Unpredictability dissociates from structured control in language agents

Unpredictable behavior is often taken as evidence of control, yet stochastic dispersion and structured action control need not coincide. This paper tests whether stochastic sampling can substitute for structured mechanisms that couple reasons, memory, self-state and inhibition to action selection in a language-agent implementation whose control components can be selectively disabled. In a seven-dataset baseline lesion matrix comprising 74,352 calls, the high-stochasticity comparator was more unpredictable than the structured-control variant in 7/7 datasets, whereas targeted reason and veto lesions reduced the expected structured-control profiles in 7/7 datasets each. In a matched-interface control spanning 26,946 generations, the structured agent maintained stronger action-field coupling than all stochastic, post-hoc, scrambled and verbosity controls across every dataset. The primary behavioral test removed free-form trace wording from the evaluation: 57,816 scored records showed the structured-control variant exceeding the high-stochasticity comparator or the reason/veto lesions in 7/7 datasets for all predefined behavioral components. Later open-weight runs extended the no-context controls to Qwen2.5 7B, 14B and 32B and to an independent Mistral-7B family across 20 task families and three agent scaffolds; no-fields, scrambled-context and distribution-matched controls failed to recover structured action control. A three-annotator blinded audit over 1,200 overlap items preserved high agreement. Strict entropy matching, strict token/compute matching and a formal counterfactual-flip stress test did not meet their gates and are treated as limitations. Stochastic unpredictability did not reproduce structured, action-coupled control in this implemented agent family.

NeuroState-Bench: A Human-Calibrated Benchmark for Commitment Integrity in LLM Agent Profiles

Outcome-only evaluation under-specifies whether an evaluated agent profile preserves the commitments required to solve a multi-turn task coherently. NeuroState-Bench is a human-calibrated benchmark that operationalizes commitment integrity through benchmark-defined side-query probes rather than inferred hidden activations. The released inventory contains 144 deterministic tasks and 306 benchmark-defined side-query probes spanning eight cognitively motivated failure families, paired clean and distractor variants, and three difficulty bands. The main 32-profile evaluation contains a fixed 16-profile local subset and a matched 16-profile hosted large-model subset evaluated through the same benchmark pipeline. Human calibration uses the final merged reporting scope: 104 sampled task units, 216 raw annotations, and 108 adjudicated task rows, with weighted kappa = 0.977 and ICC(2,1) = 0.977. Empirically, task success and commitment integrity diverge across this expanded grid: the success leader is not the integrity leader, 31 of 32 profiles change rank when integrity replaces task success, and integrity rankings are more stable under distractor perturbation. The primary confidence-free score HCCIS-CORE reaches 0.8469 AUC and 0.6992 PR-AUC for post-probe diagnostic discrimination of terminal task failure; the legacy full heuristic variant HCCIS-FULL reaches 0.7997 AUC and 0.6410 PR-AUC. Probe accuracy and state drift achieve slightly higher ROC-AUC, 0.8587, and better Brier/ECE, while HCCIS-CORE has substantially higher point-estimate PR-AUC and remains more closely tied to the benchmark's intended construct. The exploratory neural-augmented variant HCCIS+N is weaker overall, and a randomized subspace control approaches chance. NeuroState-Bench therefore contributes a calibrated evaluation axis for exposing commitment failures over a broader model grid than the original local-only subset.

Weakly Supervised Online Action Detection for Infant General Movements

To make the earlier medical intervention of infants' cerebral palsy (CP), early diagnosis of brain damage is critical. Although general movements assessment(GMA) has shown promising results in early CP detection, it is laborious. Most existing works take videos as input to make fidgety movements(FMs) classification for the GMA automation. Those methods require a complete observation of videos and can not localize video frames containing normal FMs. Therefore we propose a novel approach named WO-GMA to perform FMs localization in the weakly supervised online setting. Infant body keypoints are first extracted as the inputs to WO-GMA. Then WO-GMA performs local spatio-temporal extraction followed by two network branches to generate pseudo clip labels and model online actions. With the clip-level pseudo labels, the action modeling branch learns to detect FMs in an online fashion. Experimental results on a dataset with 757 videos of different infants show that WO-GMA can get state-of-the-art video-level classification and cliplevel detection results. Moreover, only the first 20% duration of the video is needed to get classification results as good as fully observed, implying a significantly shortened FMs diagnosis time. Code is available at: https://github.com/scofiedluo/WO-GMA.