Agregador

arXiv:2606.31422v3 Announce Type: replace Abstract: Long-horizon language agents enable autonomous decision-making in complex environments by planning and executing sequences of interdependent actions. To act reliably, these agents must maintain an accurate world model that tracks the environment's evolving state and supports subsequent decisions. However, as interactions accumulate, stored beliefs can become outdated or incorrect, causing agents to plan and act on assumptions that no longer reflect the environment. Such errors are difficult to resolve through additional reasoning or self-reflection alone, because the information needed for correction must be obtained from the environment. Actively probing the environment therefore offers a direct way to verify and correct these beliefs. To this end, we introduce ENVPROBE, a framework that enables agents to actively explore their environments and correct inaccurate beliefs during long-horizon execution. Specifically, ENVPROBE prioritizes belief fields for verification based on task relevance and potential errors. It then selectively probes the environment to retrieve the actual value of a chosen field and incorporates the observation into the agent's world model. With the updated beliefs, the agent revises its plan and continues execution based on the newly acquired environmental information. Extensive experiments across three controlled environments demonstrate that ENVPROBE substantially improves terminal world-state accuracy compared with agents operating without environment probing, leading to more reliable long-horizon decision-making. These results demonstrate the importance of actively verifying and correcting beliefs through environmental interactions for maintaining accurate world models in long-horizon language agents.

arXiv:2606.27806v4 Announce Type: replace Abstract: Language agents, such as ReAct, Reflexion, and RAP, implicitly predict how the world will change as they plan and act, using imagined state transitions to guide their decisions. These predictions are encoded in the agents' backbone LLMs as internal beliefs about the environment, effectively forming an implicit world model for multi-step planning. However, hallucinated state information can become embedded in this world model, leading to persistent inaccuracies that influence subsequent decisions. Traditional parametric world models, such as Dreamer and MuZero, learn environment dynamics through explicit transition models, but their limited model capacity makes them difficult to apply directly to modern language-agent planning. We study this tradeoff in graph-structured planning environments and seek to combine the strengths of both paradigms. We propose Hybrid World-Model Planning (HYBRID-WM), which preserves the language agent's semantic planning capabilities while leveraging a lightweight parametric world model for more accurate transition predictions. Specifically, the parametric model constructs a compact planning skeleton to guide action generation while maintaining consistency between the agent's imagined transitions and the model's predictions. Experiments across four graph-structured planning benchmarks demonstrate that HYBRID-WM consistently improves task success and reduces hallucinated-state propagation with modest computational overhead. These findings highlight the effectiveness of combining semantic planning with lightweight parametric grounding for reliable long-horizon language-agent planning.

arXiv:2606.07000v2 Announce Type: replace Abstract: Recent post-training methods, particularly Reinforcement Learning with Verifiable Rewards (RLVR), have significantly enhanced the reasoning ability of Large Vision-Language Models (LVLMs). However, the sparse nature of verifiable rewards provides little token-level supervision for failed rollouts, often leading to inefficient exploration in complex multimodal reasoning tasks. Although policy distillation can offer dense guidance, external teacher based methods introduce substantial computational overhead, while answer conditioned tuning methods may expose answer-level information and induce shortcut-like generation behavior. To address these limitations, we propose PTD-PO, a Privileged Tutoring Distillation Policy Optimization framework for RLVR that provides dense guidance without exposing the answer to the student policy. Specifically, PTD-PO constructs structured privileged hints from spatial attention guidance and intermediate textual reasoning steps, and uses them through in-context learning to produce step-wise token-distribution supervision. The student is still optimized under the original answer-free context, and its failed rollouts are aligned with the hint-augmented reference model at the token-distribution level. To further stabilize distillation under the distribution shift between guided and unguided contexts, we introduce a Top-K Jensen-Shannon divergence objective that focuses alignment on informative token probabilities while reducing memory overhead. Experiments on LVLMs ranging from 2B to 8B parameters show that PTD-PO consistently outperforms RLVR and distillation baselines, mitigates entropy collapse, and improves complex multimodal reasoning performance.

