification. By applying precise refinement on partially correct rollout, our method effectively salvages partially correct trajectories and increases diversity score by 13.5%, thereby sustaining a broad exploration space. Extensive experiments demonstrate that our approach establishes new state-of-the-art results, achieving an average accuracy of 46.6% on math reasoning and exhibiting robust generalization with 53.4% accuracy on out-of-distribution reasoning tasks.

Reinforcement learning (RL) is essential for enhancing the complex reasoning capabilities of large language models (LLMs). However, existing RL training pipelines are computationally inefficient and resource-intensive, with the rollout phase accounting for over 70% of total training time. Quantized RL training, particularly using FP8 precision, offers a promising approach to mitigating this bottleneck. A commonly adopted strategy applies FP8 precision during rollout while retaining BF16 precision for training. In this work, we present the first comprehensive study of FP8 RL training and demonstrate that the widely used BF16-training + FP8-rollout strategy suffers from severe training instability and catastrophic accuracy collapse under long-horizon rollouts and challenging tasks. Our analysis shows that these failures stem from the off-policy nature of the approach, which introduces substantial numerical mismatch between training and inference. Motivated by these observations, we propose Jet-RL, an FP8 RL training framework that enables robust and stable RL optimization. The key idea is to adopt a unified FP8 precision flow for both training and rollout, thereby minimizing numerical discrepancies and eliminating the need for inefficient inter-step calibration. Extensive experiments validate the effectiveness of Jet-RL: our method achieves up to 33% speedup in the rollout phase, up to 41% speedup in the training phase, and a 16% end-to-end speedup over BF16 training, while maintaining stable convergence across all settings and incurring negligible accuracy degradation.

Computer-use agents (CUAs) interact with graphical interfaces through screenshots and low-level mouse and keyboard actions, yet the causal error may precede the terminal failure. We present CUADebug, a framework for localizing root causes in CUA trajectories and guiding re-execution. CUADebug includes a five-category, 30-subtype taxonomy; CUAErrorBench, a benchmark of 204 failed OSWorld trajectories with human root-cause annotations; and CUADebugger, a ReAct-style agent for root-cause analysis (RCA). CUADebugger iteratively selects trajectory steps, inspects paired before/after screenshots and action traces, and submits a structured diagnosis containing the causal step, taxonomy label, grounded evidence, and correction. CUADebugger performs RCA without per-trajectory human intervention; human annotations are used to evaluate RCA predictions and, in controlled re-rollout comparisons, to fix restart points. Task reasoning and control is the largest annotated failure category (110/204). CUADebugger improves L2 and Tag+Step Exact across three debugger backbones on the Claude-agent split; with Gemini 2.5 Pro, Tag+Step Exact rises from 11.1% to 19.4%. Single re-execution improves failure recovery from 13.89% to 29.86% (overall 61.77% to 68.14%); controlled continual re-execution improves it from 12.50% to 25.69% (overall 61.22% to 66.48%). Project page: cuadebug.github.io.

While vision-language-action (VLA) models have advanced rapidly, running them at every control step remains expensive. Prior work accelerates VLA inference using techniques like action chunking and visual-token pruning, typically evaluating based on latency and average task success. However, acceleration may discard information and break tasks the original policy would solve, a risk hidden by average metrics. Measuring these failures is challenging because action deviations compound over closed-loop trajectories, meaning task failure is only observable across full episodes. We therefore define an acceleration-induced failure via paired rollouts from identical initial conditions, tracking when the reference succeeds but the accelerated policy fails. To manage this, we introduce CARE, an approach for certified accelerator selection. CARE uses paired rollouts on a calibration set to provide finite-sample guarantees that acceleration-induced failure risk stays below a user-specified budget. It deploys the fastest certified candidate, falling back to the reference if none qualify. By relying only on terminal outcomes and measured compute, CARE applies unchanged across diverse acceleration mechanisms, while sequential testing and failure-triggered reference rollouts keep certification affordable. On four LIBERO suites with OpenVLA-OFT, CARE certifies 9.0--10.8times speedups while guaranteeing (at 95% confidence) that at least 85.8% of reference-solved episodes are preserved. Under tight budgets, selectors without guarantees exceed the budget in up to 75% of trials, whereas CARE stays within budget and its sequential form uses 78.9% fewer rollouts than exhaustive evaluation. CARE further generalizes to flow-step reduction for π_{0.5}, and to Qwen3.5-9B and Llama-3.1-8B agents in Crafter.

Autoregressive (AR) video diffusion models enable low-latency streaming generation by synthesizing videos chunk by chunk with cached visual context, but this chunk-wise formulation makes temporal instruction following ambiguous. A single global prompt does not specify which sub-event should be realized in each chunk, while naively switching to step-wise prompts often leads to delayed reactions, blended step semantics, and error propagation across prompt transitions. These failures are difficult to address with supervised fine-tuning or distillation alone: SFT suffers from exposure bias, while rollout-based distillation still optimizes low-level denoising or teacher-distribution matching rather than directly enforcing action ordering and prompt-transition correctness. We address these challenges with TempAct, a planner--executor reinforcement learning framework that jointly optimizes temporal decomposition and step-conditioned execution for temporally plausible AR video generation. TempAct uses an LLM planner to explore span-aware step prompts that are executable by the video model, and trains an AR diffusion executor to follow these prompts under its own generated histories. Its key mechanism is hierarchical group exploration: candidate plans form planning groups, and each plan induces an execution group of multiple continuations from a shared visual context, enabling plan-level credit assignment for long-horizon temporal outcomes and executor-level credit assignment for prompt-switch behavior. We further design hierarchical rewards that combine plan-quality and full-video temporal feedback for the planner with local transition-level step-following rewards, aesthetic regularization, and KL constraints for the executor. Experiments on Self-Forcing and LongLive show that TempAct improves temporal consistency while preserving overall visual quality.

Large language models (LLMs) are increasingly deployed as customer-facing agents, yet evaluating their reliability remains challenging due to stochastic, multi-turn interactions. Current evaluation protocols rely on linear Monte Carlo rollouts of complete agent-user conversations to estimate success. However, this approach is computationally inefficient, repeatedly regenerating identical early prefixes, and often fails to uncover deep failure modes that arise from rare user behaviors. We introduce DIVERT (Diversity-Induced Evaluation via Branching of Trajectories), an efficient, snapshot-based, coverage-guided user simulation framework for systematic exploration of agent-user interactions. DIVERT captures the full agent-environment state at critical decision points and resumes execution from these snapshots, enabling reuse of shared conversation prefixes and reducing redundant computation. From each junction, the framework branches using targeted, diversity-inducing user responses, allowing directed exploration of alternative interaction paths. By focusing evaluation on semantically diverse and underexplored trajectories, DIVERT improves both efficiency and coverage. Empirical results show that it discovers more failures per token compared to standard linear rollout protocols, while expanding the set of tasks on which failures are identified.

Recommendation algorithm iteration is moving from an artisanal, engineer-bound process toward an industrialized research loop, but this transition remains blocked by a structural execution bottleneck: the idea-to-launch cycle still depends on human engineers to generate hypotheses, modify production code, launch A/B experiments, and attribute online results. Innovation therefore scales linearly with headcount rather than compounding with evidence, compute, and accumulated experimental knowledge. We present AgentX, a production-deployed multi-agent system that fundamentally restructures this production function. AgentX operates as a self-evolving development engine: it autonomously generates, implements, evaluates, and learns from recommendation experiments at a scale and pace that no manual workflow can sustain. The system orchestrates four tightly coupled stages in a closed loop. A Brainstorm Agent synthesizes evidence from historical experiments, system architecture, data analysis, and external research into ranked, executable proposals. A Developing Agent translates each proposal into production-ready code through repository-grounded generation and multi-dimensional reliability verification. An Evaluation Agent conducts safe online rollout with guardrail-vetoed A/B judgment, converting both successes and failures into structured knowledge assets. A Harness Evolution layer (SGPO) then distills execution trajectories into semantic-gradient updates that continuously sharpen the agents themselves -- making the system not merely automated, but self-improving.

Agent harness evolution improves frozen language-model agents by modifying the executable structures around them. We study this paradigm as a form of sample-efficient fast adaptation: instead of updating model weights, an agent can acquire task-specific competence by changing its external harness, while leaving the base model's general capabilities intact. Prior work shows that self-generated rollouts can support harness search, suggesting that agents may acquire new task competence through practice. Yet in long-horizon stochastic environments, self-practice becomes fragile: rewards are sparse, outcomes are high-variance, and failures are hard to attribute to concrete harness mechanisms. We introduce DemoEvolve, a demonstration-bootstrapped approach to harness evolution. When reward-only search is too broad and noisy, competent human trajectories serve as expert reference experience for the coding proposer, guiding harness-level diagnosis and editing. Experiments on Liar's Dice show that self-rollout evolution can work when episodes are short and failures are attributable. In contrast, Balatro exposes a harder long-horizon stochastic regime, where self-rollout evolution is misled by sparse feedback and candidate-selection noise, while tutorial-like textual knowledge alone does not yield stable improvement. Under the same limited budget, DemoEvolve produces more effective and auditable harness edits and achieves better performance. Overall, demonstrations make sparse-feedback harness evolution more diagnosable, localizable, and stable.

An unsuccessful LLM agent rollout contains more information than its final reward: the observations available to the agent, the actions it chose, and the environment's responses. Reusing this experience for learning requires identifying a decision to revise and testing a concrete alternative. We introduce the Agent Error Dataset (AED), comprising 50,228 error-diagnosis pairs from 9,961 source tasks across 33 environments, 19 harness families, and 23 policy models in text-based agent systems. We retain source traces and execution metadata to support cross-setting failure analysis and re-diagnosis without repeating the original rollout. Our five-stage Agentic Error-to-Training (AET) pipeline collects natural failures, generates diagnoses and proposed corrections, and checks them against recorded evidence. Where replay is supported, we compare corrections with original-action retries from the same checkpoint under matched execution settings. We then construct separate training views for diagnosis and actor recovery. Across 3,062 matched replay pairs, first-proposal corrections raise verifier pass rates from 18.4% to 51.1%, a gain of 32.7 percentage points. Using a separately frozen diagnosis release, full-diagnosis fine-tuning on 1,656 source tasks raises Qwen3-8B's exact-step agreement with internal teacher labels from 47.2% to 63.6%, averaged over three seeds on a 943-case holdout. The strongest prompted reference in this comparison scores 54.7%, and mean agreement improves at each of four increasing training-set sizes. In a single-seed comparison of actor-training recipes, action-only repair training scores 6.67 percentage points higher on WebShop-lite than success-only training.

In long-horizon LLM agent reinforcement learning, weak policies often repeat similar failures, producing uninformative rollout trajectories and limiting effective policy optimization. Existing skill-centric methods improve exploration by optimizing, filtering, or internalizing reusable skills. However, they remain centered on the skills themselves rather than being designed as adaptive training-time support for the evolving policy. To address this, we propose a policy-centric training paradigm that reframes skills as a dynamic training scaffold. Our framework, Pats, converts rollout groups from the latest policy into evidence cards and uses task-specific evaluation to adjust the context used in subsequent rollouts. Concrete guidance helps weak policies to complete challenging tasks. As policy improves, redundant context is revised or removed to reduce reliance on explicit guidance while preserving useful rollout variation. The policy is optimized with environmental rewards using standard RLVR, and the training scaffold is discarded at deployment. On ALFWorld and WebShop, Pats improves over strong baselines by up to 18.6%. Across seven search-augmented QA benchmarks, it remains competitive while using 32.1% fewer prompt tokens than the baseline.

Reinforcement learning (RL) is increasingly used to post-train vision-language-action (VLA) models, but every update consumes robot rollouts that are slow and costly to collect, making sample efficiency a central concern. Manipulation tasks typically provide only sparse rewards, so a weak policy fails almost every rollout early in training and has little to learn from, even when those failures execute coherent behavior. Such a failure, however, is a success at a different task. We present Learning from Hindsight (LfH), which brings hindsight relabeling to RL post-training of VLAs by scoring failed rollouts against the tasks they actually achieved. A single vision-language model relabels both the instruction and the reward, proposing a hindsight instruction for a group of failed rollouts and scoring how well each satisfies it, and the policy trains on the relabeled and original rollouts jointly. Because VLAs generalize across language, relabeling in language lets the policy learn more from the same trajectories. On out-of-distribution LIBERO-PRO tasks, where standard RL improves only slowly, LfH achieves 5times improvement in sample efficiency, and outperforms a dense progress-reward baseline. The gains hold across VLA backbones and on a physical Franka robot.

Reinforcement learning is a natural way to post-train LLM agents for long-horizon interactive tasks judged only by end-of-task verification, yet a shared belief holds that outcome-only RL soon hits a ceiling on small open models. Recent work therefore compensates around the training with denser rewards, SFT priors, skill libraries, curated memory, or multi-agent orchestration. We argue the ceiling is an artifact of two failures of common practice. Signal starvation: group-relative RL with sparse outcome-only rewards yields a gradient only when a task's rollout group mixes successes and failures, so under-scaled exploration silences exactly the hardest, most instructive tasks. Policy drift: squeezing many updates out of a small task pool degrades the policy itself, as an unanchored objective lets the sampling distribution collapse exactly when saturation has already made informative groups rare. We present CANOPY (Coverage-ANchored On-PolicY RL), a minimalist protocol attacking both directly: scale same-task exploration until the natural signal reappears, keep every update on-policy, KL-anchored, and confined to the agent's own action tokens, then cash in an enlarged interaction budget at test time. On AppWorld, a long-horizon interactive coding benchmark, a Qwen3-14B policy trained with CANOPY through environment interaction alone--without task-specific supervision, auxiliary credit signals, or elaborate agent scaffolding--topped the public leaderboard (Feb. 2026; Test-Normal TGC 86.9, Test-Challenge 67.6), and the same design principles lift Qwen3.5-9B on SWE-bench Verified by 16.6 points. Agentic RL alone thus internalizes long-horizon capability directly into a small open model; we plan to release the complete training stack at https://github.com/AlibabaResearch/SignalCoverageRL.

Agentic reinforcement learning (RL) enables LLM agents to improve continuously from environment rewards, yet the resulting policies do not systematically accumulate reusable strategies that generalize across tasks. Modular skills can provide such reusable strategies, yet existing skill-augmented RL methods decouple skill creation from policy optimization, risking adopting skills that conflict with the evolving policy. Inspired by Anthropic's Skill Creator, we introduce ReSkill, an RL-in-the-loop skill creation framework that reconciles skill evolution with policy learning. ReSkill exploits the group-wise structure of GRPO to naturally embed three mechanisms with only marginal additional overhead: (1) an assertion-driven skill creator that diagnoses failures from past experience and proposes conditional, trigger-based skill revisions; (2) within-group rollout sampling that enables controlled comparison of skill versions, capturing which version best supports the policy's ongoing learning; and (3) Thompson Sampling with adaptive discounting to balance exploration and exploitation in skill version selection as the policy evolves. Across several domains, ReSkill consistently outperforms existing memory and skill-based RL methods, with the largest gains on unseen tasks. Analysis of the skill lifecycle shows skills being automatically created, tested, refined, and pruned as the policy improves, demonstrating reconciled skill-policy co-evolution.

LLM agents are predominantly benchmarked on short, single-task trajectories, yet real deployments run for hours under contention, surfacing a different class of failures. We use the Clinical Environment Simulator (CES), in which an agent manages an entire emergency-department shift under continuous time and resource pressure, as a testbed: long-horizon execution failures manifest measurably in a single rollout under structured, multi-dimensional grading. On CES, current agents reach the correct diagnosis in most cases yet fail to deliver complete and timely critical actions, revealing an execution gap. We attribute this gap to three long-horizon failure modes, each operationalized as a per-trace counter: instruction-adherence drift, treatment incompleteness, and a severity-equity gap in timeliness. We then introduce Asclepius, an adaptive agent scaffolding with a self-evolving harness that rewrites the operating manual between shifts from trace-level feedback, an externalized clinical skills library for high-stakes regimen knowledge, and three isolated subagents that partition per-turn decisions across the patient queue. On held-out batches never observed during harness evolution, Asclepius improves critical-action correctness by 22% (p = 0.024) over a strong baseline agent framework while preserving diagnostic accuracy, with consistent gains across five LLM judges from three model families; on the full ten-batch set, improvements reach 25% on critical actions and 13% on timeliness. The three failure modes form a coupled bottleneck: decisive reductions appear only when all three components act together.

Long-horizon agentic RL is constrained by the futures visited by the current policy. When rollouts fail to reach useful behaviors, improved credit assignment cannot provide supervision for states and actions that were never encountered. On-policy distillation (OPD) introduces teacher guidance on policy-induced states, and recent multi-turn variants improve how this guidance is scheduled, replayed, refined, or validated. However, they do not directly use the oracle-controlled future itself as the training trajectory. We introduce State-Rooted Executable Rollout Optimization (SERO), which augments on-policy agentic RL with executable oracle continuations. SERO restores a state actually reached by the current policy and lets a training-only oracle continue interacting with the real environment, making the downstream observations and oracle actions themselves available for learning. It then applies policy-reachability weighting, which weights each oracle step by the frozen behavior policy's compatibility with the preceding oracle actions. This allows deeper supervision to become effective as the policy becomes more compatible with the oracle branch. The oracle loss is optimized separately from the base RL estimator and no oracle is used at inference time. We evaluate SERO on two challenging interactive-agent benchmarks, ALFWorld and WebShop, using Qwen2.5-1.5B-Instruct and Qwen2.5-7B-Instruct. On ALFWorld, learning the full executable continuation is substantially more effective than supervising only the root oracle action, while policy-reachability weighting further improves out-of-distribution performance.

