Code and derived results for procedure-level conformal calibration of step-recognition outputs in cataract surgery videos

This record accompanies the manuscript “Procedure-Level Reliability Assessment for Cataract Surgery Video Step Recognition Using Conformal Prediction” It contains the analysis scripts, deterministic procedure-level split definitions, machine-readable results underlying the main tables and figures, and metadata for the representative video frames shown in Fig. 1. The record does not redistribute source surgical videos, manual annotations, or released model-output files. The annotations and model outputs used in the analysis are available from version 2 of the Johns Hopkins Research Data Repository record at https://doi.org/10.7281/T1ZZX1NJ. The original Cataract-1K videos are available at https://doi.org/10.7303/SYN52540135. The deposited code covers fixed soft ensembling, probability calibration, procedure-level conformal prediction, sensitivity analyses, cluster-dependence simulation, and generation of the three manuscript figures.