arXiv:2606.01202v2 Announce Type: replace Abstract: When a language model answers a factual question, where does its internal state go? We asked four open models (Llama-3.1-8B, Qwen2.5-7B, Gemma-2-9B and Llama-3.2-1B) 2,256 factual questions whose answers recur, such as the country a person is a citizen of. At every layer, we recorded the state from which the answer is produced. Questions that receive the same answer converge on a shared region of state space, which we summarise by its centre, the answer centroid. The answer becomes readable within a single layer. In every model, the share of questions whose state lies nearest its own answer's centroid jumps at one layer, between 47% and 71% of the way through the network. The centroids follow the answer the model gives rather than the question: the centroid of a country learned from questions about people's citizenship also identifies that country in questions about towns and films. The model also uses the centroids. After the jump, moving the state of a wrongly answered question to the right answer's centroid makes the model give the right answer in 83%-97% of cases late in the network, against 2%-4% for a random move of the same size. Before the jump, moves of the same size almost never work. Near misses, in which the model answers wrongly but picks the right option in multiple choice, can be seen from the onset layer, where the right answer's centroid is already among the nearest to the state. This signal is about as informative as the right answer's rank in the model's final output, and more informative than the model's confidence in its reply.

arXiv:2606.00729v4 Announce Type: replace Abstract: AI sovereignty is commonly associated with control over data, infrastructure and technological dependencies. This integrative review examines how institutions convert those resources into capabilities that remain accountable and open to revision. It brings together 72 references on digital sovereignty, absorptive capacity, innovation systems and sociotechnical governance, using six reading dimensions: subject, object, scale, mechanism, legitimacy and evidence. The synthesis proposes five conditions of collective learning capacity: recognizing and retaining knowledge, transmitting it across institutions, protecting consequential challenge, authorizing correction and preserving feasible alternatives. Authoritative corrigibility connects corrective evidence to decision authority; retained option value concerns the practical alternatives that survive a technological commitment. Comparisons of the United States, China and France within Europe distinguish resources and formal arrangements from documented operation and corrective outcomes. Building on this review, Human-Centered Learning Mechanics (HCLM) organizes a descriptive analysis of OECD, IMF and World Bank indicators around information resources, absorption and institutional coordination. The indicators show different resource profiles and trajectories, while leaving the operation of correction and feasible exit largely unmeasured. HCLM connects these observations to questions for institutional investigation; its national-level parameters are not estimated. Purposeful source selection, uneven case coverage and aggregate measurement constrain the conclusions. The resulting research agenda combines reproducible indicator analysis with decision-level evidence of learning, correction and retained choice.

arXiv:2605.30803v2 Announce Type: replace Abstract: LLM judges are increasingly used to evaluate open-ended responses, but their scores depend strongly on the rubrics that condition them. A vague rubric asking for a response to be ``helpful and factual'' can reward polished answers that invent facts or violate user intent. We treat reusable rubrics as measurement specifications: changing the rubric changes the response quality measurement induced by a fixed judge. We introduce PReMISE, a framework that, given pairwise human-preference data, (i) discovers a policy-level rubric set, and (ii) audits any rubric set under LLM-judge use along four axes: structural adequacy, reliability, preference fit, and adversarial robustness. Across rubric sources no raw source is simultaneously reliable, preference-predictive, and adversarially robust; and high inter-rater agreement does not imply low exploitability. PReMISE is the only rubric source to score non-trivially on applicability, specificity, and effective dimensionality simultaneously. We contribute two audit-targeted repair operations: preference-rank selection raises judge accuracy on paired responses from $65.0\%$ to $68.6\%$, competitive with the strongest rubric-discovery baselines and leading on two of three judges in our cross-judge sweep; reliability-constrained refinement reduces the rate at which exploit responses receive high scores from $46.4\%$ to $36.0\%$ with little change in inter-judge agreement ($\alpha{=}.531\to.519$).

arXiv:2605.26772v2 Announce Type: replace Abstract: Large reasoning models (LRMs) generate chain-of-thought (CoT) traces before producing final outputs, introducing a dynamic internal state that may complicate control mechanisms such as refusal. Unlike instruction-tuned LLMs, where refusal is mediated by a single directional subspace, refusal in large reasoning models (LRMs) additionally depends on the CoT. In DeepSeek-R1-Distill-LLaMA-8B, activation steering reverses refusal in only 39% of cases when the CoT is kept fixed, but removing the CoT entirely increases this to 70%, indicating that the CoT actively reinforces refusal. In a two-stage intervention where the model regenerates its CoT under activation steering, refusal is reversed in 94% of cases, while the resulting CoT alone retains 48% of this effect even after steering is removed. This suggests that the CoT can carry and reconstruct the compliance signal independently. These findings indicate that refusal in LRMs is jointly encoded in residual stream activations and CoT. This joint activation makes LRM more robust against activation-level interventions alone, but exposes CoT to a possible alternative surface attack.