Multi-turn agents solve complex tasks through extended sequences of tool interactions before producing a final answer, making credit assignment a fundamental challenge during post-training. Outcome rewards provide reliable supervision for short-horizon reasoning, but become sparse and high-variance as trajectories grow to tens or hundreds of tool calls. They can also be misleading: a failed rollout may contain many useful actions that move the agent closer to the goal, yet outcome-only training assigns them the same negative advantage as the eventual mistake. We propose TRACE (Turn-level Reward Assignment via Credit Estimation), a dense credit-assignment method for agentic reinforcement learning. TRACE represents rollouts as state transitions at tool-call boundaries, obtains gold-answer log-probabilities from a frozen reference model, transforms them into log-ratio state values, and derives per-action rewards as Temporal-Difference changes in those values. This requires no additional critic or process-label training, and its one-step log-ratio TD component telescopes across redundant tool calls. On long-horizon complex search, TRACE substantially improves base-model tool-use ability using pure RL, without a cold-start supervised fine-tuning stage, an agentic mid-training stage, or training on live-web data. On the closed-web BrowseComp-Plus benchmark, it raises Qwen3-4B from 7.2 to 35.6 and Qwen3-30B-A3B from 8.4 to 42.6. The learned search behavior also transfers to open-web benchmarks, and the learning curves show earlier improvement and faster convergence during RL training.

Large language model (LLM) reinforcement learning (RL) pipelines are often bottlenecked by rollout generation, making end-to-end training slow. Recent work mitigates this by running rollouts with quantization to accelerate decoding, which is the most expensive stage of the RL loop. However, these setups destabilize optimization by amplifying the training-inference gap: rollouts are operated at low precision, while learning updates are computed at full precision. To address this challenge, we propose QaRL (Rollout Alignment Quantization-Aware RL), which aligns training-side forward with the quantized rollout to minimize mismatch. We further identify a failure mode in quantized rollouts: long-form responses tend to produce repetitive, garbled tokens (error tokens). To mitigate these problems, we introduce TBPO (Trust-Band Policy Optimization), a sequence-level objective with dual clipping for negative samples, aimed at keeping updates within the trust region. On Qwen3-30B-A3B MoE for math problems, QaRL outperforms quantized-rollout training by +5.5 while improving stability and preserving low-bit throughput benefits.

We introduce ROLL, an efficient, scalable, and user-friendly library designed for Reinforcement Learning Optimization for Large-scale Learning. ROLL caters to three primary user groups: tech pioneers aiming for cost-effective, fault-tolerant large-scale training, developers requiring flexible control over training workflows, and researchers seeking agile experimentation. ROLL is built upon several key modules to serve these user groups effectively. First, a single-controller architecture combined with an abstraction of the parallel worker simplifies the development of the training pipeline. Second, the parallel strategy and data transfer modules enable efficient and scalable training. Third, the rollout scheduler offers fine-grained management of each sample's lifecycle during the rollout stage. Fourth, the environment worker and reward worker support rapid and flexible experimentation with agentic RL algorithms and reward designs. Finally, AutoDeviceMapping allows users to assign resources to different models flexibly across various stages.

The continuous delivery of modern software requires the execution of many automated pipeline jobs. These jobs ensure the frequent release of new software versions while detecting code problems at an early stage. For TELUS, our industrial partner in the telecommunications field, reliable job execution is crucial to minimize wasted time and streamline Continuous Deployment (CD). In this context, flaky job failures are one of the main issues hindering CD. Prior studies proposed techniques based on machine learning to automate the detection of flaky jobs. While valuable, these solutions are insufficient to address the waste associated with the diagnosis of flaky failures, which remain largely unexplored due to the wide range of underlying causes. This study examines 4,511 flaky job failures at TELUS to identify the different categories of flaky failures that we prioritize based on Recency, Frequency, and Monetary (RFM) measures. We identified 46 flaky failure categories that we analyzed using clustering and RFM measures to determine 14 priority categories for future automated diagnosis and repair research. Our findings also provide valuable insights into the evolution and impact of these categories. The identification and prioritization of flaky failure categories using RFM analysis introduce a novel approach that can be used in other contexts.

Reinforcement Learning with Verifiable Rewards (RLVR) methods such as GRPO rely on successful self-generated trajectories, but finite rollout budgets can produce all-fail groups with no reward-based policy-gradient signal. While additional rollouts improve the chance of success at higher cost, successful trajectories missing from one model's rollouts may already have been discovered by another. Indeed, we observe that heterogeneous models often succeed on complementary prompts, creating opportunities for mutual learning without a designated stronger teacher. To exploit this complementarity, we propose GRAFT (Gated Replacement of Answer-Failed groups with peer Trajectories), an off-policy-aware framework that replaces all-fail groups with informative peer groups. GRAFT transfers both successful and unsuccessful peer responses with peer-computed advantages, while controlling cross-model mismatch through sequence-level compatibility weighting and token-level importance ratio clipping. Across three heterogeneous model pairs and five mathematical reasoning benchmarks, GRAFT consistently improves both models over GRPO with the same per-model rollout budget, gaining 2.1 points on average and up to 4.5 points in model-level average performance. Stored peer trajectories preserve most of the gains, improving over GRPO by 1.8 points on average without simultaneous co-training.

Rollout generation is a major bottleneck in Reinforcement Learning (RL) for Mixture-of-Experts (MoE) Large Language Models, motivating low-precision rollout acceleration such as FP8. As an emerging low-precision format, NVFP4 combines fine-grained scaling for accuracy preservation with native W4A4 FP4 GEMMs for higher throughput than FP8. However, we find that directly applying NVFP4 to MoE RL rollout is impractical. NVFP4 rollout with BF16 training collapses after roughly 150 steps, accompanied by rapidly growing rollout-trainer log-probability gaps. Through training-inference error analysis and controlled ablations, we identify activation error, rather than weight error, as the dominant source of FP4 RL instability: weights can be synchronized and aligned by a shared quantization-dequantization path, whereas activations are recomputed online and error is amplified by the coarse E2M1 grid. Therefore, to stabilize NVFP4 RL for MoE, we propose QUantization-error Alignment across Dual Sides (QUADS). On the trainer side, we introduce Asymmetric Quantization-Aware Training fake-quantizing weights while keeping activations unquantized for better alignment. On the rollout side, Residual Activation Compensation corrects high-error activation channels while preserving native W4A4 GEMMs. In our MoE RL experiments on several benchmarks, QUADS achieves BF16-level accuracy, improves average pass@1 by 21.49 points over naive NVFP4 RL, and delivers ~16% higher rollout throughput than FP8.

Despite the success of reinforcement learning for large language models, a common failure mode is reduced sampling diversity, where the policy repeatedly generates similar erroneous behaviors. Classical entropy regularization encourages randomness under the current policy, but does not explicitly discourage recurrent failure patterns across rollouts. We propose MEDS, a Memory-Enhanced Dynamic reward Shaping framework that incorporates historical behavioral signals into reward design. By storing and leveraging intermediate model representations, we capture features of past rollouts and use density-based clustering to identify frequently recurring error patterns. Rollouts assigned to more prevalent error clusters are penalized more heavily, encouraging broader exploration while reducing repeated mistakes. Across five datasets and three base models, MEDS consistently improves average performance over existing baselines, achieving gains of up to 4.13 pass@1 points and 4.37 pass@128 points. Additional analyses using both LLM-based annotations and quantitative diversity metrics show that MEDS increases behavioral diversity during sampling.

Reinforcement learning (RL) has become a central post-training tool for improving the reasoning abilities of large language models (LLMs). In these systems, the rollout, the trajectory sampled from a prompt to termination, including intermediate reasoning steps and optional tool or environment interactions, determines the data the optimizer learns from, yet rollout design is often underreported. This survey provides an optimizer-agnostic view of rollout strategies for RL-based post-training of reasoning LLMs. We formalize rollout pipelines with unified notation and introduce Generate-Filter-Control-Replay (GFCR), a lifecycle taxonomy that decomposes rollout pipelines into four modular stages: Generate proposes candidate trajectories and topologies; Filter constructs intermediate signals via verifiers, judges, critics; Control allocates compute and makes continuation/branching/stopping decisions under budgets; and Replay retains and reuses artifacts across rollouts without weight updates, including self-evolving curricula that autonomously generate new training tasks. We complement GFCR with a criterion taxonomy of reliability, coverage, and cost sensitivity that characterizes rollout trade-offs. Using this framework, we synthesize methods spanning RL with verifiable rewards, process supervision, judge-based gating, guided and tree/segment rollouts, adaptive compute allocation, early-exit and partial rollouts, throughput optimization, and replay/recomposition for self-improvement. We ground the framework with case studies in math, code/SQL, multimodal reasoning, tool-using agents, and agentic skill benchmarks that evaluate skill induction, reuse, and cross-task transfer. Finally, we provide a diagnostic index that maps common rollout pathologies to GFCR modules and mitigation levers, alongside open challenges for building reproducible, compute-efficient, and trustworthy rollout pipelines.

One of the main challenges developers face in the use of continuous integration (CI) and deployment pipelines is the occurrence of intermittent job failures, which result from unexpected non-deterministic issues (e.g., flaky tests or infrastructure problems) rather than regular code-related errors such as bugs. Prior studies developed machine learning (ML) models trained on large datasets of job logs to classify job failures as either intermittent or regular. As an alternative to costly manual labeling of large datasets, the state-of-the-art (SOTA) approach leveraged a heuristic based on non-deterministic job reruns. However, this method mislabels intermittent job failures as regular in contexts where rerunning suspicious job failures is not an explicit policy, and therefore limits the SOTA's performance in practice. In fact, our manual analysis of 2,125 job failures from 5 industrial and 1 open-source projects reveals that, on average, 32% of intermittent job failures are mislabeled as regular. To address these limitations, this paper introduces a novel approach to intermittent job failure detection using few-shot learning (FSL). Specifically, we fine-tune a small language model using a few number of manually labeled log examples to generate rich embeddings, which are then used to train an ML classifier. Our FSL-based approach achieves 70-88% F1-score with only 12 shots in all projects, outperforming the SOTA, which proved ineffective (34-52% F1-score) in 4 projects. Overall, this study underlines the importance of data quality over quantity and provides a more efficient and practical framework for the detection of intermittent job failures in organizations.

Recent reinforcement learning methods for diffusion large language models (dLLMs) commonly rely on on-policy rollouts generated by the target dLLM itself. When successful on-policy rollouts are scarce, however, on-policy training may receive little positive reward and make only limited progress. To mitigate this problem, we explore incorporating higher-reward rollouts generated by a stronger external policy alongside on-policy rollouts from the target dLLM. However, directly incorporating these external rollouts introduces two practical challenges: differences in rollout length and instability when jointly processing rewards from on-policy and external rollouts. To address these challenges, we propose External Rollout Integration with Length Control and Source-Specific Processing (ERILS), which controls external-rollout length and processes the rewards of on-policy and external rollouts separately. Experiments on Sudoku, Countdown, and MATH500 under zero-shot evaluation show that ERILS improves multi-sample performance across all three tasks, with the largest gains on Sudoku. On Sudoku, ERILS achieves 98.4% best-of-4 completion accuracy, compared with 40.3% for the strongest baseline. ERILS also maintains approximately 90% deterministic single-completion accuracy on Sudoku across generation lengths of 128, 256, and 512 tokens. Our component analysis further shows that length-controlled external rollouts are more effective than uncontrolled external rollouts, and that source-specific reward processing avoids the training collapse observed with joint reward processing. These results show that rollout construction and reward processing are important design dimensions when integrating external rollouts into dLLM reinforcement learning.

Tool-using LLM agents are mostly evaluated assuming all tools work. When a tool times out, returns a week-stale value, or has its description poisoned in deployment, the developer needs a controlled way to reproduce the failure, test a fix, and confirm the fix worked before deployment. We present AgentCheck, an open-source web workbench that turns an MCP server into an intervention surface. AgentCheck runs an agent against its real tools and records every tool response, then re-runs the agent with the response perturbed by a fault (12 types) injector. Matching tool calls are replayed from cache, and later tool calls go live after the agent diverges. This yields a reproduce-intervene-confirm loop: the developer toggles a mitigation, re-runs against the identical fault, and sees if the failure goes away. Scoring has two parts: deterministic pass/fail rules, plus an LLM judge for interpretive labels, validated against human annotations. Across five agents, the best passes 105/120 scenarios and the weakest only 77. The failures are usually silent, confident use of incorrect tool outputs rather than crashes. On the weakest agent, a retry mitigation raises success on timeout error faults from as few as 30% of cases to 100%, whereas stale-data faults remain near 3-4 of 10 regardless of the mitigation. AgentCheck makes these failure modes reproducible, comparable, and verifiable before deployment.

When an off-the-shelf coding agent is used directly as a robot policy, observing a browser-based 3D interface through screenshots and acting by posing a virtual target gripper through a few tools, the agent's harness, its prompts, tools, and control rules, largely determines success, and until now it has been written by hand. We show that this harness can be improved automatically by another coding agent, the optimizer agent, and report two findings about what makes it work. First, the number of rollouts the optimizer agent sees per round governs whether the evolved harness is trustworthy, generalizes, and improves steadily. A single rollout is a noisy binary outcome, so with few rollouts per round a revision can be promoted on luck; enlarging the batch raises the signal-to-noise ratio of every promotion decision. Holding rounds fixed and growing the training set from 5 to 100 rollouts, held-out success rises from 47% to 67%, while small training sets overfit, reaching 70% on training tasks but only 54% held-out. Second, the optimizer agent must not be given free rein. With every revision it proposes accepted unconditionally, performance drifts downward within ten rounds as ill-judged edits accumulate; adding the most basic safeguard, Champion-Challenger selection that promotes a revision only if it strictly beats the incumbent on the same fixed evaluation set, turns the same loop into one that raises held-out success from 51% to 67% over 30 rounds. Automatic harness evolution for visual-interface robot agents is thus feasible, but its gains hinge on the rollout scale behind each decision and on how the optimizer agent's revisions are selected.

Autoregressive models accumulate error over long rollouts, yet at deployment there is no ground truth to measure it against. We train a single conditional latent diffusion model that steps a dynamical system forward or backward in time via a direction flag, and show that this bidirectionality supplies a measurement-free test-time error signal: rolling forward i steps and then backward i steps must return the model to its start, so the round-trip discrepancy C_i is a self-supervised proxy for the unobservable rollout error: no ensembles, no held-out data, no governing equations, for one extra rollout. We validate on compressible magnetohydrodynamics (MHD), an astrophysical turbulent radiative mixing layer, and natural face videos (CelebV-HQ). On held-out MHD trajectories, C_i ranks rollout error (Spearman 0.91-0.98 at fixed depth; 0.69 pm 0.16 within trajectories), and a simple calibrator fit on training rollouts predicts its magnitude to within 1.14times (68%) and 1.29times (95%) with near-nominal coverage - one nat beyond a depth-only predictor, transferring to all six decoded physical fields. The same signal flags the out-of-distribution Orszag-Tang vortex (AUROC 0.98; 1.0 by depth 10) exactly where sampling-dispersion baselines invert, and it cuts incurred error by 15% at 80% coverage - three times the depth-only baseline. Bidirectional training comes at negative cost, beating direction specialists in both directions, and the backward direction doubles as a fast inverse solver. On LE-PDE-UQ's turbulent Navier-Stokes benchmark, a single bidirectional model reaches accuracy within 1.3times of their ten-model ensemble at a tenth of the training cost, with the best training-free pixel-level calibration. Round-trip consistency turns reversibility into a practical trust signal for generative models.

Agentic coding assistants are increasingly given extra capabilities, such as browser based testing tools and design oriented system prompts, on the assumption that more capability yields better software. This study tested that assumption directly. Ninety independent agent runs built the same application, a real time retrospective board, from one detailed specification, each scored on a fixed 14 criterion functional rubric (42 point maximum) and a visual quality review. The runs spanned several model generations, two agent harnesses, two reasoning effort levels, a testing tool, and two design oriented prompts. Capability tier dominated: frontier models clustered near the ceiling while a low cost local model fell to 24 to 37 points. A criterion level analysis revealed what run totals conceal. Container deployment was the dominant defect, failing first try in 44 percent of runs, with its failure rate shifting sharply across model generations while mean totals moved less than a point. The testing tool raised cost by 42 to 68 percent without improving functional score or reliability, even on interface visible criteria. Raising reasoning effort from High to xHigh lifted first try perfect runs from 28 percent to 89 percent and cut corrective prompts about five fold, for 9 to 29 percent more cost. A design oriented prompt raised visual quality, 4.5 versus 3.0 on a 5 point scale, without lifting function, and a one paragraph paraphrase of its directive reproduced the entire lift. The practical lesson is to match the fix to the failure: most first run failures came from weak reasoning, which a stronger model or more effort prevents, not from visible flaws a checking tool would catch.

