We replicate and extend Volkhonskiy et al. (2017) on Inductive Conformal Martingales (ICM) for change-point detection in time series. The original study uses k-nearest neighbors (KNN) as the nonconformity measure; here we propose median absolute deviation (MAD) and interquartile range (IQR), which are more robust to outliers, and compare all three. We implement the full pipeline in R using dplyr, tidyr, purrr and ggplot2. We also extend ICM to multiple detections by restarting or updating the martingale after each alarm. Our experiments assess KNN, MAD, and IQR in terms of average detection delay, false-alarm probability, and stability under various settings, with reproducible visualization and validation routines. In addition, we contrast multiple-detection strategies, highlighting strengths and limitations across simulated scenarios and discussing their practical applicability.