ClarusC64/clinical-compensated-instability-collapse-v0.1

该数据集用于评估模型是否能在明显崩溃前检测到补偿性临床不稳定性。每行数据代表一个短临床轨迹，其中表面测量可能看起来稳定，而潜在的补偿机制正在减弱。任务是根据轨迹判断是保持补偿状态还是接近补偿失败。数据集包含多个轨迹变量（如血压轨迹、心率轨迹、休克指数轨迹等）和干扰变量（如测量噪声、图表噪声）。预测目标为标签1（补偿失败接近）或0（补偿稳定轨迹）。数据集是Clarus稳定性推理基准的一部分，支持补偿性不稳定性、延迟崩溃检测、基于轨迹的推理等研究领域。

This dataset evaluates whether models can detect compensated clinical instability before overt collapse. Each row represents a short clinical trajectory where surface measurements may appear stable while underlying compensation is weakening. The task is to determine whether the scenario remains compensated or is approaching compensation failure. The dataset includes multiple trajectory variables (e.g., MAP trajectory, heart-rate trajectory, shock-index trajectory) and decoy variables (e.g., measurement_noise, chart_noise). The prediction target is label = 1 (compensation failure approaching) or label = 0 (compensated and stable trajectory). This dataset is part of the Clarus Stability Reasoning Benchmark and supports research into compensated instability, delayed collapse detection, trajectory-based reasoning, and other areas.