In this work, we reframe agentic trajectories as paths through a shared embedding space of actions and environmental observations. We show that successful and failed rollouts of the same task separate in this space, with simple per-task discriminators predicting held-out rollout outcomes at up to 0.95 AUROC across three agent benchmarks and three policy models. To exploit this signal across tasks, we introduce Precog-RM, a transformer-based model that meta-learns a representation in which outcome prediction can be performed in a few-shot setting. We show that Precog-RM generalizes to unseen tasks and models, and gives confident predictions of trajectory success long before termination. Deployed as an early stopping controller for Group Relative Policy Optimization (GRPO) on Tau2-Bench, Precog-RM cuts training FLOPs by up to 47.3\% while maintaining performance comparable to baseline GRPO.

AI systems can fail silently. The failure propagates through training loops, evaluation pipelines, and production monitoring stacks until downstream harm makes it visible. This paper introduces evaluation blindness: a measurement function M exhibits evaluation blindness with respect to failure class F when it produces readings indistinguishable from a healthy state while the system is actually failing, with no auxiliary signal flagging the gap. The problem surfaces at two lifecycle stages the literature has treated separately. At training time, reward models are gamed, importance-sampling corrections are silently miscalculated, and benchmark contamination inflates fine-tuning evaluations, all while loss curves look healthy and gradient updates proceed normally. At deployment time, monitoring fails to catch six classes of production failure, including an Operational category that is 100% silent by structural definition. We provide a formal detectability predicate unifying both stages. Four training-time case studies trace concrete breakdowns, including a real implementation bug in TRL PR #6594 where gradients are corrupted as loss decreases normally. A six-class taxonomy validated against 50 real-world incidents from court documents and regulatory filings finds that 53% of verifiable public failures were silent. A failure budget framework ties acceptable failure rates to use-case risk class. The implication is direct: measurement infrastructure is a correctness concern across the full AI lifecycle, not just at evaluation time. Data, code, and taxonomy schema are at https://github.com/priyanka25aug/llm-failure-taxonomy.

Reinforcement learning with verifiable rewards (RLVR) has emerged as a promising paradigm for enhancing the reasoning capabilities of large language models (LLMs). In this context, models explore reasoning trajectories and exploit rollouts with correct answers as positive signals for policy optimization. However, these rollouts might involve flawed patterns such as answer-guessing and jump-in-reasoning. Such flawed-positive rollouts are rewarded identically to fully correct ones, causing policy models to internalize these unreliable reasoning patterns. In this work, we first conduct a systematic study of flawed-positive rollouts in RL and find that they enable rapid capability gains during the early optimization stage, while constraining reasoning capability later by reinforcing unreliable patterns. Building on these insights, we propose Flawed-Aware Policy Optimization (FAPO), which presents a parameter-free reward penalty for flawed-positive rollouts, enabling the policy to leverage them as useful shortcuts in the warm-up stage, securing stable early gains, while gradually shifting optimization toward reliable reasoning in the later refinement stage. To accurately and comprehensively detect flawed-positive rollouts, we introduce a generative reward model (GenRM) with a process-level reward that precisely localizes reasoning errors. Experiments show that FAPO is effective in broad domains, improving outcome correctness, process reliability, and training stability without increasing the token budget.

Reinforcement learning (RL) has become a cornerstone in advancing large-scale pre-trained language models (LLMs). Successive generations, including GPT-o series, DeepSeek-R1, Kimi-K1.5, Grok 4, and GLM-4.5, have relied on large-scale RL training to enhance reasoning and coding capabilities. To meet the community's growing RL needs, numerous RL frameworks have been proposed. However, RL training remains computationally expensive, with rollout generation accounting for more than 90% of total runtime. In addition, its efficiency is often constrained by the long-tail distribution of rollout response lengths, where a few lengthy responses stall entire batches, leaving GPUs idle and underutilized. As model and rollout sizes continue to grow, this bottleneck increasingly limits scalability. To address this challenge, we propose Active Partial Rollouts in Reinforcement Learning (APRIL), which mitigates long-tail inefficiency. In the rollout phase, APRIL over-provisions rollout requests, terminates once the target number of responses is reached, and recycles incomplete responses for continuation in future steps. This strategy ensures that no rollouts are discarded while substantially reducing GPU idle time. Experiments show that APRIL improves rollout throughput by 22.5% on average (at most 44%) across commonly used RL algorithms (GRPO, DAPO, GSPO), accelerates convergence, and achieves 2.1% on average(at most 8%) higher final accuracy across tasks. Moreover, APRIL is both framework and hardware agnostic, already integrated into the slime RL framework, and deployable on NVIDIA and AMD GPUs alike. Taken together, this work unifies system-level and algorithmic considerations in proposing APRIL, with the aim of advancing RL training efficiency and inspiring further optimizations in RL systems. Our codebase is available at https://github.com/RLsys-Foundation/APRIL

Software engineers develop, fine-tune, and deploy deep learning (DL) models. They use and re-use models in a variety of development frameworks and deploy them on a range of runtime environments. In this diverse ecosystem, engineers use DL model converters to move models from frameworks to runtime environments. However, errors in converters can compromise model quality and disrupt deployment. The failure frequency and failure modes of DL model converters are unknown. In this paper, we conduct the first failure analysis on DL model converters. Specifically, we characterize failures in model converters associated with ONNX (Open Neural Network eXchange). We analyze past failures in the ONNX converters in two major DL frameworks, PyTorch and TensorFlow. The symptoms, causes, and locations of failures (for N=200 issues), and trends over time are also reported. We also evaluate present-day failures by converting 8,797 models, both real-world and synthetically generated instances. The consistent result from both parts of the study is that DL model converters commonly fail by producing models that exhibit incorrect behavior: 33% of past failures and 8% of converted models fell into this category. Our results motivate future research on making DL software simpler to maintain, extend, and validate.

Large Language Model (LLM) agents increasingly solve long-horizon tasks through multi-turn interactions with users and external tools. In these settings, relevant task information often unfolds over time rather than being fully specified at the initial prompt. Service agents make this challenge especially concrete: users may clarify or revise their goals, while tool responses provide information needed for subsequent decisions. Thus, a final reward alone cannot indicate which actions contributed to resolving the task. Recent methods rely on comparative evidence from other trajectories or resampled continuations, or on separately constructed step-level learning signals, to refine credit. However, a completed rollout already records how information and errors flow between agent actions. We introduce Influence-Aware Policy Optimization (IAPO), which represents each rollout as a typed influence-dependency graph over trainable agent actions, with user and tool observations serving as evidence. IAPO converts support-use and failed-use structure into routing weights that redistribute the same trajectory-level advantage. Experiments with Qwen3-4B and Qwen3-8B demonstrate superior performance over multi-turn reinforcement learning (RL) baselines across three service-agent benchmarks: {τ^2}-Bench, UserBench, and AgentChangeBench. BFCL-v4 Multi-Turn further shows that these gains do not compromise multi-turn function-calling performance. This work advances the understanding of credit assignment in multi-turn user interactions and provides a principled approach to training service agents from sparse outcome feedback.

While cloud-native microservice architectures have revolutionized software development, their inherent operational complexity makes failure Root Cause Analysis (RCA) a critical yet challenging task. Numerous data-driven RCA models have been proposed to address this challenge. However, we find that the benchmarks used to evaluate these models are often too simple to reflect real-world scenarios. Our preliminary study reveals that simple rule-based methods can achieve performance comparable to or even surpassing state-of-the-art (SOTA) models on four widely used public benchmarks. This finding suggests that the oversimplification of existing benchmarks might lead to an overestimation of the performance of RCA methods. To further investigate the oversimplification issue, we conduct a systematic analysis of popular public RCA benchmarks, identifying key limitations in their fault injection strategies, call graph structures, and telemetry signal patterns. Based on these insights, we propose an automated framework for generating more challenging and comprehensive benchmarks that include complex fault propagation scenarios. Our new dataset contains 1,430 validated failure cases from 9,152 fault injections, covering 25 fault types across 6 categories, dynamic workloads, and hierarchical ground-truth labels that map failures from services down to code-level causes. Crucially, to ensure the failure cases are relevant to IT operations, each case is validated to have a discernible impact on user-facing SLIs. Our re-evaluation of 11 SOTA models on this new benchmark shows that they achieve low Top@1 accuracies, averaging 0.21, with the best-performing model reaching merely 0.37, and execution times escalating from seconds to hours.

GPU training jobs fail often, roughly two in five on large production clusters, yet the operator typically learns of a failure only by reconnecting hours later. Experiment trackers require editing the training script and maintaining a cloud connection; the scheduler's mail hook delivers a single status line with no cause and no logs. GPUAlert is a command-line wrapper that monitors any training command at the process boundary, and with no change to that command, emails a structured notification on completion carrying a classified failure cause, durable logs, and output artifacts. The tool is organized around three reliability primitives: a pre-launch log guarantee that establishes the durable destination before the child process can crash, notifier isolation that makes the wrapper's exit code a pure function of the child's status regardless of whether the email succeeds, and a non-silent artifact budget that bounds attachment size without ever dropping output silently. We release a labelled corpus of 474 GPU training logs across 15 failure classes and a reproducible evaluation harness. On the twelve hardware-reproduced classes, the ordered-rule classifier reaches 0.997 macro-F1, against 0.830 for unordered keyword matching and 0.133 for exit-code inspection. Wrapper overhead is a constant approximately 3ms per job; the pre-launch guarantee preserves a log where a shell redirect yields nothing; and across all 15 failure modes the wrapper returns the child's exit code unchanged even when the SMTP relay is unreachable.

Vision-language-action (VLA) models provide a promising paradigm for scalable robotic manipulation, yet their reliance on success-only behavioral cloning leaves them brittle; lacking corrective training signals, minor execution errors rapidly compound into unrecoverable, out-of-distribution failures. To address this limitation, we propose Adaptive Failure-Informed Learning (AFIL), an end-to-end framework that leverages failure trajectories as adaptive negative guidance for diffusion- and flow-based VLA policies. AFIL uses a pretrained VLA to generate failure rollouts online, avoiding the need for handcrafted failure-mode design or human-in-the-loop recovery. It then jointly trains Dual Action Generators (DAGs) for successful and failed behaviors while sharing a common vision-language backbone, enabling efficient failure-aware policy learning with limited parameter overhead. During sampling, the failure generator adaptively steers action generation away from failure-prone regions and toward more reliable success modes, with guidance strength determined by the per-diffusion-step distance between success and failure distributions. Experiments across in-domain and out-of-domain robotic manipulation tasks, covering both short- and long-horizon settings, show that AFIL consistently improves task success rates and robustness over existing VLA baselines, demonstrating its effectiveness, efficiency, and generality.

Reinforcement learning (RL) has emerged as a standard post-training paradigm for shaping large language models (LLMs) into capable agents. In agentic RL, the rollout stage generates trajectories while invoking tools, producing long-tailed and non-stationary workloads that expose two fundamental challenges in resource management. First, due to the long-tail distribution, a small fraction of trajectories dominates rollout makespan. Second, rollout and training are subject to cross-stage imbalance, as they exhibit strong asymmetry in compute patterns, memory demands, and sensitivity to sequence length. Compounding this asymmetry, the sequence length distribution drifts continuously as the policy evolves, rendering any static resource split progressively suboptimal. We present Libra, a resource management system to address both challenges via two core mechanisms. The first is a global resource planner that jointly optimizes GPU allocation across rollout and training clusters. It leverages an elastic hybrid pool to enable lightweight, non-blocking worker reallocation between stages. The second is a causality-driven multi-level feedback queue (C-MLFQ) scheduler, which routes requests to heterogeneous rollout buckets based on causal signals derived from tool-return outcomes, rather than relying on fragile length predictions. Evaluated on 48 A800 GPUs, Libra achieves up to 3.0x higher throughput and converges up to 2.5x faster in reward compared to the baselines.

Large Language Models (LLMs) are being increasingly used in software engineering tasks, with an increased focus on bug report resolution over the past year. However, most proposed systems fail to properly handle uncertain or incorrect inputs and outputs. Existing LLM-based tools and coding agents respond to every issue and generate a patch for every case, even when the input is vague or their own output is incorrect. There are no mechanisms in place to abstain when confidence is low. This leads to unreliable behaviour, such as hallucinated code changes or responses based on vague issue reports. We introduce BouncerBench, a benchmark that evaluates whether LLM-based software agents can refuse to act when inputs are ill-defined or refuse to respond when their own outputs are likely to be incorrect. Unlike prior benchmarks that implicitly incentivize models to generate responses even when uncertain, BouncerBench aims to improve precision by targeting two overlooked failure points: (1) vague or underspecified issue descriptions in tickets and (2) logically or functionally incorrect code patches created by the system. It measures whether proposed systems can distinguish actionable issues from vague tickets and valid patches from untrustworthy ones. We also implement a basic input and output bouncer, evaluating how well current LLMs can abstain when needed. Our results show that most models fail to abstain from underspecified inputs or incorrect outputs. Hence, we conclude that there is significant room for improvement before LLMs can be trusted to make correct decisions and recommendations in real-world software engineering workflows. BouncerBench provides a first step toward evaluating and building more cautious, trustworthy code agents. The replication package, dataset, and leaderboard can be found at bouncerbench.com

Large language model post-training generates self-generated rollouts through reinforcement learning and on-policy distillation, yet this experience is often treated as stale once the policy advances. Historical rollouts can remain compatible with a later policy while preserving behaviors that the policy no longer expresses reliably. However, they may also contain mistakes, abandoned attempts, and redundant actions that should not be imitated, motivating finer-grained selective supervision. We introduce ROSS (Relearning from Self-Generated Rollouts through Selective Supervision), which preserves the full historical trajectory as context while applying loss only to selected model-generated continuations. Across domain-specific reinforcement learning, multi-teacher on-policy distillation, and agentic reinforcement learning, ROSS consistently improves upstream checkpoints and outperforms baselines across mathematics, code generation, instruction following, and software engineering. On Qwen3.6-35B-A3B, ROSS improves the six-benchmark MOPD average from 58.40% to 62.20% and SWE-bench Verified from 64.20% to 68.40%. These results show that self-rollout training leaves behind reusable behavioral experience that can yield further gains through offline supervised fine-tuning (SFT), without additional policy rollouts.

Large Language Models (LLMs) often generate code with subtle but critical bugs, especially for complex tasks. Existing automated repair methods typically rely on superficial pass/fail signals, offering limited visibility into program behavior and hindering precise error localization. In addition, without a way to learn from prior failures, repair processes often fall into repetitive and inefficient cycles. To overcome these challenges, we present TraceCoder, a collaborative multi-agent framework that emulates the observe-analyze-repair process of human experts. The framework first instruments the code with diagnostic probes to capture fine-grained runtime traces, enabling deep insight into its internal execution. It then conducts causal analysis on these traces to accurately identify the root cause of the failure. This process is further enhanced by a novel Historical Lesson Learning Mechanism (HLLM), which distills insights from prior failed repair attempts to inform subsequent correction strategies and prevent recurrence of similar mistakes. To ensure stable convergence, a Rollback Mechanism enforces that each repair iteration constitutes a strict improvement toward the correct solution. Comprehensive experiments across multiple benchmarks show that TraceCoder achieves up to a 34.43\% relative improvement in Pass@1 accuracy over existing advanced baselines. Ablation studies verify the significance of each system component, with the iterative repair process alone contributing a 65.61\% relative gain in accuracy. Furthermore, TraceCoder significantly outperforms leading iterative methods in terms of both accuracy and cost-efficiency.

In Agentic AI, Large Language Models (LLMs) are increasingly used in the orchestration layer to coordinate multiple agents and to interact with external services, retrieval components, and shared memory. In this setting, failures are not limited to incorrect final outputs. They also arise from long-horizon interaction, stochastic decisions, and external side effects (such as API calls, database writes, and message sends). Common failures include non-termination, role drift, propagation of unsupported claims, and attacks via untrusted context or external channels. This paper presents an assurance framework for such Agentic AI systems. Executions are instrumented as Message-Action Traces (MAT) with explicit step and trace contracts. Contracts provide machine-checkable verdicts, localize the first violating step, and support deterministic replay. The framework includes stress testing, formulated as a budgeted counterexample search over bounded perturbations. It also supports structured fault injection at service, retrieval, and memory boundaries to assess containment under realistic operational faults and degraded conditions. Finally, governance is treated as a runtime component, enforcing per-agent capability limits and action mediation (allow, rewrite, block) at the language-to-action boundary. To support comparative evaluations across stochastic seeds, models, and orchestration configurations, the paper defines trace-based metrics for task success, termination reliability, contract compliance, factuality indicators, containment rate, and governance outcome distributions. More broadly, the framework is intended as a common abstraction to support testing and evaluation of multi-agent LLM systems, and to facilitate reproducible comparison across orchestration designs and configurations.

