ClarusC64/clinical-recovery-window-detection-v0.1

该数据集用于评估模型是否能识别临床轨迹是否保持在恢复窗口内或已超出恢复窗口。任务不仅仅是检测严重性，还包括恢复能力。核心稳定性理念涉及轨迹方向、响应能力和时间之间的相互作用。数据集测试了压力恢复、乳酸逆转、肾输出趋势、对升压药的反应、对液体的反应、氧合反应和干预延迟等多个方面的推理能力。预测目标为`label = 1`表示恢复窗口已失败或关闭，`label = 0`表示轨迹仍可恢复或稳定。每行数据包括`scenario_id`、MAP轨迹、乳酸轨迹、尿量轨迹、升压药反应、液体反应、氧合反应、干预延迟、恢复趋势标记、干扰变量和标签。干扰变量包括`observation_noise`和`chart_noise`，这些变量看似有意义但单独不决定标签。

This dataset evaluates whether models can identify when a clinical trajectory remains inside or has crossed a recovery window. The task is not simply to detect severity but also recovery capacity. The core stability idea involves the interaction between trajectory direction, response capacity, and timing. The dataset tests reasoning across pressure recovery, lactate reversal, renal output trend, response to pressors, response to fluids, oxygenation response, and intervention delay. The prediction target is `label = 1` means the recovery window has failed or is effectively closed, `label = 0` means the trajectory remains recoverable or stabilizing. Each row includes `scenario_id`, MAP trajectory, lactate trajectory, urine output trajectory, pressor response, fluid response, oxygenation response, intervention delay, recovery trend marker, decoy variables, and label. Decoy variables include `observation_noise` and `chart_noise`, which appear meaningful but do not determine the label alone.