Transformer Field Theory: A Response-Theoretic Approach to Mechanistic Interpretability — Reproducibility Data and Code

Overview This archive provides the experimental data, analysis code, computational notebooks, metadata, and reference figures associated with the research paper "Transformer Field Theory: A Response-Theoretic Approach to Mechanistic Interpretability." The research investigates a response-theoretic framework for studying transformer representations and their behavior under perturbations, with the aim of developing quantitative tools for mechanistic interpretability. Contents The reproducibility archive is organized as follows: data/ — Experimental data supporting the analyses. metadata/ — Metadata documenting experimental configurations and associated information. notebooks/ — Jupyter notebooks used for computational analysis and visualization. scripts/ — Supporting scripts for the experimental workflow. reference_figures/ — Reference figures associated with the manuscript. README.md — Documentation describing the archive and its use. requirements.txt and requirements-experiments.txt — Python dependency specifications. SHA256SUMS — Checksums for verifying file integrity. LICENSE — Licensing information. Reproducibility The archive is intended to support the verification, inspection, and reproduction of the computational analyses presented in the accompanying manuscript. Users should consult the README documentation for instructions on the computational environment, dependencies, data organization, and available reproduction procedures. Associated Publication Transformer Field Theory: A Response-Theoretic Approach to Mechanistic Interpretability. Version Version 1.0.0 — Initial reproducibility archive. Citation Researchers using these materials should cite both the associated research paper and this Zenodo dataset: https://doi.org/10.5281/zenodo.23734366.