arXiv:2605.12682v2 Announce Type: replace Abstract: Large language models (LLMs) are increasingly used as reasoning modules in many applications, yet they often struggle to produce human-aligned decisions. Human-aligned decision-making requires accounting for both explicitly stated goals and latent user preferences that shape how ambiguous situations should be resolved. Existing approaches either learn these preferences through continual per-interaction feedback, which is costly for both the user and the agent, or keep them implicit in dialogue transcripts or example sets tied to a single model. We present CLIPR (Conversational Learning for Inferring Preferences and Reasoning), a framework that represents user's preferences as an explicit, compact set of natural-language rules used at inference time for human-aligned decision-making. CLIPR elicits these rules from conversational input, monitors decision accuracy, and consults the user when performance degrades unexpectedly. CLIPR revises the learned rules, based on the conversation, only if it does not reduce accuracy on scenarios already processed. Evaluations on three preference-ambiguous datasets show that CLIPR matches or outperforms baselines while using up to 96\% fewer LLM calls than continual-feedback preference learning. Our results also show that rules learned by one LLM transfer unchanged to others, and that our approach can recover from corrupted rules. In a user study with 30 participants, CLIPR outperformed baselines and was rated as requiring less effort than continual-feedback learning.

arXiv:2605.08827v3 Announce Type: replace Abstract: The safety of mental health AI is often judged at the wrong temporal scale. Current evaluations typically score isolated responses, endpoint outcomes, or aggregate dialogue quality, while clinically consequential failures may arise from the order and accumulation of interactions themselves, including delayed escalation, repeated reinforcement, dependency formation, failed repair, and gradual deterioration across turns. This paper argues that this mismatch is not merely a limitation of evaluation coverage but a source of invalid safety conclusions. We introduce Temporal Safety Non-Identifiability, a formal account of why safety properties that depend on sequence, timing, accumulation, or recovery cannot be certified by protocols that discard those features. From this formalization, we develop SCOPE (Safety Claims Over Preserved Evidence) as a general principle for aligning safety claims with the evidence an evaluation actually retains, and instantiate it as SCOPE-MH, a mental-health instantiation of this reporting standard. We operationalize SCOPE-MH through a proof-of-concept on the AnnoMI dataset of expert-annotated motivational interviewing conversations, which reveals mechanisms of failure that per-turn behavior scoring does not represent. We propose SCOPE-MH as a diagnostic complement to existing evaluation infrastructure and argue that evaluation preserving temporal evidence is necessary, not optional, for safety-critical mental health AI deployment.

arXiv:2604.23716v5 Announce Type: replace Abstract: Reporting guidance for information-theoretic measures is rarely tested against ground truth. We test one guardrail in two multi-agent reinforcement learning games, a social dilemma and a coordination race, where directed influence between selected agent pairs is zero by construction, over 100 seeds. Omitting one precondition, exclusion of the non-stationary training transient, gives false-positive rates of 100.00% and 99.95%: agents annealing exploration independently, in runs that never met, are flagged as influencing one another. Excluding the transient reaches 3.05% in the social dilemma but 11.8% in the coordination game, so the non-stationarity must be treated and exclusion is neither the only way nor sufficient. What we recommend instead changes the null model. Series, statistic and estimand are left untouched: permuting the source within blocks of training time reaches 5.25% and 5.50%, the only one of the three at or below the size of the test in both games. The construction is not new, being a restricted permutation local to blocks of training time; the measurement is. A third null control, four learners in the social dilemma, defeats all three, and there conditioning costs three times what ours does, 15.20 points against 4.55. Titrating injected links separates it from conditioning on identical pairs, 82.5% against 58.0%, while the ablated test reports a positive with or without a link, so its apparent sensitivity is uninformative. Five block counts spanning 16 to 256 land in the nominal region. Code and data are released.