As browser-use agents improve, benchmarks keep pace by collecting new tasks, websites, and applications, often making tasks longer or more novel. This makes difficulty expensive to refresh and difficult to control: when many aspects change at once, it is unclear what actually makes a task challenging. We instead construct challenging instances from tasks agents already solve, turning difficulty into a programmable property of the environment. BreakingWeb pairs every base task with an intervention condition that preserves the user instruction, latent target, and backend success criterion while changing the environment at different web stack layers. Each intervention is deterministic, detectable, and recoverable, and is annotated with the cognitive primitive it primarily loads. The benchmark contains 519 clean/intervention task pairs across seven self-hosted websites and 29 intervention families, all graded against outcomes. We evaluate six strong browser-use agents, three GUI-only agents that see only screenshots, and humans. The construction is effective: interventions cut agent pass rate by 22.9% on average and overturn nearly half of the tasks each agent solves cleanly, whereas humans lose 10.0% on a first attempt and 5.7% after one familiarisation attempt. The dominant failure is belief failure: 75% of the six agents' failures end with a declared success although the required change never happened. Our code, data and environment are publicly available at www.breakingweb.app.

Working agents need to read diverse files, coordinate tools, and produce deliverables. Training such agents requires tasks built on many real files with verifiable results, but few pipelines exist to synthesize this kind of data. Existing pipelines either generate files with models, which lack realism and diversity, or build tasks on real files without task-specific verifiers, leaving result quality unchecked. We introduce GraphForge, an evidence-graph based framework that grounds both the task and its verification in real files. Starting from occupation-grounded seeds for controlled diversity, GraphForge assembles a workspace of real files for each seed and builds an evidence graph over their relations. Since the task statement and rubrics are both derived from this graph, task requirements are backed by the workspace files and each criterion is anchored to the files needed to verify it. An initial rollout further tests executability, and a revision agent repairs the task and rubrics against the original files before trajectories are collected. Fine-tuning Qwen3.6-27B on 2,169 GraphForge trajectories brings GDPVal to 1445.7 (+65.7) under OpenHands, and Workspace-Bench-Lite and SpreadsheetBench II to 63.7 (+7.7) and 24.0 (+13.7) under Claude Code. Rejection fine-tuning on the SFT model's own rollouts, with candidates selected by the evidence-anchored rubrics, yields further improvements on all three benchmarks, suggesting that the rubrics provide a useful selection signal. The data and models are available.

Tool-using automation systems, from scripts and CI bots to agentic assistants, fail in recurring patterns. Common failures include unsafe side effects, invalid arguments, uncontrolled retries, and leakage of sensitive outputs. Many mitigations are model-centric and prompt-dependent, so they are brittle and do not generalize to non-LLM callers. We present Policy-First Tooling, a model-agnostic permission layer that mediates tool invocation through explicit constraints, risk-aware gating, recovery controls, and auditable explanations. The paper contributes a compact policy DSL, a runtime enforcement architecture with actionable rationale and fix hints, and a reproducible benchmark based on trace replay with controlled fault and misuse injection. In 225 controlled runs across five policy packs and three fault profiles, stricter packs improve violation prevention from 0.000 in P0 to 0.681 in P4, while task success drops from 0.356 to 0.067. Retry amplification decreases from 3.774 in P0 to 1.378 in P4, and leakage recall reaches 0.875 under injected secret outputs. These results make safety to utility trade-offs explicit and measurable.

Test-time reinforcement learning (TTRL) enables large language models (LLMs) to self-improve on unlabeled inputs, but its effectiveness critically depends on how reward signals are estimated without ground-truth supervision. Most existing TTRL methods rely on majority voting (MV) over rollouts to produce deterministic rewards, implicitly assuming that the majority rollout provides a reliable learning signal. We show that this assumption is fragile: MV reduces the rollout distribution into a single outcome, discarding information about non-majority but correct actions candidates, and yields systematically biased reward estimates. To address this, we propose Distribution-AwareReward Estimation (DARE), which shifts reward estimation from a single majority outcome to the full empirical rollout distribution. DARE further augments this distribution-based reward with an exploration bonus and a distribution pruning mechanism for non-majority rollout exploration and reward denoise, yielding a more informative and robust reward estimation. Extensive experiments on challenging reasoning benchmarks show that DARE improves optimization stability and final performance over recent baselines, achieving relative improvements of 25.3% on challenging AIME 2024 and 5.3% on AMC.

When adapting Vision Language Action (VLA) models to downstream tasks, multiple rounds of post-training are often required to progressively address policy weaknesses. In this report, we focus on maximizing human efficiency during this iterative process, measured by policy improvement and task throughput per unit of human labor and time. We propose HELP, a Human-Efficient Large-scale robot Post-training pipeline in which two specialized operators supervise twelve robots concurrently. A trained Teleoperator provides high-value remote interventions and recovery demonstrations, while a Floor Operator monitors the robot fleet, triggers takeovers, and performs physical resets. This role specialization improves human efficiency by reducing task switching, lowering operator training costs, and expanding robot interaction coverage. Beyond increasing rollout volume, concurrent supervision also broadens the range of policy behaviors observed by the human team, making recurring failure modes easier to identify and enabling more targeted takeovers, resets, and recovery demonstrations. To efficiently utilize the large and mixed-quality rollout data, HELP incorporates \vlac, an automatic rollout segmentation critic specifically designed for this setting. It separates autonomous trajectories into progress-making, idle, failure-inducing, and recovery segments. Useful rollout segments are retained and combined with Human-in-the-Loop data for the next post-training round. Across four real-world manipulation tasks, HELP achieves 80\%--95\% success rates and improves task throughput by 1.7times--4.2times over the base model. Under matched HITL recovery budgets, VLAC-CUT further amplifies throughput gains by 1.20times--3.43times and success-rate gains by 1.50times--3.00times over HITL-only updates.

Modern web dashboards and enterprise applications increasingly rely on complex, distributed microservices architectures. While these architectures offer scalability, they introduce significant challenges in debugging and observability. When failures occur, they often manifest as opaque error messages to the end-user such as Something went wrong. This masks the underlying root cause which may reside in browser side exceptions, API contract violations, or server side logic failures. Existing monitoring tools capture these events in isolation but fail to correlate them effectively or provide intelligible explanations to non technical users. This paper proposes a novel system for Automated Multi Source Debugging and Natural Language Error Explanation. The proposed framework automatically collects and correlates error data from disparate sources such as browser, API, server logs and validates API contracts in real time, and utilizes Large Language Models to generate natural language explanations. This approach significantly reduces Mean Time to Resolution for support engineers and improves the user experience by transforming cryptic error codes into actionable insights.

Long-horizon agents must coordinate reasoning, tool use, and environment feedback over extended interactions. Learning from sparse outcome rewards requires weighting trajectories of different lengths and updating the policy while rollouts are still being collected. We examine a single-rollout asynchronous recipe combining Muon, token-averaged loss, truncated importance sampling, and auxiliary length and tool-error penalties. Single-rollout sampling allows asynchronous learning without waiting for multiple trajectories for the same issue. We study optimizer choices and recipe components for repository-level issue resolution across two model families. The evaluated training runs show higher training success with the full recipe than with the baselines.

Joint-embedding world models enable visual planning by learning action-conditioned dynamics in latent space. Yet they are commonly trained for one-step prediction on encoded states, while planning recursively applies the learned transition to its own predictions. One-step accuracy therefore does not capture how prediction errors propagate under recursive rollout. We decompose multi-step rollout error into the errors introduced at individual steps and their propagation through subsequent transitions. We show that state-affine dynamics are precisely the differentiable transitions with state-independent Jacobians, eliminating the nonlinear propagation residual and making the error propagation operators depend only on the action sequence. Guided by this result, we introduce SALT (State-Affine Latent Transition), an action-conditioned state-affine dynamics model in which the action modulates both the state transformation and the additive update. We train SALT through recursive multi-step rollout supervision, feeding each predicted latent state back into the transition so that training matches how the model is used during planning. Across four visual planning environments, SALT exhibits 1.48--2.19times higher one-step prediction error than the matched LeWM baseline, yet improves closed-loop success in every environment by 10.0 percentage points on average. On OGBench-Cube, the fraction of episodes that fail with a sharp rise in model-predicted cost after execution decreases from 23.3% to 2.0%.

Autonomous agents increasingly perform bounded software tasks under an orchestrator that retries, resumes, and budgets them. The machinery such orchestrators reach for is the service mesh's: retry, timeout, and error-rate circuit breaking. We report a failure study of a production agentic software-delivery platform over 147 numbered incidents spanning 81 runs, each with a measured cost and, in most cases, a mutation proof reproducing the failure. All three assumptions those primitives rest on are violated in practice, and we quantify the consequences: a loop of fifty-four consecutive successful tool calls no error-rate breaker could see; a progress signal constant by construction, guaranteeing a false trip on the third repair round and driving one run from six of six components to three; twenty-one events accumulated across six invocations of one delegation, making a correct, idempotent component unwinnable; a misrouted failure that woke five components for a two-component fault, leaving three bystanders regressing working code; and twelve incidents in which the enforcement layer blocked correct work, the most expensive costing 107 agent turns and zero accepted writes. We find one cross-cutting cause and its dual. Identity adequacy: in five separate subsystems an identity that failed to discriminate produced a confident wrong answer, and two of them derived the corrective rule independently. Evidence adequacy: a reliability decision may be taken only on evidence capable of moving, attributable to what it measures, and deterministic under identical conditions. From the findings we derive seven reliability primitives whose enforcement unit is the delegation rather than the message, and specify the controlled evaluation the study motivates but does not constitute.

When a large language model under reinforcement learning commits a wrong reasoning step early in a trajectory, standard algorithms force it to keep generating until the maximum horizon, spending compute on tokens that never receive positive reward and polluting advantage estimates with post-failure noise. We propose ESPO (Early-Stopping Proximal Policy Optimization), which detects trajectory failure on-the-fly and terminates rollouts early. At each generation step, ESPO computes a surrogate regret using only the logits already computed during sampling, and terminates when the smoothed cumulative regret significantly exceeds its estimated values. Truncated trajectories are treated as absorbing failure states with a terminal reward, concentrating negative temporal-difference (TD) errors near the detected failure step without any additional reward model or human annotation. On DeepSeek-R1-Distill-Qwen-7B trained for mathematical reasoning, ESPO surpasses PPO on AIME~2024 (46.28% vs. 45.25%), AMC~2023 (85.83% vs. 82.94%), and MATH-500 (87.42% vs. 85.43%), while saving more than 20% rollout tokens cumulatively.

We present WebGym, the largest-to-date open-source environment for training realistic visual web agents. Real websites are non-stationary and diverse, making artificial or small-scale task sets insufficient for robust policy learning. WebGym contains nearly 300,000 tasks with rubric-based evaluations across diverse, real-world websites and difficulty levels. We train agents with a simple reinforcement learning (RL) recipe, which trains on the agent's own interaction traces (rollouts), using task rewards as feedback to guide learning. To enable scaling RL, we speed up sampling of trajectories in WebGym by developing a high-throughput asynchronous rollout system, designed specifically for web agents. Our system achieves a 4-5x rollout speedup compared to naive implementations. Second, we scale the task set breadth, depth, and size, which results in continued performance improvement. Fine-tuning a strong base vision-language model, Qwen-3-VL-8B-Instruct, on WebGym results in an improvement in success rate on an out-of-distribution test set from 26.2% to 42.9%, significantly outperforming agents based on proprietary models such as GPT-4o and GPT-5-Thinking that achieve 27.1% and 29.8%, respectively. This improvement is substantial because our test set consists only of tasks on websites never seen during training, unlike many other prior works on training visual web agents.

Java remains central to enterprise software, and many applications outlive their original architecture. Migrating them across frameworks is a behavior-preserving refactoring spanning build configuration, dependency injection, persistence, request handling, and deployment. Existing software-engineering benchmarks cover bug fixing, feature implementation, and language or version modernization, but leave cross-framework refactoring largely unmeasured. We introduce ScarfBench, a benchmark for behavior-preserving cross-framework refactoring of enterprise Java applications. It is built from expert-written implementation triples across Spring, Jakarta EE, and Quarkus: 34 applications (29 focused single-layer, 5 whole) yielding 102 variants (~151K lines across 1946 source and test files) and 204 directed refactoring tasks. Each task gives an agent a working source application and a target framework; the agent must synthesize a target implementation preserving the source behavior. Correctness is evaluated by an application-specific executable oracle: the candidate must compile, deploy in a containerized target runtime, and pass behavioral tests over the application's observable interface. We evaluate five state-of-the-art coding agents on ScarfBench. The strongest achieves only 15.3% aggregate test pass on focused-layer migrations and 12.2% on whole applications, and only one of the 204 tasks yields a fully behaviorally equivalent target. Difficulty is asymmetric across framework directions and architectural layers: Spring<->Quarkus is the most tractable pair, and Jakarta-targeted migrations are hardest. From LLM-as-a-judge and expert adjudication of failed-task traces, we derive a taxonomy of recurring failure categories spanning build, deploy, and test stages. We release the benchmark, harness, and agent traces at https://scarfbench.info.

Neural operators provide fast surrogates for time-dependent partial differential equations (PDEs) by applying a learned evolution operator recursively to its own predictions, but this autoregressive rollout feeds every prediction error back as input, so local errors accumulate. Existing rollout-training strategies reduce the mismatch between training inputs and self-generated states, yet their supervision still measures only the absolute discrepancy from the ground-truth trajectory. Such supervision is therefore uninformative about whether the operator has overcome the long-horizon failure behaviors it exhibited earlier during optimization. We propose history-enriched rollout training (HERO), which augments conventional absolute trajectory supervision with relative supervision derived from the model's optimization history. HERO ranks detached candidate rollouts from a periodically refreshed lagged operator, the current model, and a perturbed input by rollout error, spectral discrepancy, energy drift, and error growth, and selects the strongest failure trajectory as reference. This reference enters a margin-based objective as a fixed comparison baseline, inducing a bounded, sample-dependent reweighting of the ground-truth rollout gradient rather than an independent gradient direction, which we further analyze theoretically. Experiments on nine PDE benchmarks with spectral and attention-based backbones show that HERO consistently improves long-horizon accuracy, stable rollout length, and out-of-distribution robustness at no inference-time cost. These results indicate that history-enriched relative supervision is effective for stabilizing long-horizon autoregressive prediction.

We introduce SWE-Bench Pro, a substantially more challenging benchmark that builds upon the best practices of SWE-BENCH [25], but is explicitly designed to capture realistic, complex, enterprise-level problems beyond the scope of SWE-BENCH. SWE-BENCH PRO contains 1,865 problems sourced from a diverse set of 41 actively maintained repositories spanning business applications, B2B services, and developer tools. The benchmark is partitioned into a public set with open access to problems sourced from 11 repositories, a held-out set of 12 repositories and a commercial set of 18 proprietary repositories where we have formal partnership agreements with early-stage startups. Problems in the held-out and the commercial set are not publicly accessible, but we release results on the commercial set. Our benchmark features long-horizon tasks that may require hours to days for a professional software engineer to complete, often involving patches across multiple files and substantial code modifications. All tasks are human-verified and augmented with sufficient context to ensure resolvability. In our evaluation of widely used coding models, under a unified scaffold, we observe that their performance on SWE-Bench PRO remains below 25% (Pass@1), with GPT-5 achieving the highest score to date at 23.3%. To better understand these limitations, we cluster the failure modes observed in the collected agent trajectories for a clearer characterization of the error patterns exhibited by current models. Overall, SWE-BENCH PRO provides a contamination-resistant testbed that more faithfully captures the complexity and diversity of real-world software development, advancing the pursuit of truly autonomous software engineering agents at a professional level.

CLI agents are the closest thing language models have to an embodied setting: the model emits commands, the terminal executes them, and the returned stream -- stdout, errors, files, logs, and traces -- records the consequences. We argue that this stream is a supervision signal, but standard agent RL discards it: GRPO-style training updates action tokens with sparse outcome-level rewards while ignoring environment responses already in the rollout. Failed rollouts provide little policy-gradient signal despite containing rich evidence about how the environment responds. We introduce ECHO (Environment Cross-entropy Hybrid Objective), a hybrid objective that combines the standard policy-gradient loss on action tokens with an auxiliary loss that trains the policy to predict environment observation tokens resulting from its own actions. ECHO reuses the same forward pass as GRPO, requires no additional rollouts, and turns terminal feedback into dense supervision for all rollouts. ECHO doubles GRPO pass@1 on TerminalBench-2.0: Qwen3-8B improves from 2.70% to 5.17%, and Qwen3-14B from 5.17% to 10.79%. ECHO also produces policies that better predict terminal dynamics, even on trajectories they did not generate: across held-out rollouts, it sharply reduces environment-token cross-entropy while GRPO alone barely changes it. From base Qwen3-8B, ECHO matches expert-SFT-then-GRPO performance on held-out terminal tasks without expert demonstrations, and recovers roughly half of the expert-SFT initialization benefit on TerminalBench-2.0. In some settings, the environment prediction loss alone enables verifier-free self-improvement, allowing policies to improve on unseen OOD tasks by learning only from environment interactions. Together, these results suggest that environment observations are not merely context for future actions, but a dense, on-policy supervision signal already present in every rollout.

