ClarusC64/clinical-acid-base-instability-v0.1

该数据集用于评估模型是否能够从简短的临床代理轨迹中检测酸碱不稳定性。每个数据行代表三个时间点的简化酸碱场景，任务是根据pH轨迹、碳酸氢盐轨迹、pCO2轨迹、乳酸轨迹、肾缓冲代理、呼吸补偿代理和校正延迟等指标，判断系统是否保持缓冲或趋向代谢不稳定。数据集还包含了一些干扰变量，这些变量看似有意义但单独使用时并不决定标签。数据集的评估方法包括准确率、精确率、召回率、F1分数、混淆矩阵和数据集完整性诊断。该数据集是Clarus稳定性推理基准的一部分，旨在作为酸碱稳定性推理的紧凑基准，而非临床决策工具。

This dataset evaluates whether models can detect acid–base instability from short clinical proxy trajectories. Each row represents a simplified acid–base scenario across three time points. The task is to determine whether the system remains buffered or is moving toward metabolic instability based on pH trajectory, bicarbonate trajectory, pCO2 trajectory, lactate trajectory, renal buffer proxy, respiratory compensation proxy, and correction delay. The dataset also includes decoy variables that appear meaningful but do not determine the label alone. Evaluation methods include accuracy, precision, recall, F1 score, confusion matrix, and dataset integrity diagnostics. This dataset is part of the Clarus Stability Reasoning Benchmark and is intended as a compact benchmark for acid–base stability reasoning, not as a clinical decision tool.