arXiv:2604.07667v2 Announce Type: replace Abstract: Multi-agent debate improves LLM reasoning, yet agreement is not evidence of correctness. When agents converge on a wrong answer through social reinforcement, consensus stopping commits that error to automated action with no recourse. We introduce Conformal Social Choice, a post-hoc layer that converts debate outputs into calibrated act-versus-escalate decisions. Verbalized probability distributions from heterogeneous agents are aggregated via a linear opinion pool and calibrated with split conformal prediction, yielding fixed-round marginal coverage: at any pre-specified round, the correct answer is included with probability ${\geq}\,1{-}\alpha$, without assumptions on individual model calibration. A hierarchical action policy maps singletons to autonomous action and larger sets to human escalation. Across eight MMLU-Pro domains and two additional benchmarks, fixed-round coverage stays close to target. The key finding is not that debate becomes more accurate, but that the conformal layer makes its failures actionable: at the final round, 81.9\% of wrong-consensus cases are flagged by non-singleton sets at $\alpha{=}0.05$. Separately, among cases that reach a singleton before the final round, accuracy reaches 90.0--96.8\% on MMLU-Pro (up to 22.1pp above consensus stopping): a selection effect, not a reasoning improvement. Simple probability- and entropy-threshold rules attain comparable interception; conformal's distinct value is a finite-sample fixed-round coverage guarantee together with explicit candidate sets for escalation. Adaptive first-singleton stopping is evaluated empirically and not covered by this guarantee.

arXiv:2602.14622v3 Announce Type: replace Abstract: Association Rule Mining (ARM) is a fundamental task for knowledge discovery in tabular data and is widely used in high-stakes decision-making. Classical ARM methods rely on frequent itemset mining, leading to rule explosion and poor scalability, while recent neural approaches mitigate these issues but suffer from degraded performance in low-data regimes. Tabular foundation models (TFMs), pretrained on diverse tabular data with strong in-context generalization, provide a basis for addressing these limitations. We introduce a model-agnostic association rule learning framework that extracts association rules from any conditional probabilistic model over tabular data, enabling us to leverage TFMs. We then introduce TabProbe, an instantiation of our framework that utilizes TFMs as conditional probability estimators to learn association rules out-of-the-box without frequent itemset mining. We evaluate our approach on tabular datasets of varying sizes based on standard ARM rule quality metrics and downstream classification performance. The results show that TFMs consistently produce concise, high-quality association rules with strong predictive performance and remain robust in low-data settings without task-specific training. Source code is available at https://github.com/DiTEC-project/tabprobe.

arXiv:2602.02589v2 Announce Type: replace Abstract: Large language model (LLM) evaluation commonly relies on fixed human-authored benchmarks or centralized human/model judges. Fixed benchmarks can age and enter model training corpora, while centralized judges introduce their own scoring preferences. We cast LLM evaluation as a multi-agent peer-assessment problem and introduce Agents-as-Peers, a distributed evaluation protocol in which every agent generates tasks, answers the shared task set, and judges the answers of its peers, with no privileged evaluator. Self-ratings are excluded from the ranking, and judging is repeated under controlled identity and presentation regimes so that self bias, identity effects, presentation-position sensitivity, and judge severity are measured rather than left implicit. In an instantiation with 12 LLMs and 420 agent-generated questions, Agents-as-Peers aggregates 55,440 non-self peer judgments per regime. Most agents assign their own answers higher scores than those answers receive from peer judges; visible identity changes scores for some agents; the first fixed answer slot is associated with scores 0.39 points higher than the shuffled baseline; and judges differ markedly in severity. On labeled benchmarks, peer scores align strongly with accuracy on TruthfulQA (Pearson r=0.904) and GSM8K (r=0.873). On TruthfulQA, the peer-score correlation is substantially higher than the self-score correlation (r=0.538). A benchmark-overlap screen and item-level human audit further characterize the endogenous task set. These results support Agents-as-Peers as a cohort-relative measurement protocol whose ranking is reported together with the biases and rater effects that shape it. Code and dataset: https://github.com/caura-ai/caura-PeerRank