We study whether demonstration-curation metrics that detect defective training episodes also improve the downstream behavior-cloning policy that trains on the curated data. On a contact-rich LIBERO pick-and-place benchmark with a controlled structural defect (early gripper release during the carry phase), we find that the two quantities are sharply decoupled. The metric with the highest defect-detection AUROC (0.804) produces the worst curated policy (13.3% task success), while a metric with a substantially lower AUROC (0.638) produces a policy that nearly matches the oracle trained on ground-truth clean data (90.0% vs. 93.3%). We further show that five of the seven metrics we evaluate exploit episode length as a trivial proxy for the defect label, a confound that inflates reported AUROCs to near-perfect values and disappears once episode length is controlled. Across all conditions, the contaminated baseline succeeds on only 3.3% of rollouts, and the two best curation methods close this to within 3 percentage points of the 93.3% oracle ceiling. Our results argue that curation methods should be evaluated by the policy they produce, not the defects they flag, and that any curation benchmark must control for episode length before reporting detection accuracy. We release the testbed, all metric implementations, and the evaluation pipeline.

Deployed multimodal systems can fail in ways that evaluators did not anticipate. In order to find these failures before deployment, we introduce MultiMon, a system that automatically identifies systematic failures -- generalizable, natural-language descriptions of patterns of model failures. To uncover systematic failures, MultiMon scrapes a corpus for examples of erroneous agreement: inputs that produce the same output, but should not. It then prompts a language model (e.g., GPT-4) to find systematic patterns of failure and describe them in natural language. We use MultiMon to find 14 systematic failures (e.g., "ignores quantifiers") of the CLIP text-encoder, each comprising hundreds of distinct inputs (e.g., "a shelf with a few/many books"). Because CLIP is the backbone for most state-of-the-art multimodal systems, these inputs produce failures in Midjourney 5.1, DALL-E, VideoFusion, and others. MultiMon can also steer towards failures relevant to specific use cases, such as self-driving cars. We see MultiMon as a step towards evaluation that autonomously explores the long tail of potential system failures. Code for MULTIMON is available at https://github.com/tsb0601/MultiMon.

Reinforcement learning (RL) post-training has become a trending paradigm for enhancing the capabilities of large language models (LLMs). Most existing RL systems for LLMs operate in a fully synchronous manner, where training must wait for the rollout of an entire batch to complete. This design leads to severe inefficiencies, as extremely long trajectories can stall the entire rollout process and leave many GPUs idle. To address this issue, we propose Concurrency- Controlled Partial Rollout with Importance Sampling (CoPRIS), which mitigates long-tail inefficiencies by maintaining a fixed number of concurrent rollouts, early-terminating once sufficient samples are collected, and reusing unfinished trajectories in subsequent rollouts. To mitigate the impact of off-policy trajectories, we introduce Cross-stage Importance Sampling Correction, which concatenates buffered log probabilities from the previous policy with those recomputed under the current policy for importance sampling correction. Experiments on challenging mathematical reasoning benchmarks show that CoPRIS achieves up to 1.94x faster training while maintaining comparable or superior performance to synchronous RL systems. The code of CoPRIS is available at https://github.com/777pomingzi/CoPRIS.

AI systems fail silently far more often than they fail visibly. In an analysis of 100K human-AI interactions from the WildChat dataset, we find that 79% of AI failures are invisible: something went wrong but the user gave no overt indication that there was a problem. These invisible failures cluster into eight archetypes that help us characterize where and how AI systems are failing to meet users' needs. In addition, the archetypes show systematic co-occurrence patterns indicating higher-level failure types. To address the question of whether these archetypes will remain relevant as AI systems become more capable, we also created and annotated a counterfactual dataset in which WildChat's 2024-era responses are replaced by those from three present-day frontier LMs. This analysis indicates that failure rates have dropped substantially, but that the vast majority of failures remain invisible in our sense, and the distribution of failure archetypes seems stable. Finally, we illustrate how the archetypes help us to identify systematic and variable AI limitations across different usage domains. Overall, we argue that our invisible failure taxonomy can be a key component in reliable failure monitoring for product developers, scientists, and policy makers. Our code and data are available at https://github.com/bigspinai/bigspin-invisible-failure-archetypes

Multi-step enterprise agent tasks fail in a characteristic way: single-pass inference has no checkpoint between deciding an answer and committing to it. We study one production system (Leni) whose architecture installs such checkpoints: verification loops (execute, observe, compare, correct) staffed by lightweight task-specialized post-trained models. We evaluate the unmodified production configuration on three public benchmarks stressing distinct failure modes: SpreadsheetBench Verified (silent computation error), BullshitBench v2 (premise confabulation), and the GAIA validation split (cascade error over long tool chains). The full system improves over its frontier base model by +11.0 percentage points on SpreadsheetBench (91.25% vs 80.25%, n=400, p<0.001), +7 to +10 percentage points on BullshitBench (98% vs 91%, n=100), and roughly +15 points on GAIA validation (75.2% pass@1, n=165; 83.0% best-of-k). Our central contribution is a decomposition of that uplift: most of it comes from scaffolding, routing, and specialist models rather than from the verification step itself, whose isolated contribution is small (+1.5 points) but concentrated at the top of the score distribution, where it converts otherwise-failing tasks. We instrument the loop end-to-end, yielding an empirical verifier confusion matrix (catch rate about 0.20, fix rate 0.75, no false-alarm regressions) that grounds a compounding-reliability model. Specialist-swap ablations suggest that the loop's value depends on who observes it: replacing the small trained verifier with the generating frontier model eliminates most rescues. A valid-premise control shows zero over-rejections in 100 expert-level questions.

We report a case study of four end-to-end attempts to autonomously generate ML research papers using a pipeline of six LLM agents mapped to stages of the scientific workflow. Of these four, three attempts failed during implementation or evaluation. One completed the pipeline and was accepted to Agents4Science 2025, an experimental inaugural venue that required AI systems as first authors, passing both human and multi-AI review. From these attempts, we document six recurring failure modes: bias toward training data defaults, implementation drift under execution pressure, memory and context degradation across long-horizon tasks, overexcitement that declares success despite obvious failures, insufficient domain intelligence, and weak scientific taste in experimental design. We conclude by discussing four design principles for more robust AI-scientist systems, implications for autonomous scientific discovery, and we release all prompts, artifacts, and outputs at https://github.com/Lossfunk/ai-scientist-artefacts-v1

Reinforcement learning with verifiable rewards (RLVR) has emerged as a highly effective framework for improving LLM reasoning, with methods such as GRPO among its most successful instantiations. However, GRPO relies on repeated generation of long chain-of-thought rollouts. Training time scales with the number of rollouts, a large fraction of which are uninformative. Thus, GRPO is computationally expensive and unstable. To mitigate this, existing approaches either generate a larger pool of rollouts and filter the most informative prompts, or leverage historical signals for filtering at later stages of training. These strategies offer modest performance gains, but slow down the overall process. To address this, we propose VarIance Guided Online Rollout allocation (VIGOR) which instead of allocating a fixed rollout budget per example, begins with a small number of rollouts for all examples in a batch and iteratively allocates additional rollouts to those with the highest group reward variance until a fixed total rollout budget is reached. Theoretically, we show that under RLVR, reward variance controls the gradient magnitude, and derive VIGOR's closed-form speedup ratio over GRPO, which grows with refinement rounds under Pareto-distributed reward variance. Experiments on mathematical reasoning and coding tasks show that VIGOR reaches target accuracy with up to 2.3times fewer rollouts on math, reaches GRPO's final coding full pass rate with 1.49times fewer rollouts, and improves the coding average test pass rate by 3.4 points.

Autonomous coding agents (e.g., OpenAI Codex, Devin, GitHub Copilot) are increasingly used to generate fix-related pull requests (PRs) in real world software repositories. However, their practical effectiveness depends on whether these contributions are accepted and merged by project maintainers. In this paper, we present an empirical study of AI agent involved fix related PRs, examining both their integration outcomes, latency, and the factors that hinder successful merging. We first analyze 8,106 fix related PRs authored by five widely used AI coding agents from the AIDEV POP dataset to quantify the proportions of PRs that are merged, closed without merging, or remain open. We then conduct a manual qualitative analysis of a statistically significant sample of 326 closed but unmerged PRs, spending approximately 100 person hours to construct a structured catalog of 12 failure reasons. Our results indicate that test case failures and prior resolution of the same issues by other PRs are the most common causes of non integration, whereas build or deployment failures are comparatively rare. Overall, our findings expose key limitations of current AI coding agents in real world settings and highlight directions for their further improvement and for more effective human AI collaboration in software maintenance.

As AI becomes more capable, we entrust it with more general and consequential tasks. The risks from failure grow more severe with increasing task scope. It is therefore important to understand how extremely capable AI models will fail: Will they fail by systematically pursuing goals we do not intend? Or will they fail by being a hot mess, and taking nonsensical actions that do not further any goal? We operationalize this question using a bias-variance decomposition of the errors made by AI models: An AI's incoherence on a task is measured over test-time randomness as the fraction of its error that stems from variance rather than bias in task outcome. Across all tasks and frontier models we measure, the longer models spend reasoning and taking actions, the more incoherent their failures become. Incoherence changes with model scale in a way that is experiment dependent. However, in several settings, larger, more capable models are more incoherent than smaller models. Consequently, scale alone seems unlikely to eliminate incoherence. Instead, as more capable AIs pursue harder tasks, requiring more sequential action and thought, our results predict failures to be accompanied by more incoherent behavior. This suggests a future where AIs sometimes cause industrial accidents (due to unpredictable misbehavior), but are less likely to exhibit consistent pursuit of a misaligned goal. This increases the relative importance of alignment research targeting reward hacking or goal misspecification.

AI systems rewrite information constantly: conversations become stored memories, documents become answers. The rewrite can keep a claim while washing away what made it checkable, who said it, how sure they were, when it held. We call that failure factwashing, and release factwash, an open-source write-time gate that catches it deterministically, with named flags and evidence rather than an LLM judge. Building it answers a practical question: when does a cheap check suffice, and when do you need a model? What decides is whether the property has a bounded surface-cue inventory. Explicit negation cues are close to enumerable, so a word list finishes and transfers, reaching 0.91 F1 on untuned text. Hedging and attribution have open-ended realizations, so vocabulary plateaus near half recall, and a one-question LLM witness recovers +17 and +15 points of cue-detection recall at equal precision. Deployed, that witness may only lower a verdict, so it buys precision rather than coverage. We measure cue detection on 105,596 independently annotated sentences. A blind-labelled corpus of memory writes then locates the failure: 55% of bad writes in conversational hearsay, 7% in business email (p < 0.001), so the first deployment question is not which detector to use but whether the failure occurs at all. On unmodified mem0 2.0.7, the gate flags 5 of 8 hedged-hearsay writes.

Recent prominent post-training methods, such as Reinforcement Learning (RL) and On-Policy Self-Distillation (OPSD), have driven rapid progress in mathematical reasoning for large language models, yet their reliance on ground-truth labels precludes test-time training (TTT). Replacing ground truth with majority-vote pseudo-labels is a natural alternative, yet it is fragile: an incorrect vote corrupts the teacher and misleads every token. We observe that this failure mode is asymmetric: rollouts that disagree with the pseudo-label are typically wrong regardless of whether the vote itself is correct. Building on this observation, we propose Test-Time Policy Optimization (TTPO), an asymmetric objective that distills agreeing rollouts via OPSD and penalizes disagreeing rollouts with Grouped RL. Token-level selection further refines both branches: distillation down-weights already-converged positions, while RL penalizes only confident errors. Both updates remain well-grounded even under frequent pseudo-label errors, and majority-vote routing yields tighter self-supervision as the model improves. Without any labels, TTPO matches label-supervised OPSD on five competition-level benchmarks, raises Qwen3-1.7B from 38.0% to 45.2% in TTT, yields +25.2% to +36.4% without thinking, and shows strong cross-task generalization.

Despite growing enthusiasm for Multi-Agent Systems (MAS), where multiple LLM agents collaborate to accomplish tasks, their performance gains across popular benchmarks remain minimal compared to single-agent frameworks. This gap highlights the need to analyze the challenges hindering MAS effectiveness. In this paper, we present the first comprehensive study of MAS challenges. We analyze five popular MAS frameworks across over 150 tasks, involving six expert human annotators. We identify 14 unique failure modes and propose a comprehensive taxonomy applicable to various MAS frameworks. This taxonomy emerges iteratively from agreements among three expert annotators per study, achieving a Cohen's Kappa score of 0.88. These fine-grained failure modes are organized into 3 categories, (i) specification and system design failures, (ii) inter-agent misalignment, and (iii) task verification and termination. To support scalable evaluation, we integrate MASFT with LLM-as-a-Judge. We also explore if identified failures could be easily prevented by proposing two interventions: improved specification of agent roles and enhanced orchestration strategies. Our findings reveal that identified failures require more complex solutions, highlighting a clear roadmap for future research. We open-source our dataset and LLM annotator.

Reinforcement learning has been central to recent advances in large language model reasoning, but most algorithms rely on on-policy training that demands fresh rollouts at every update, limiting efficiency and scalability. Asynchronous RL systems alleviate this by decoupling rollout generation from training, yet their effectiveness hinges on tolerating large staleness in rollout data, a setting where existing methods either degrade in performance or collapse. We revisit this challenge and uncover a prosperity-before-collapse phenomenon: stale data can be as informative as on-policy data if exploited properly. Building on this insight, we introduce M2PO (Second-Moment Trust Policy Optimization), which constrains the second moment of importance weights to suppress only extreme outliers while preserving informative updates. Notably, M2PO sharply reduces the fraction of clipped tokens under high staleness (from 1.22% to 0.06% over training), precisely masking high-variance tokens while maintaining stable optimization. Extensive evaluation across six models (from 1.7B to 32B) and eight benchmarks shows that M2PO delivers stable off-policy training even with data stale by at least 256 model updates and matches on-policy performance.

Despite algorithm-level innovations for multi-agent reinforcement learning (MARL), the underlying networked infrastructure for large-scale MARL training remains underexplored. Existing training frameworks primarily optimize for single-agent scenarios and fail to address the unique system-level challenges of MARL, including rollout-training synchronization barriers, rollout load imbalance, and training resource underutilization. To bridge this gap, we propose FlexMARL, the first end-to-end training framework that holistically optimizes rollout, training, and their orchestration for large-scale LLM-based MARL. Specifically, FlexMARL introduces the joint orchestrator to manage data flow under the rollout-training disaggregated architecture. Building upon the experience store, a novel micro-batch driven asynchronous pipeline eliminates the synchronization barriers while providing strong consistency guarantees. Rollout engine adopts a parallel sampling scheme combined with hierarchical load balancing, which adapts to skewed inter/intra-agent request patterns. Training engine achieves on-demand hardware binding through agent-centric resource allocation. The training states of different agents are swapped via unified and location-agnostic communication. Empirical results on a large-scale production cluster demonstrate that FlexMARL achieves up to 7.3x speedup and improves hardware utilization by up to 5.6x compared to existing frameworks.

Reinforcement learning (RL) is a critical stage in post-training large language models (LLMs), involving repeated interaction between rollout generation, reward evaluation, and centralized learning. Distributing rollout execution offers opportunities to leverage more cost-efficient inference resources, but introduces challenges in wide-area coordination and policy dissemination. We present ECHO-2, a distributed RL framework for post-training with remote inference workers and non-negligible dissemination latency. ECHO-2 combines centralized learning with distributed rollouts and treats bounded policy staleness as a user-controlled parameter, enabling rollout generation, dissemination, and training to overlap. We introduce an overlap-based capacity model that relates training time, dissemination latency, and rollout throughput, yielding a practical provisioning rule for sustaining learner utilization. To mitigate dissemination bottlenecks and lower cost, ECHO-2 employs peer-assisted pipelined broadcast and cost-aware activation of heterogeneous workers. Experiments on GRPO post-training of 4B and 8B models under real wide-area bandwidth regimes show that ECHO-2 significantly improves cost efficiency while preserving RL reward comparable to strong baselines.

Tool-using large language model agents frequently replan, retry failed operations, delegate tasks, and resume after crashes. These behaviors can cause one user authorization to be requested and executed multiple times under freshly issued token identifiers, even when each individual token is single-use. We call this failure semantic replay: exceeding the execution budget of a token-independent authorization instance rather than merely reusing an old token identifier. We show that identifier-local token consumption cannot prevent fresh reissuance unless the issuer retains monotonic durable state over the authorized action, confirmation event, and remaining execution budget. We introduce CapLease, an authorization-consumption layer that follows proposal- and authority-level defenses, binds an authenticated user confirmation to a canonical action, and enforces transactional Issue-Prepare-Commit transitions. Across LLM-agent replanning, retry, delegation, concurrency, confirmation-replay, and crash-recovery scenarios, identifier-local tokens permit fresh semantic reissuance, whereas CapLease and an equally stateful Server Ledger prevent duplicate admission and, with an idempotent sink, duplicate external effects. Our results identify durable authorization state, rather than token representation alone, as the systems requirement for replay-resistant agent execution.

Agentic LLM coding tools compress long session histories into compaction summaries that subsequent sessions inherit as ground truth. This paper documents a failure mode in Claude Code where partial standard output from timed-out commands (exit code 143) is recorded in compaction summaries as confirmed results, propagating false positives across sessions and model versions without re-verification. The underlying mechanism is a conflation of observation and persistence, where information that appeared in the terminal is treated as equivalent to information written to durable storage. This finding extends the analysis of LLM self-evaluation failures reported in prior work on non-determinism in LLM-as-judge grading by showing that agentic tools exhibit analogous reliability deficits when reporting on their own operational outcomes. The failure has direct implications for any workflow that relies on agentic session continuity for data processing, scientific computation, or multi-step automation.

