Supporting Data and Code for "Anticipating Instability in Learning Systems: A Comparative Study of Functional Coherence Metrics"

This dataset contains the experimental results, analysis code, and figures supporting the manuscript "Anticipating Instability in Learning Systems: A Comparative Study of Functional Coherence Metrics" by Angelina Davini Hintsanen (2026). Contents:- Raw experimental data (CSV files) for all four test scenarios (A1-A4)- 20-seed statistical comparison results- High-resolution figures (300 DPI, publication-ready)- Analysis scripts for reproducing all results- README with methodology and data format descriptions Experiments cover:A1: Fine-tuning drift detection (transformer, conflicting labels)A2: Catastrophic forgetting detection (continual learning)A3: Policy drift detection (reinforcement learning, reward hacking)A4: Semantic drift detection (LLM, adversarial prompts) All experiments validate ZC(t) and H-CAM coherence metrics for early instability detection in machine learning systems. Intended Use This archive is intended for independent verification, methodological review, and non‑commercial research use by qualified researchers. machine learning, instability detection, early warning systems, AI safety, catastrophic forgetting, reward hacking, semantic drift, coherence metrics, continual learning, reinforcement learning https://orcid.org/0009-0009-7709-4336