arXiv:2512.23067v3 Announce Type: replace Abstract: Personalized alignment from preference data has focused primarily on improving personal reward model (RM) accuracy, with the implicit assumption that better preference ranking translates to better personalized behavior. However, in deployment, computational constraints necessitate inference-time adaptation such as reward-guided decoding (RGD) rather than per-user policy fine-tuning. This creates a critical but overlooked requirement: reward models must not only rank preferences accurately but also effectively guide generation. We demonstrate that standard RM accuracy fails catastrophically as a selection criterion for deployment-ready personalized rewards. We introduce policy accuracy; a metric quantifying whether RGD-adapted LLMs correctly discriminate between preferred and dispreferred responses and show that upstream RM accuracy correlates only weakly with downstream policy accuracy (Kendall's tau = 0.08--0.31). More critically, we introduce Pref-LaMP the first personalized alignment benchmark with ground-truth user completions, enabling direct behavioural evaluation. On Pref-LaMP, we expose a complete decoupling between discriminative ranking and generation metrics: methods with 20-point RM accuracy differences produce almost identical output quality, and methods with high ranking accuracy can fail to generate behaviorally aligned responses. These findings reveal that the field has been optimizing for proxy metrics that do not predict deployment performance, and that current personalized alignment methods fail to operationalize preferences into behavioral adaptation under realistic deployment constraints. In contrast, we find simple in-context learning (ICL) to be highly effective - dominating all reward-guided methods for models $\geq$3B parameters, achieving $\sim$3 point ROUGE-1 gains over the best reward method at 7B scale.

arXiv:2510.14697v2 Announce Type: replace Abstract: Model merging aims to integrate task-specific abilities from individually fine-tuned models into a single model without extra training. In recent model merging methods, task vector has become a fundamental building block, as it can encapsulate the residual information from finetuning. However, the merged model often suffers from notable performance degradation due to the conflicts caused by task-irrelevant redundancy in task vectors. Existing efforts in overcoming redundancy by randomly dropping elements in the parameter space involves randomness and lacks knowledge awareness. To address these challenges, in this study, we propose Purifying TAsk Vectors (PAVE) in knowledge-aware subspace. Concretely, we sample some training examples from each task, and feed them into their corresponding fine-tuned models to acquire the covariance matrices before linear layers. We then perform a context-oriented singular value decomposition, which accentuates the weight components most relevant to the target knowledge. As a result, we can split fine-tuned model weights into task-relevant and redundant components in the knowledge-aware subspace, and purify the task vector by pruning the redundant components. To induce fair pruning efforts across models, we further introduce a spectral rank allocation strategy by optimizing a normalized activated pruning error. The task vector purification by our method as a plug-and-play scheme is applicable across various task vector-based merging methods to improve their performance. In experiments, we demonstrate the effectiveness of PAVE across a diverse set of merging methods, tasks, and model architectures.

arXiv:2510.07073v2 Announce Type: replace Abstract: Designing high-performing heuristics for vehicle routing problems (VRPs) is a complex task that requires both intuition and deep domain knowledge. Large language model (LLM)-based code generation has recently shown promise across many domains, but it still falls short of producing heuristics that rival those crafted by human experts. In this paper, we propose VRPAgent, a framework for VRPs that integrates LLM-generated components into a metaheuristic and refines them through a novel genetic search. By using the LLM to generate problem-specific operators, embedded within a shared metaheuristic framework for VRPs, VRPAgent keeps tasks manageable, guarantees correctness, and still enables the discovery of novel and powerful strategies. Across multiple VRPs, including the capacitated VRP, the VRP with time windows, and the prize-collecting VRP, our method discovers heuristic operators that outperform handcrafted methods and recent learning-based approaches while requiring only a single CPU core. To our knowledge, VRPAgent is one of the first LLM-based methods to advance the state-of-the-art in VRPs, highlighting a promising future for automated heuristics discovery.

arXiv:2509.22407v3 Announce Type: replace Abstract: The generalization of vision-language-action (VLA) models heavily relies on diverse training data. However, acquiring large-scale robot manipulation data across varied object appearances is costly and labor-intensive. To address this, we introduce Embodied Manipulation Media Adaptation (EMMA), a framework for augmenting VLA policies that combines a generative data engine with an effective training pipeline. We introduce DreamTransfer, a diffusion Transformer-based architecture for generating multi-view consistent and geometrically plausible embodied manipulation videos. It supports text-driven video editing on foreground, background and lighting while preserving 3D structure and geometric fidelity. We also utilize a hybrid training set of real and generated data and propose AdaMix to enhance the training process. AdaMix is a training strategy that adaptively weights samples according to policy performance to emphasize challenging samples. Evaluations show DreamTransfer outperforms previous video generation models in multi-view consistency, geometric fidelity, and text-conditioning precision. In real-world robotic tasks with zero-shot visual settings, our framework achieves a relative performance increase of over 92% compared to training with real data alone, and improves by an additional 17% with AdaMix.