We study goal misgeneralization, a type of out-of-distribution generalization failure in reinforcement learning (RL). Goal misgeneralization failures occur when an RL agent retains its capabilities out-of-distribution yet pursues the wrong goal. For instance, an agent might continue to competently avoid obstacles, but navigate to the wrong place. In contrast, previous works have typically focused on capability generalization failures, where an agent fails to do anything sensible at test time. We formalize this distinction between capability and goal generalization, provide the first empirical demonstrations of goal misgeneralization, and present a partial characterization of its causes.

A prevalent assumption in LLM agent deployment holds that more structured harnesses universally improve reliability, and that higher-capability models need proportionally less structural guidance -- together implying a monotone inverse relationship between model capability tier and optimal harness complexity. We test this hypothesis through a controlled 432-run experiment crossing six models across four capability tiers with three harness conditions (light, balanced, strict) on HEAT-24, a 24-task synthetic benchmark with git-based workspace verification. Our results refute the monotone inverse relationship on two fronts. First, for the frontier chat model evaluated (Gemini 2.5 Flash), increased harness verbosity lowers VTSR by 29-38 percentage points -- a harness-complexity paradox. Second, for the frontier reasoning model evaluated (Qwen3.5-122B, extended thinking enabled), strict harness achieves the highest VTSR (91.7%) and the lowest latency, the opposite of the prediction. Within the constrained tier, a 2B model (Gemma4:e2B) matches strong-open-tier stability at 91.7% across all harnesses. Because each tier is represented by a single model in this study, these results should be interpreted as model-specific observations; harness sensitivity appears non-monotone across the models evaluated, and depends critically on model type (chat vs. reasoning). We introduce a six-label failure taxonomy showing that format_violation dominates capable-model failures while wrong_file dominates low-capability failures, and we derive practical tier-aware harness selection guidelines.

Long-horizon terminal agents may solve a task in one run yet fail across repeated runs. Stochastic errors can propagate through actions and context, creating a gap between capability and reliable execution. We present RiseHarness, a framework for iteratively improving agent harnesses with model weights fixed. It focuses on flipping tasks with both successful and failed runs under the same configuration, using successful runs as references for failure analysis. RiseHarness combines agent trajectories with internal harness traces from a runtime monitor to examine how model actions interact with command execution, feedback, context management, and completion decisions. A Harness Handbook links this evidence to harness behaviors and source code, helping identify recurring problems in mechanisms shared across tasks. RiseHarness then searches other harness frameworks for relevant implementations and adapts them as design references to the current model and execution protocol. Candidate changes undergo behavioral checks and repeated task evaluation; versions with measured gains seed the next iteration. This loop integrates suitable mechanisms from different frameworks into a single harness and uses new execution evidence to guide further changes. We evaluate task success and reliability across models, ablate harness execution evidence and external research, and test transfer to unseen benchmarks with the same model. RiseHarness treats a specific model as the unit of harness adaptation, studying how changes to shared interaction mechanisms can make existing capabilities more reliable across tasks.

Imitation learning (IL) with generative models, such as diffusion and flow matching, has enabled robots to perform complex, long-horizon tasks. However, distribution shifts from unseen environments or compounding action errors can still cause unpredictable and unsafe behavior, leading to task failure. Early failure prediction during runtime is therefore essential for deploying robots in human-centered and safety-critical environments. We propose FIPER, a general framework for Failure Prediction at Runtime for generative IL policies that does not require failure data. FIPER identifies two key indicators of impending failure: (i) out-of-distribution (OOD) observations detected via random network distillation in the policy's embedding space, and (ii) high uncertainty in generated actions measured by a novel action-chunk entropy score. Both failure prediction scores are calibrated using a small set of successful rollouts via conformal prediction. A failure alarm is triggered when both indicators, aggregated over short time windows, exceed their thresholds. We evaluate FIPER across five simulation and real-world environments involving diverse failure modes. Our results demonstrate that FIPER better distinguishes actual failures from benign OOD situations and predicts failures more accurately and earlier than existing methods. We thus consider this work an important step towards more interpretable and safer generative robot policies. Code, data and videos are available at https://tum-lsy.github.io/fiper_website.

Reward-guided search methods have demonstrated strong potential in enhancing tool-using agents by effectively guiding sampling and exploration over complex action spaces. As a core design, those search methods utilize process reward models (PRMs) to provide step-level rewards, enabling more fine-grained monitoring. However, there is a lack of systematic and reliable evaluation benchmarks for PRMs in tool-using settings. In this paper, we introduce ToolPRMBench, a large-scale benchmark specifically designed to evaluate PRMs for tool-using agents. ToolPRMBench is built on top of several representative tool-using benchmarks and converts agent trajectories into step-level test cases. Each case contains the interaction history, a correct action, a plausible but incorrect alternative, and relevant tool metadata. We respectively utilize offline sampling to isolate local single-step errors and online sampling to capture realistic multi-step failures from full agent rollouts. A multi-LLM verification pipeline is proposed to reduce label noise and ensure data quality. We conduct extensive experiments across large language models, general PRMs, and tool-specialized PRMs on ToolPRMBench. The results reveal clear differences in PRM effectiveness and highlight the potential of specialized PRMs for tool-using. Code and data will be released at https://github.com/David-Li0406/ToolPRMBench.

Autonomous improvement loops are hard to trust because the improvement process is usually external scaffolding bolted onto the agent: failures go unlogged, diagnoses cannot be replayed, and promote-or-discard decisions land in a side database rather than the agent's own history. We show that an event-sourced agent runtime removes that friction and turns controlled improvement into a first-class workflow. When the agent's state is a deterministic projection of an append-only event log, failures are recorded, a run replays exactly from its log, candidate patches scope to typed pipeline seams, gates are auditable, and every promotion or discard is itself an event. We demonstrate this with Regimes, a loop on the ActiveGraph runtime that diagnoses failed evaluations, proposes a repair at a pipeline point, and promotes it only after static checks, sandbox execution, in-sample evaluation, and held-out validation. The loop is target-agnostic: the same control flow runs against different tasks through a common interface. On LongMemEval-S the dominant failure is not retrieval but reconciliation: the evidence is already in the assembled context, yet the reader answers incorrectly. Across five seeded held-out splits, Regimes discovers reader-prompt repairs that improve final held-out accuracy by +0.05 to +0.10 in four splits and +0.01 in one over-promotion split; two splits are individually significant (seed 5 unadjusted for its sequential promotion structure), and the pooled count is descriptive only, since the splits share one 500-question pool. The durable contributions are ActiveGraph as an auditable substrate that makes controlled improvement loops tractable, the held-out-gated loop it supports, the failure-regime taxonomy routing each failure to a pipeline location (whose marginal value over an unrouted baseline is the primary open question), and the prompt-as-discovery-probe hypothesis.

Reinforcement-Learning-based post-training has recently emerged as a promising paradigm for aligning text-to-image diffusion models with human preferences. In recent studies, increasing the rollout group size yields pronounced performance improvements, indicating substantial room for further alignment gains. However, scaling rollouts on large-scale foundational diffusion models (e.g., FLUX.1-12B) imposes a heavy computational burden. To alleviate this bottleneck, we explore the integration of FP4 quantization into Diffusion RL rollouts. Yet, we identify that naive quantized pipelines inherently introduce risks of performance degradation. To overcome this dilemma between efficiency and training integrity, we propose Sol-RL (Speed-of-light RL), a novel FP4-empowered Two-stage Reinforcement Learning framework. First, we utilize high-throughput NVFP4 rollouts to generate a massive candidate pool and extract a highly contrastive subset. Second, we regenerate these selected samples in BF16 precision and optimize the policy exclusively on them. By decoupling candidate exploration from policy optimization, Sol-RL integrates the algorithmic mechanisms of rollout scaling with the system-level throughput gains of NVFP4. This synergistic algorithm-hardware design effectively accelerates the rollout phase while reserving high-fidelity samples for optimization. We empirically demonstrate that our framework maintains the training integrity of BF16 precision pipeline while fully exploiting the throughput gains enabled by FP4 arithmetic. Extensive experiments across SANA, FLUX.1, and SD3.5-L substantiate that our approach delivers superior alignment performance across multiple metrics while accelerating training convergence by up to 4.64times, unlocking the power of massive rollout scaling at a fraction of the cost.

Reinforcement Learning with Verifiable Rewards (RLVR) is an effective paradigm for improving the reasoning capabilities of large language models. However, existing RLVR methods utilize rollouts in an indiscriminate and short-horizon manner: responses of heterogeneous quality within each prompt are treated uniformly, and historical rollouts are discarded after a single use. This leads to noisy supervision, poor sample efficiency, and suboptimal policy updates. We address these issues by formulating rollout scheduling in RLVR as a contextual bandit problem and proposing a unified neural scheduling framework that adaptively selects high-value rollouts throughout training. Each rollout is treated as an arm whose reward is defined by the induced performance gain between consecutive optimization steps. The resulting scheduler supports both noise-aware intra-group selection and adaptive global reuse of historical rollouts within a single principled framework. We provide theoretical justification by deriving sublinear regret bounds and showing that enlarging the rollout buffer improves the achievable performance upper bound. Experiments on six mathematical reasoning benchmarks demonstrate consistent gains in performance and training efficiency across multiple RLVR optimization methods.

Multi-provider LLM gateways reverse proxies that route, load-balance, and rate-limit requests across foundation-model APIs have become critical production infrastructure. Yet the failure modes specific to this architectural layer remain undocumented, scattered across issue trackers and post-mortems with no unifying framework. We introduce , a two-axis taxonomy that classifies failures by their origin layer (Network/Transport, Streaming/Protocol, State/Session, Model~Behavior, Governance/Cost) and their detectability (Loud vs.\ Silent). We populate this taxonomy with five verified catalog entries sourced from public bug reports and first-hand stress testing, each accompanied by a mechanistic root-cause analysis. Three entries include standalone reproduction scripts. Our principal finding is that the most operationally severe failures are silent: they return HTTP~200, pass every standard health check, and corrupt application state in ways that require semantic-level observability to detect. Two such silent failures a concurrency race condition causing history loss and a streaming index collision corrupting tool-call payloads were discovered first-hand during evaluation campaigns.

We present Miles v0.1, a full-stack, production-ready system for frontier post-training. Building upon the clean design of slime, Miles designs each stage of the reinforcement-learning (RL) training loop around a single principle: components should be verified, clean, and customizable. With accuracy, efficiency, reliability, and scalability as first-class goals, Miles aims to make frontier-scale RL accessible to researchers and enterprises alike. This report walks through the system end to end: rollout engines built on SGLang, a trainer with a choice of two backends (NVIDIA Megatron-LM and PyTorch FSDP), and three weight-synchronization transports for different deployment topologies. Beyond full-parameter RL, Miles also supports LoRA RL, on-policy distillation, supervised fine-tuning, and true-on-policy rollout-training alignment, and extends the same architecture to diffusion models. We close with an end-to-end case study: fully asynchronous agentic RL on a GLM-5.2 744B-A40B model over terminal-use coding tasks, running on 64 NVIDIA GB300 GPUs with a median step time of 263 seconds over the first 30 measured steps. Miles is open-sourced at https://github.com/radixark/miles, with the project website at https://miles.radixark.com.

Large Language Models (LLMs) increasingly rely on reinforcement learning with verifiable rewards (RLVR) to elicit reliable chain-of-thought reasoning. However, the training process remains bottlenecked by the computationally expensive rollout stage. Existing acceleration methods-such as parallelization, objective- and data-driven modifications, and replay buffers-either incur diminishing returns, introduce bias, or overlook redundancy across iterations. We identify that rollouts from consecutive training epochs frequently share a large portion of overlapping segments, wasting computation. To address this, we propose SPEC-RL, a novel framework that integrates SPECulative decoding with the RL rollout process. SPEC-RL reuses prior trajectory segments as speculative prefixes and extends them via a draft-and-verify mechanism, avoiding redundant generation while ensuring policy consistency. Experiments on diverse math reasoning and generalization benchmarks, including GSM8K, MATH-500, OlympiadBench, MMLU-STEM, and others, demonstrate that SPEC-RL reduces rollout time by 2-3x without compromising policy quality. As a purely rollout-stage enhancement, SPEC-RL integrates seamlessly with mainstream algorithms (e.g., PPO, GRPO, DAPO), offering a general and practical path to scale RLVR for large reasoning models. Our code is available at https://github.com/ShopeeLLM/Spec-RL

Software maintenance faces a persistent challenge with crash bugs, especially across diverse release channels catering to distinct user bases. Nightly builds, favoured by enthusiasts, often reveal crashes that are cheaper to fix but may differ significantly from those in stable releases. In this paper, we emphasize the need for a data-driven solution to predict the impact of crashes happening on nightly channels once they are released to stable channels. We also list the challenges that need to be considered when approaching this problem.

In rapid release development processes, patches that fix critical issues, or implement high-value features are often promoted directly from the development channel to a stabilization channel, potentially skipping one or more stabilization channels. This practice is called patch uplift. Patch uplift is risky, because patches that are rushed through the stabilization phase can end up introducing regressions in the code. This paper examines patch uplift operations at Mozilla, with the aim to identify the characteristics of uplifted patches that introduce regressions. Through statistical and manual analyses, we quantitatively and qualitatively investigate the reasons behind patch uplift decisions and the characteristics of uplifted patches that introduced regressions. Additionally, we interviewed three Mozilla release managers to understand organizational factors that affect patch uplift decisions and outcomes. Results show that most patches are uplifted because of a wrong functionality or a crash. Uplifted patches that lead to faults tend to have larger patch size, and most of the faults are due to semantic or memory errors in the patches. Also, release managers are more inclined to accept patch uplift requests that concern certain specific components, and-or that are submitted by certain specific developers.

Large language model (LLM) agents offer a promising data-driven approach to automating Site Reliability Engineering (SRE), yet their enterprise deployment is constrained by three challenges: restricted access to proprietary data, unsafe action execution under permission-governed environments, and the inability of closed systems to improve from failures. We present AOI (Autonomous Operations Intelligence), a trainable multi-agent framework formulating automated operations as a structured trajectory learning problem under security constraints. Our approach integrates three key components. First, a trainable diagnostic system applies Group Relative Policy Optimization (GRPO) to distill expert-level knowledge into locally deployed open-source models, enabling preference-based learning without exposing sensitive data. Second, a read-write separated execution architecture decomposes operational trajectories into observation, reasoning, and action phases, allowing safe learning while preventing unauthorized state mutation. Third, a Failure Trajectory Closed-Loop Evolver mines unsuccessful trajectories and converts them into corrective supervision signals, enabling continual data augmentation. Evaluated on the AIOpsLab benchmark, our contributions yield cumulative gains. (1) The AOI runtime alone achieves 66.3% best@5 success on all 86 tasks, outperforming the prior state-of-the-art (41.9%) by 24.4 points. (2) Adding Observer GRPO training, a locally deployed 14B model reaches 42.9% avg@1 on 63 held-out tasks with unseen fault types, surpassing Claude Sonnet 4.5. (3) The Evolver converts 37 failed trajectories into diagnostic guidance, improving end-to-end avg@5 by 4.8 points while reducing variance by 35%.

Large language model (LLM) coding agents are increasingly deployed to autonomously perform software engineering tasks in terminal-based environments, making their reliability a growing concern. Existing empirical studies investigate why coding agents fail, yet they largely treat failure as a final outcome rather than a temporal process, providing limited insight into how failures emerge, evolve, and become unrecoverable. We present the first large-scale empirical study of CLI coding-agent failure trajectories, introducing a process-oriented framework that analyzes failure through its onset, evolution, and recovery across execution trajectories. We first collect 3,843 execution trajectories generated by seven frontier models across three coding-agent scaffolds (OpenHands, MiniSWE, and Terminus2) on Terminal-Bench, then carefully filter them to obtain 1,794 complete and valid trajectories for manual annotation (over 63,000 execution steps), from which we derive 14 findings spanning failure occurrence, root causes, recovery, and cross-system consistency. Our findings show that coding-agent failures are predominantly driven by epistemic errors, typically begin within the first few execution steps, and often remain hidden until recovery is no longer possible, suggesting that improving coding-agent reliability requires earlier validation and intervention rather than relying solely on final-outcome evaluation.

Test-time reinforcement learning generates multiple candidate answers via repeated rollouts and performs online updates using pseudo-labels constructed by majority voting. To reduce overhead and improve exploration, prior work introduces tree structured rollouts, which share reasoning prefixes and branch at key nodes to improve sampling efficiency. However, this paradigm still faces two challenges: (1) high entropy branching can trigger rollout collapse, where the branching budget concentrates on a few trajectories with consecutive high-entropy segments, rapidly reducing the number of effective branches; (2) early pseudo-labels are noisy and biased, which can induce self-reinforcing overfitting, causing the policy to sharpen prematurely and suppress exploration. To address these issues, we propose Entropy Confidence Hybrid Group Relative Policy Optimization (ECHO). During rollout, ECHO jointly leverages local entropy and group level confidence to adaptively control branch width, and further introduces online confidence-based pruning to terminate persistently low confidence branches, avoiding high entropy traps and mitigating collapse. During policy updates, ECHO employs confidence adaptive clipping and an entropy confidence hybrid advantage shaping approach to enhance training robustness and mitigate early stage bias. Experiments demonstrate that ECHO achieves consistent gains on multiple mathematical and visual reasoning benchmarks, and generalizes more effectively under a limited rollout budget.

