ClarusC64/clinical-monitoring-failure-instability-v0.1

该数据集用于评估模型是否能检测由监测失败而非直接生理崩溃引起的不稳定性。每个数据行代表一个简化的临床监测场景，包含三个时间点。任务目标是判断护理系统是否仍能检测到病情恶化或正在走向检测失败。数据集测试了传感器丢失轨迹、监测延迟轨迹、警报抑制、患者变异性、工作人员响应延迟、敏锐度水平和升级路径负载等多个方面的交互推理。预测目标为label=1表示监测失败不稳定，label=0表示监测和响应状态稳定。数据集结构包括多个变量，其中有些是干扰变量。评估方法包括准确率、精确率、召回率、F1分数等。数据集是Clarus稳定性推理基准的一部分。

This dataset evaluates whether models can detect instability caused by monitoring failure rather than direct physiological collapse. Each row represents a simplified clinical monitoring scenario across three time points. The task is to determine whether the care system remains able to detect deterioration or is moving toward detection failure. The dataset tests interaction reasoning across: sensor dropout trajectory, monitoring latency trajectory, alarm suppression, patient variability, staff response delay, acuity level, and escalation pathway load. The prediction target is label = 1 → monitoring failure instability, label = 0 → stable monitoring and response state. The dataset includes multiple variables, some of which are decoy variables. Evaluation metrics include accuracy, precision, recall, f1, confusion matrix, and dataset integrity diagnostics. This dataset is part of the Clarus Stability Reasoning Benchmark.