This repository contains the code and partial results used for the misalignment analysis presented in the main body and appendix of our research paper. The scripts provided here generate key results related to action sequences, hierarchical structures, and state-space graph analysis. Note that user trajectory data is not included due to confidentiality.
action_sequence_parser.py: Parses and processes action sequences for analysis.classify.py: Classifies data based on predefined criteria for misalignment detection.degree_dist_calculator.py: Computes degree distributions of graphs to analyze structural properties.degree_dist.zip: Precomputed degree distribution data used in the analysis.grapher_with_abstract_actions.py: Generates graphs incorporating abstract actions.hierarchical_analysis.ipynb: Jupyter Notebook containing hierarchical analysis of the state-space.state_space_graph_maker_with_abstract_actions.py: Creates state-space graphs using abstract actions.state_space_graphs_with_abstract_actions.zip: Precomputed state-space graphs for further analysis.trajectory_visualization.py: Visualizes action trajectories to examine misalignment patterns.