arXiv:2508.20784v3 Announce Type: replace Abstract: Bus bunching remains a challenge for urban transit due to stochastic traffic and passenger demand. Traditional solutions rely on multi-agent reinforcement learning (MARL) in loop-line settings, which overlook realistic operations characterized by heterogeneous routes, timetables, fluctuating demand, and varying fleet sizes. We propose a novel single-agent reinforcement learning (RL) framework for bus holding control that avoids the data imbalance and convergence issues of MARL under near-realistic simulation. A bidirectional timetabled network with dynamic passenger demand is constructed. The key innovation is reformulating the multi-agent problem into a single-agent one by augmenting the state space with categorical identifiers (vehicle ID, station ID, time period) in addition to numerical features (headway, occupancy, velocity). This high-dimensional encoding enables single-agent policies to capture inter-agent dependencies, analogous to projecting non-separable inputs into a higher-dimensional space. We further design a structured reward function aligned with operational goals: instead of exponential penalties on headway deviations, a ridge-shaped reward balances uniform headways and schedule adherence. Experiments show that our modified soft actor-critic (SAC) achieves more stable and superior performance than benchmarks, including MADDPG (e.g., -430k vs. -530k under stochastic conditions). These results demonstrate that single-agent deep RL, when enhanced with categorical structuring and schedule-aware rewards, can effectively manage bus holding in non-loop, real-world contexts. This paradigm offers a robust, scalable alternative to MARL frameworks, particularly where agent-specific experiences are imbalanced.

arXiv:2508.11850v3 Announce Type: replace Abstract: Integer programming (IP) is central to many combinatorial optimization tasks but remains challenging due to its NP-hard nature. A practical way to improve IP solvers is to manually design inequality families that exploit problem structure. However, this creative process requires deep expertise and has therefore been difficult to automate. Our proposed framework, Symbolic Acceleration via Generative Evolution (SAGE), automates symbolic discovery of acceleration inequality families at the modeling level. It reasons over a symbolic IP formulation and a natural language description of the problem to discover a reusable set of acceleration inequality families that can be instantiated for each problem instance. These acceleration inequality families are selected through empirical admissibility and impact screening rather than mathematical proofs. SAGE (i) initializes a population of candidate acceleration inequality families via an initializer agent that uses an LLM; (ii) it screens candidates for admissibility on a small development admissibility set for solution feasibility and the separation of the relaxation solution; and (iii) it iteratively refines the population of inequalities through evolutionary crossover and mutation agents. Across seven MILP benchmarks, SAGE reduces optimality gaps by up to $62.1\%$ relative to baseline MILP formulations solved with a fixed time budget and reaches target gaps up to $5.51\times$ faster according to shifted geometric mean speedup. Comprehensive ablations show its robustness across different LLM backends and across solvers and internal cut settings. Code: https://github.com/milad1378yz/sage

arXiv:2508.01300v2 Announce Type: replace Abstract: AI planning provides agents the ability to act autonomously towards achieving a goal in a given environment. Recent advances in Large Language Models (LLMs) have introduced the prospect of using unstructured input for planning, making LLMs in planning an active area of research, especially for supporting autonomous agents. However, LLMs remain unreliable as long-horizon planners, since some of the plans they produce contain mistaken or hallucinated actions; executing such plans could lead to unexpected and harmful outcomes. Existing benchmarks for LLMs in planning primarily rely on binary evaluation metrics, providing limited insight into the quality of plans that do not achieve the goal, and how far those are from becoming valid. In this paper, we frame the evaluation of LLMs in planning as a natural language processing (NLP) task, as LLMs are themselves NLP models. We propose a pipeline comprising three NLP-based steps that assess plan quality from complementary perspectives, followed by a symbolic planning step to achieve a valid plan, to provide a finer-grained understanding of the quality of invalid plans. In our experiments, we find no clear evidence of logical derivation or state-tracking during LLM plan generation -- on average, for all non-empty plans, only 3.55 actions per plan are executable, despite symbolically-generated optimal plans having an average length of 9.47 actions, which suggests that LLMs are still far from the reliability of symbolic planners. Our suite of metrics is further used to investigate patterns in LLM plan generation over time, and can provide a useful metric for assessing the quality of individual invalid plans.