Software developers attempt to reproduce software bugs to understand their erroneous behaviours and to fix them. Unfortunately, they often fail to reproduce (or fix) them, which leads to faulty, unreliable software systems. However, to date, only a little research has been done to better understand what makes the software bugs non-reproducible. In this paper, we conduct a multimodal study to better understand the non-reproducibility of software bugs. First, we perform an empirical study using 576 non-reproducible bug reports from two popular software systems (Firefox, Eclipse) and identify 11 key factors that might lead a reported bug to non-reproducibility. Second, we conduct a user study involving 13 professional developers where we investigate how the developers cope with non-reproducible bugs. We found that they either close these bugs or solicit for further information, which involves long deliberations and counter-productive manual searches. Third, we offer several actionable insights on how to avoid non-reproducibility (e.g., false-positive bug report detector) and improve reproducibility of the reported bugs (e.g., sandbox for bug reproduction) by combining our analyses from multiple studies (e.g., empirical study, developer study).

Failure attribution, i.e., identifying the responsible agent and decisive step of a failure, is particularly challenging in LLM-based multi-agent systems (MAS) due to their natural-language reasoning, nondeterministic outputs, and intricate interaction dynamics. A reliable benchmark is therefore essential to guide and evaluate attribution techniques. Yet existing benchmarks rely on partially observable traces that capture only agent outputs, omitting the inputs and context that developers actually use when debugging. We argue that failure attribution should be studied under full execution observability, aligning with real-world developer-facing scenarios where complete traces, rather than only outputs, are accessible for diagnosis. To this end, we introduce TraceElephant, a benchmark designed for failure attribution with full execution traces and reproducible environments. We then systematically evaluate failure attribution techniques across various configurations. Specifically, full traces improve attribution accuracy by up to 76\% over a partial-observation counterpart, confirming that missing inputs obscure many failure causes. TraceElephant provides a foundation for follow-up failure attribution research, promoting evaluation practices that reflect real-world debugging and supporting the development of more transparent MASs.

AI agents often fail in ways that are difficult to localize because executions are probabilistic, long-horizon, multi-agent, and mediated by noisy tool outputs. We address this gap by manually annotating failed agent runs and release a novel benchmark of 115 failed trajectories spanning structured API workflows, incident management, and open-ended web/file tasks. Each trajectory is annotated with a critical failure step and a category from a grounded-theory derived, cross-domain failure taxonomy. To mitigate the human cost of failure attribution, we present AGENTRX, an automated domain-agnostic diagnostic framework that pinpoints the critical failure step in a failed agent trajectory. It synthesizes constraints, evaluates them step-by-step, and produces an auditable validation log of constraint violations with associated evidence; an LLM-based judge uses this log to localize the critical step and category. Our framework improves step localization and failure attribution over existing baselines across three domains.

Knowledge distillation transfers a teacher's competence to a small student but is brittle in the small-student regime: forcing the student to imitate logits from a much larger teacher concentrates it on the teacher's sharpest modes, hurting generalization on benchmark families beyond the training corpus. Reinforcement learning (RL) avoids logit imitation by training on the student's own rollouts. However, on questions where every rollout fails-yielding zero advantage and being silently discarded-injecting a stronger teacher's response into the policy gradient breaks the on-policy assumption and induces drift. We introduce Zone of Proximal Policy Optimization (ZPPO), inspired by Vygotsky's zone of proximal development, which keeps the teacher inside the prompt rather than the policy gradient. On hard questions, ZPPO constructs two reformulated prompts: a Binary Candidate-included Question (BCQ) pairs one correct teacher response with one incorrect student response as anonymized candidates the student must discriminate, and a Negative Candidate-included Question (NCQ) aggregates the student's wrong rollouts into a single prompt to surface their shared failure modes. A prompt replay buffer recirculates each hard question until it either graduates-the student's mean rollout accuracy on it reaches half- or is FIFO-evicted under finite capacity, amplifying BCQ and NCQ inside the student's current zone of proximal development. On the Qwen3.5 family at four student scales (0.8B-9B) with a 27B teacher, post-trained as vision-language models and evaluated on a 31-benchmark suite (16 VLM, 10 LLM, 5 Video), ZPPO outperforms off/on-policy distillation and GRPO, with the largest gains at the smallest scale.

Tool-using agents make consequential changes to external state, yet correct outcomes do not guarantee that their actions were supported by evidence established beforehand. We study where this evidence-to-action chain breaks as agents move from deciding whether to act to executing single actions and dependent workflows. Across ten model-harness configurations, strong static action assessment can coexist with much weaker interactive execution. Failures often begin before execution: agents stop with incomplete investigation or act before required evidence is established. Once required evidence is obtained, single-action execution is usually reliable, while multi-action workflows additionally expose unresolved prerequisites and incomplete execution. For this analysis, we introduce SafeActBench, comprising 656 cases across six operational domains and five protocols that progress from static action judgment and investigated non-action to single- and multi-action workflows. A provenance-bound Evidence Ledger and deterministic trajectory evaluator track what information was established, when actions occurred, and whether downstream dependencies were satisfied. These results show that failures arise not only from missing information, but also from how agents use established evidence when deciding and executing actions.

Autonomous coding agents built on large language models (LLMs) are rapidly being integrated into development workflows, yet their operational safety properties remain poorly understood beyond evaluations of explicitly malicious inputs. In practice, high-impact failures arise during benign, goal-directed use through environment breakage, fabricated success reports, etc. that current benchmarks do not capture. What categories of operational safety failures actually occur when coding agents are used for everyday development tasks and what is their impact? We present an incident-driven empirical study grounded in two complementary evidence streams. We screen 68,816 papers from 22 premier venues, curating 185 safety-relevant studies, and mine 16,586 GitHub issues from widely deployed LLM-powered coding tools, manually confirming 547 genuine safety failures. Applying systematic open coding over both corpora, we derive a multi-dimensional safety taxonomy of 33 operational risk types organized across seven dimensions, and annotate each incident with contributing factors, task context, severity, and downstream impact. Our findings show that coding-agent failures are often severe, with 326 of 547 incidents rated high or critical. The dominant risks are constraint violations, destructive operations, authorization bypasses, and deception, and over 65% of incidents arise in bug fixing and setup or configuration, patterns largely missing from prior literature. These results have direct implications for SE tool designers and benchmark developers: guardrails must go beyond adversarial-prompt defenses to enforce environmental constraints, failure transparency, and safe-halt behaviors.

Large Language Model (LLM) agents, which integrate planning, memory, reflection, and tool-use modules, have shown promise in solving complex, multi-step tasks. Yet their sophisticated architectures amplify vulnerability to cascading failures, where a single root-cause error propagates through subsequent decisions, leading to task failure. Current systems lack a framework that can comprehensively understand agent error in a modular and systemic way, and therefore fail to detect these errors accordingly. We address this gap with three contributions. First, we introduce the AgentErrorTaxonomy, a modular classification of failure modes spanning memory, reflection, planning, action, and system-level operations. Second, we construct AgentErrorBench, the first dataset of systematically annotated failure trajectories from ALFWorld, GAIA, and WebShop, grounding error analysis in real-world agent rollouts. Third, we propose AgentDebug, a debugging framework that isolates root-cause failures and provides corrective feedback, enabling agents to recover and iteratively improve. Experiments on AgentErrorBench show that AgentDebug achieves 24% higher all-correct accuracy and 17% higher step accuracy compared to the strongest baseline. Beyond detection, the targeted feedback generated by AgentDebug enables LLM agents to iteratively recover from failures, yielding up to 26% relative improvements in task success across ALFWorld, GAIA, and WebShop. These results establish principled debugging as a pathway to more reliable and adaptive LLM agents. The code and data will be available at https://github.com/ulab-uiuc/AgentDebug

Agentic systems consisting of multiple LLM-driven agents coordinating through tools and structured interactions, are increasingly deployed for complex reasoning and problem-solving tasks. At the same time, emerging low-code and template-based agent development platforms (e.g., Dify) enable users to rapidly build and orchestrate agentic systems, which we refer to as platform-orchestrated agentic systems. However, these systems are also fragile and it remains unclear how to systematically identify their potential failure root cause. This paper presents a study of root cause identification of these platform-orchestrated agentic systems. To support this initiative, we construct a dataset AgentFail containing 307 failure logs from ten agentic systems, each with fine-grained annotations linking failures to their root causes. We additionally utilize counterfactual reasoning-based repair strategy to ensure the reliability of the annotation. Building on the dataset, we develop a taxonomy that characterizes failure root causes and analyze their distribution across different platforms and task domains. Furthermore, we introduce a benchmark that leverages LLMs for automatically identifying root causes, in which we also utilize the proposed taxonomy as guidance for LLMs. Results show that the taxonomy can largely improve the performance, thereby confirming its utility. Nevertheless, the accuracy of root cause identification reaches at most 33.6%, which indicates that this task still remains challenging. In light of these results, we also provide actionable guidelines for building such agentic systems. In summary, this paper provides a reliable dataset of failure root cause for platform-orchestrated agentic systems, corresponding taxonomy and benchmark, which serves as a foundation for advancing the development of more reliable agentic systems.

Self-improving AI agents are designed to learn from their mistakes. We show they can also hallucinate mistakes that never happened. We study this failure mode in automated harness optimization, where an LLM-based proposer edits an agent's scaffold, including prompts, parsers, filters, validators and guardrails, to eliminate observed failures. But this process rarely asks first: was there a real failure to fix? We introduce the Counterfactual Fabrication Lab, a deterministic micro-lab where the correct action is known: do nothing. The lab plants a candidate guardrail for a failure class that provably never occurs, presents only legal episodes, and uses a byte-exact oracle to check every cited violation. The proposer behaves as expected on real violations and abstains on featureless legal input. Yet when the legal input contains a harmless pattern resembling a familiar game rule, it invents a failure: in 15/60 runs, versus 0/60 on featureless input, it enables the nonexistent-rule guardrail and cites a violation the oracle refutes. The effect is structured, not indiscriminate. In single-shot proposals it appears only when three conditions coincide: a rule-shaped pattern, an open-ended rule set and an instruction that presupposes failures. Removing any of these conditions eliminates the fabrication. Because the invented guardrail changes no true outcome and cannot improve an already-perfect suppression score, the phenomenon is neither reward hacking nor over-refusal. It is a phantom guardrail: a fix for a failure that never happened, invisible to suppression-only acceptance. Inside an add-only accept loop it re-enters even without the failure-presupposing instruction, the loop's keep-adding role supplying the demand the instruction supplied in single shot, and once in it stays. We present the Counterfactual Fabrication Lab for measuring fabricated failures in self-improving agent harnesses.

The deployment of embodied agents in self-driving laboratories could accelerate scientific discovery, yet their reliability is constrained by the irreversible and safety-critical nature of chemical experiments. Progress is further hindered by scarce failure data and the lack of fine-grained evaluation protocols. To address these challenges, we introduce LabRobFail, a failure-centric framework for learning and evaluating robotic failure analysis in chemical laboratories. LabRobFail-Sim injects controllable failures at the control, physics, and semantic levels, enabling the construction of LabRobFail-Data, which contains over 20,000 trajectories across 70+ task scenarios, five failure categories, and 11 fine-grained failure types. LabRobFail-Bench evaluates six capabilities spanning task understanding, failure detection, temporal localization, severity assessment, failure classification, and actionable correction. We further develop LabRobFail-VLM, a domain-specialized vision-language model that generates structured failure diagnoses and recovery instructions. On seen environments, it achieves 90.83% failure-detection accuracy and 77.21% temporal-localization accuracy, substantially outperforming general-purpose VLMs. When integrated as a real-time supervisor, it improves downstream task success rates by 4-16 percentage points, demonstrating the value of fine-grained failure understanding for closed-loop recovery and reliable laboratory autonomy. Our code and data are available at https://github.com/Su-ISE-2001/SciRobo

Large-scale AI training is now fundamentally a distributed systems problem, and hardware failures have become routine operating conditions rather than rare exceptions. Public operational evidence from production training clusters, however, remains scarce. This technical report presents an empirical analysis of a 63-node NVIDIA B200 production cluster (504 GPUs), using 55 days of Prometheus time-series data and 73 days of operational logs covering 224 multi-node training sessions. The cluster operates within a cross-organizational environment in which five parties (SKT, Upstage, Lablup, NVIDIA Korea, and VAST Data) share a unified monitoring pipeline. This arrangement enabled joint diagnosis of a 60-node-scale storage I/O bottleneck that did not appear at 2-4-node scale, a production-scale phenomenon no single team could isolate alone. Drawing on a months-long pre-training campaign, we perform three quantitative analyses yielding four findings. First, statistical analysis over 751 Prometheus metrics and 10 XID-identified GPU failures achieves a 10/10 detection rate (2/10 pre-XID) at ~0.84 false positives per day. No single metric is consistently dominant across failure types, motivating a multi-signal detection strategy. Second, profiling 523 checkpoint events along the GPU VRAM to NFS path attributes the "bandwidth paradox" (1.4-10.4% utilization of 200 Gbps RoCE) to saturation of the 128-slot NFS RPC layer. Third, multi-node failure response shows concentrated exclusions (top 3 of 63 nodes account for >50% of all exclusions) and an auto-retry chain success rate of 33.3% over 12 chains (73 attempts), 2.7x the 12.5% manual recovery rate; the median retry interval is 11 min (IQR 10-11). All analyses are grounded in production infrastructure providing session-level workload management, GPU-centric scheduling, and unified observability.

Safety benchmark scores provide incomplete evidence of deployment readiness: aligned language models often adhere to rigid rules even when a situational update flips which action is safe. We term this failure brittle safety. To diagnose it, we introduce context-flip evaluation, testing 12 models across a safety benchmark (PacifAIst) and two commonsense controls using paired variants where the nominally safe action produces harm. Three findings emerge. First, brittle safety is safety-specific: all 12 models exhibit a safety-commonsense gap (mean +17.4 pp). Baseline accuracy fails to predict brittleness: among models above 90% baseline accuracy, brittleness rates range from 13.7% to 90.0%. Second, failures stem from policy override rather than miscomprehension: despite acknowledging the context change in every case, models persist via three distinct mechanisms that vary by update type and model family. Third, on a hand-audited probe of catastrophic consequence-flip scenarios, standard action-level guardrails catch none, while a state-aware validator catches all without false alarms on correct interventions. This indicates that action-level content moderation is systematically blind to consequence-flips, motivating state-aware architectural alternatives. We release our protocol, perturbed benchmarks, and deployment probe.

Large language models are increasingly deployed as autonomous agents for multi-step tasks in executable environments, yet their ability to perform realistic operations research (OR) work remains unclear. Existing OR evaluations often decouple modeling from solving, rely on pre-formalized or text-only instances, and rarely test the full workflow from operational artifacts to validated decisions. In this work, we introduce ORAgentBench, an execution-grounded benchmark for evaluating autonomous agents on challenging end-to-end operations research tasks. It contains 107 human-reviewed tasks across diverse operational scenarios, each packaged in an isolated environment with a natural-language brief, multi-file data, configuration artifacts, and a required submission schema. Agents must write and run solution code, and their submissions are evaluated by hidden validators for schema validity, hard-constraint feasibility, and normalized objective quality. Experiments with fourteen frontier agent-model configurations show that current agents remain far from reliable OR practice. The best agent passes only 35.51% of all tasks and 20.59% of hard tasks, and many feasible submissions still fall below the required quality threshold. Failure analysis further shows that errors are dominated by strategic weaknesses, including missed operational rules, brittle formulations, weak feasible-solution construction, and insufficient solution improvement. OR-specific procedural skills increase hard-task feasibility, but do not reliably improve solution quality or pass rate. These results suggest that progress in OR agents requires moving beyond plausible optimization code toward dependable, high-quality operational decision-making.

Large language model-powered coding agents are increasingly used to modify existing code repositories, for example, by adding features or fixing bugs. Yet existing repository-level benchmarks typically evaluate only whether the final patch passes tests. Satisfying a user request requires a long chain of interdependent reasoning and decisions: an agent must recover explicit and implicit requirements, formulate a repository-grounded implementation plan, and translate it into correct code. A pass/fail outcome cannot characterize how an unsuccessful trajectory diverges from the requirements and implementation process needed for a correct patch. To address this gap, we introduce SWE-RPG, a repository-level benchmark that combines executable patch evaluation with validated ground-truth references (GTs) for (1) Requirement Clarification and (2) Implementation Planning. These intermediate GTs support retrospective, GT-aligned diagnosis of complete coding-agent trajectories across clarification, planning, code generation, and artifact submission. SWE-RPG comprises 163 tasks from 31 Python and Java repositories, including 113 bug fixes and 50 feature additions. We evaluate 3 coding agents, including Claude Code, Codex, and OpenCode, with 6 large language model backends, including Claude-Sonnet-5 and GPT-5.6-Terra. Results show that the evaluated popular coding agents still struggle to implement user requests in existing repositories, achieving an average resolved rate of only 31.5% on SWE-RPG. Intermediate-GT diagnosis further identifies implicit requirement recovery as the main bottleneck, accounting for 24.5%--46.0% of agent runs. This result suggests implicit-requirement recovery as a key candidate direction for improving coding agents. The benchmark data and evaluation code are available at https://github.com/Xin-Zhou-smu/SWE-RPG-Bench.

Asynchronous reinforcement learning improves rollout throughput for large language model agents by decoupling sample generation from policy optimization, but it also introduces a critical failure mode for PPO-style off-policy correction. In heterogeneous training systems, the total importance ratio should ideally be decomposed into two semantically distinct factors: a training--inference discrepancy term that aligns inference-side and training-side distributions at the same behavior-policy version, and a policy-staleness term that constrains the update from the historical policy to the current policy. We show that practical asynchronous pipelines with delayed updates and partial rollouts often lose the required historical training-side logits, or old logits. This missing-old-logit problem entangles discrepancy repair with staleness correction, breaks the intended semantics of decoupled correction, and makes clipping and masking thresholds interact undesirably. To address this issue, we study both exact and approximate correction routes. We propose three exact old-logit acquisition strategies: snapshot-based version tracking, a dedicated old-logit model, and synchronization via partial rollout interruption, and compare their system trade-offs. From the perspective of approximate correction, we focus on preserving the benefits of decoupled correction through a more appropriate approximate policy when exact old logits cannot be recovered at low cost, without incurring extra system overhead. Following this analysis, we adopt a revised PPO-EWMA method, which achieves significant gains in both training speed and optimization performance. Code at https://github.com/millioniron/ROLL.

An agentic system issues several structurally different kinds of LLM calls. It routes intent, classifies actions, grounds language in a device registry, plans multi-agent pipelines and writes the Python code those pipelines run. The difficulty of these call sites varies by an order of magnitude, yet in practice a single model, chosen for the hardest site, serves all of them. In this work, we evaluate 9 models from 0.8B to a frontier hosted model across the five call sites of a deployed open-source home-automation framework (Wactorz), using its unmodified production prompts and two real Home Assistant installations (280 cases, 2520 scored calls). We find that capability is not ordered the same way at every site, and that larger models are not uniformly better: one 4B model is worse than its 2B sibling at grounded actuation. Paired testing shows the best local model to be statistically indistinguishable from both hosted models at four of five sites. Only code generation separates them, against a small hosted model (p = 0.039) as well as a frontier one (p = 0.002). Aggregate accuracy also hides a safety failure specific to actuation, where small models resolve the accuracy/refusal trade-off in degenerate ways: one model (Gemma4 E2B) actuates on 87.2% of requests for devices the site does not own, while another refuses every request it receives. Routing each site to its best local model reaches 91.8% against 95.4% at no per-call cost. In a live deployment judged by a user, hosting only the two generative sites matches hosting everything (39/43 against 39/43) for 28% of the spend, and the actuation gap the benchmark predicted appears as exactly one case in twenty-six. Benchmark, harness and all records are released at https://github.com/waldiez/slm-callsite-eval.

Language-model agents are increasingly deployed under standing instructions: a system prompt, a policy file, or a skills document is placed in context, and the agent is trusted to let that document govern every action that follows. Existing benchmarks rarely test this deployment pattern directly; they measure whether an agent can complete a task, not whether a long, binding policy document constrains its behavior over an extended tool-use horizon. We present HANDBOOK_md, a benchmark of 65 agentic tasks modeled on how employees follow company handbooks. Each task places an agent in a self-contained company environment (a file workspace with mock email, chat, calendar, issue-tracking, and commerce services exposed over the Model Context Protocol) and instructs it to carry out routine professional work governed by an expert-written standard operating procedure of 20-124 pages. Tasks span five domains (finance, medical billing, insurance, logistics, and HR) and 10 fictional companies. To resist memorization, every task modifies one of 10 base handbooks, altering the specific rules and thresholds on which grading depends, so no two tasks share the same set of policies. Grading is fully deterministic: each task carries a rubric of programmatic criteria (824 in total) that check both that required actions occurred and that prohibited actions did not. Under strict grading, where a trial passes only if every criterion is satisfied, the strongest evaluated model passes 36.2% of trials, and most frontier models remain below 25%. Failures follow consistent patterns: agents let a plausible but unauthorized in-environment request override the standing policy, perform a required check and then act against its result, lose rule details over long horizons, and report compliance they did not achieve. We release the tasks, environments, and evaluation harness.

Behavior-cloned diffusion policies are expressive but remain vulnerable to covariate shift: small deviations from demonstrated states can compound into task failure. Existing methods address this either by expanding the training distribution through expert corrections or synthetic augmentation, or by steering a frozen policy at test time with guidance from a learned model. The former can be expensive or assumption-dependent, while the latter discards the corrected trajectories after execution. We introduce ReGuide, a self-improving framework that treats guided rollouts as reusable on-policy recovery data. ReGuide first uses Phase-Conditioned Guidance (PCG) to generate corrective rollouts: it constructs phase-specific latent targets, applies guidance only in the drifted-but-recoverable regime, and guides through the estimated clean action to match the dynamics model's training distribution. Successful guided rollouts are then absorbed back into the policy through ReGuide-FT, which fine-tunes the current checkpoint, or ReGuide-FS, which retrains from scratch on the augmented dataset; the two can also be composed and iterated. On Robomimic Can, Square, Transport, and Tool Hang, ReGuide improves base-policy success by 1.3--7.7times, outperforms LPB in the test-time-only setting, and matched-data ablations show that the gains come from guided recovery data rather than additional rollouts alone.

Offline reinforcement learning aims to train agents from pre-collected datasets. However, this comes with the added challenge of estimating the value of behaviors not covered in the dataset. Model-based methods offer a potential solution by training an approximate dynamics model, which then allows collection of additional synthetic data via rollouts in this model. The prevailing theory treats this approach as online RL in an approximate dynamics model, and any remaining performance gap is therefore understood as being due to dynamics model errors. In this paper, we analyze this assumption and investigate how popular algorithms perform as the learned dynamics model is improved. In contrast to both intuition and theory, if the learned dynamics model is replaced by the true error-free dynamics, existing model-based methods completely fail. This reveals a key oversight: The theoretical foundations assume sampling of full horizon rollouts in the learned dynamics model; however, in practice, the number of model-rollout steps is aggressively reduced to prevent accumulating errors. We show that this truncation of rollouts results in a set of edge-of-reach states at which we are effectively ``bootstrapping from the void.'' This triggers pathological value overestimation and complete performance collapse. We term this the edge-of-reach problem. Based on this new insight, we fill important gaps in existing theory, and reveal how prior model-based methods are primarily addressing the edge-of-reach problem, rather than model-inaccuracy as claimed. Finally, we propose Reach-Aware Value Learning (RAVL), a simple and robust method that directly addresses the edge-of-reach problem and hence - unlike existing methods - does not fail as the dynamics model is improved. Code open-sourced at: github.com/anyasims/edge-of-reach.

Reinforcement learning drives recent advances in LLM reasoning and agentic capabilities, yet current approaches struggle with both exploration and exploitation. Exploration suffers from low success rates on difficult tasks and high costs of repeated rollouts from scratch. Exploitation suffers from coarse credit assignment and training instability: Trajectory-level rewards penalize valid prefixes for later errors, and failure-dominated groups overwhelm the few positive signals, leaving optimization without constructive direction. To this end, we propose R^3L, Reflect-then-Retry Reinforcement Learning with Language-Guided Exploration, Pivotal Credit, and Positive Amplification. To synthesize high-quality trajectories, R^3L shifts from stochastic sampling to active synthesis via reflect-then-retry, leveraging language feedback to diagnose errors, transform failed attempts into successful ones, and reduce rollout costs by restarting from identified failure points. With errors diagnosed and localized, Pivotal Credit Assignment updates only the diverging suffix where contrastive signals exist, excluding the shared prefix from gradient update. Since failures dominate on difficult tasks and reflect-then-retry produces off-policy data, risking training instability, Positive Amplification upweights successful trajectories to ensure positive signals guide the optimization process. Experiments on agentic and reasoning tasks demonstrate 5\% to 52\% relative improvements over baselines while maintaining training stability. Our code is released at https://github.com/shiweijiezero/R3L.

Large language models (LLMs) can call tools effectively, yet they remain brittle in multi-turn execution: following a tool call error, smaller models often degenerate into repetitive invalid re-invocations, failing to interpret error feedback and self-correct. This brittleness hinders reliable real-world deployment, where the execution errors are inherently inevitable during tool interaction procedures. We identify a key limitation of current approaches: standard reinforcement learning (RL) treats errors as sparse negative rewards, providing no guidance on how to recover, while pre-collected synthetic error-correction datasets suffer from distribution mismatch with the model's on-policy error modes. To bridge this gap, we propose Fission-GRPO, a framework that converts execution errors into corrective supervision within the RL training loop. Our core mechanism fissions each failed trajectory into a new training instance by augmenting it with diagnostic feedback from a finetuned Error Simulator, then resampling recovery rollouts on-policy. This enables the model to learn from the precise errors it makes during exploration, rather than from static, pre-collected error cases. On the BFCL v4 Multi-Turn, Fission-GRPO improves the error recovery rate of Qwen3-8B by 5.7% absolute, crucially, yielding a 4% overall accuracy gain (42.75% to 46.75%) over GRPO and outperforming specialized tool-use agents.

Interactive agent benchmarks face a tension between scalable construction and realistic workflow evaluation. Hand-authored tasks are expensive to extend and revise, while static prompt evaluation misses failures that only appear when agents operate over persistent state. Existing interactive benchmarks have advanced agent evaluation significantly, but most initialize tasks from clean state and do not systematically test how agents handle pre-existing partial, stale, or conflicting artifacts. We present ClawForge, a generator-backed benchmark framework for executable command-line workflows under state conflict. The framework compiles scenario templates, grounded slots, initialized state, reference trajectories, and validators into reproducible task specifications, and evaluates agents step by step over persistent workflow surfaces using normalized end state and observable side effects rather than exact trajectory matching. We instantiate this framework as the ClawForge-Bench (17 scenarios, 6 ability categories). Results across seven frontier models show that the best model reaches only 45.3% strict accuracy, wrong-state replacement remains below 17\% for all models, and the widest model separation (17% to 90%) is driven by whether agents inspect existing state before acting. Partial-credit and step-efficiency analyses further reveal that many failures are near-miss closures rather than early breakdowns, and that models exhibit qualitatively different failure styles under state conflict.

Despite rapid progress in LLM agents, performance on long-horizon, tool-using tasks remains fragile. To better understand this fragility, we ask a simple question: do all actions contribute equally to failure? Analyzing execution traces on τ-Bench (Airline/Retail) and SWE-Bench Verified, we decompose trajectories into mutating (environment-changing) vs.\ non-mutating steps and formalize decisive deviations, earliest action, level divergences that flip success to failure. A logistic regression reveals that each additional deviation in a mutating action reduces the odds of success by upto 92% on Airline and upto 96% on Retail for SoTA models. In contrast, deviations in non-mutating actions have little to no effect. Errors also grow with context length as agents drift from role and act on stale constraints. Motivated by these observations, we introduce , a model-agnostic, gradient-free, test-time safeguard that (i) adds mutation-gated verification, (ii) injects Targeted Reflection before mutating steps, and (iii) performs block-based context cleaning. delivers consistent gains, e.g., Qwen3-Thinking: +28\% relative on Airline, +11\% on Retail, and +7\% on SWE-Bench Verified; Claude: +9\%/+7\%. We further identify ceiling effects in τ-Bench, where annotation errors and underspecified tasks artificially cap model performance. To address this, we release τ-Bench Verified, which restores benchmark headroom through targeted revisions. Our results argue for action-level analysis, targeted safeguards, and reliable evaluations as prerequisites for robust multi-turn agents.

Agent Skills package reusable instructions and assets for tool-using language-model agents. Progressive loading creates failure boundaries poorly represented by session-, model-, or tool-centric traces: a Skill can be discovered but not activated, activated without instructions, or appear successful without an independently verified outcome. We present Skill Runtime Intelligence, a passive runtime-intelligence system that reconstructs supported Skill-lifecycle stages across heterogeneous harnesses while preserving unsupported stages as unknown. Its Run Panorama separates immutable events, deterministic relations, inferred diagnoses, and controlled outcomes with four evidence grades; optional trace import and OTLP/HTTP export support existing observability deployments. Across six frozen repository profiles, three coding agents, and seven clean or fault-injected conditions, all 126 executions preserve source worktrees and each correlates to exactly one source session. Yet adapters expose three distinct semantics: no Skill runs; complete runs but no failure-like events; or failure-like events in every operational-failure and clean session. In a seven-template diagnostic study, semantic aliases and Panorama localize the same six non-clean boundaries but differ in exact/status behavior; both Raw views emit a failure status on all 18 clean cases, while Panorama emits none. A known-rule graph conforms to 126/126 frozen contracts, whereas a second model completes only 228/378 calls. These observations motivate executable adapter qualification and show that event presence is not boundary fidelity, composite exact scores mask distinct errors, and model explanations must not overwrite deterministic facts.

As software systems grow increasingly intricate, Artificial Intelligence for IT Operations (AIOps) methods have been widely used in software system failure management to ensure the high availability and reliability of large-scale distributed software systems. However, these methods still face several challenges, such as lack of cross-platform generality and cross-task flexibility. Fortunately, recent advancements in large language models (LLMs) can significantly address these challenges, and many approaches have already been proposed to explore this field. However, there is currently no comprehensive survey that discusses the differences between LLM-based AIOps and traditional AIOps methods. Therefore, this paper presents a comprehensive survey of AIOps technology for failure management in the LLM era. It includes a detailed definition of AIOps tasks for failure management, the data sources for AIOps, and the LLM-based approaches adopted for AIOps. Additionally, this survey explores the AIOps subtasks, the specific LLM-based approaches suitable for different AIOps subtasks, and the challenges and future directions of the domain, aiming to further its development and application.

Structured pruning is a hardware-friendly way to compress LLMs, but it is mostly validated on multiple-choice recognition tasks, while the same compressed checkpoints can collapse on the free-form generation that deployment actually requires. Two observations trace this gap. First, greedy pass@1 nearly vanishes after compression, yet pass@k recovers substantially under repeated sampling: useful generations are demoted, not erased. Second, the recoverable regime fails mainly through suffix repetition. Recovery should therefore train on the compressed model's own on-policy states with dense token-level supervision, which On-Policy Distillation (OPD) provides by reusing the pre-compression model as a frozen teacher. However, long on-policy rollouts spend early recovery budget on low-information repetitive suffixes, delaying loss descent. To mitigate this waste, we propose \shortopd, a short-to-long OPD schedule that detects teacher-confirmed repetitive suffixes, treats the surviving prefix as each rollout's effective length, and allocates future rollout budgets to the effective lengths the policy can currently use. Across math, code, and open-ended generation, \shortopd\ raises the compressed model's score to about 9times its unrecovered value and 1.6--4.4times standard recovery recipes (SFT w/o KD, KD, and SeqKD), and it matches a fixed 8192-token rollout horizon within two points using a quarter of the training time (8.5 vs.\ 35.9 hours) and 71% fewer rollout tokens. We hope this recipe helps move structured pruning beyond marginal gains on perplexity and multiple-choice benchmarks, a step closer to deployment-ready generation quality.

Large language models (LLMs) are being rapidly integrated into decision-support tools, automation workflows, and AI-enabled software systems. However, their behavior in production environments remains poorly understood, and their failure patterns differ fundamentally from those of traditional machine learning models. This paper presents a system-level taxonomy of fifteen hidden failure modes that arise in real-world LLM applications, including multi-step reasoning drift, latent inconsistency, context-boundary degradation, incorrect tool invocation, version drift, and cost-driven performance collapse. Using this taxonomy, we analyze the growing gap in evaluation and monitoring practices: existing benchmarks measure knowledge or reasoning but provide little insight into stability, reproducibility, drift, or workflow integration. We further examine the production challenges associated with deploying LLMs - including observability limitations, cost constraints, and update-induced regressions - and outline high-level design principles for building reliable, maintainable, and cost-aware LLM systems. Finally, we outline high-level design principles for building reliable, maintainable, and cost-aware LLM-based systems. By framing LLM reliability as a system-engineering problem rather than a purely model-centric one, this work provides an analytical foundation for future research on evaluation methodology, AI system robustness, and dependable LLM deployment.

In real world software development, improper or missing exception handling can severely impact the robustness and reliability of code. Exception handling mechanisms require developers to detect, capture, and manage exceptions according to high standards, but many developers struggle with these tasks, leading to fragile code. This problem is particularly evident in open source projects and impacts the overall quality of the software ecosystem. To address this challenge, we explore the use of large language models (LLMs) to improve exception handling in code. Through extensive analysis, we identify three key issues: Insensitive Detection of Fragile Code, Inaccurate Capture of Exception Types, and Distorted Handling Solutions. These problems are widespread across real world repositories, suggesting that robust exception handling practices are often overlooked or mishandled. In response, we propose Seeker, a multi agent framework inspired by expert developer strategies for exception handling. Seeker uses agents: Scanner, Detector, Predator, Ranker, and Handler to assist LLMs in detecting, capturing, and resolving exceptions more effectively. Our work is the first systematic study on leveraging LLMs to enhance exception handling practices, providing valuable insights for future improvements in code reliability.