A fixed-cohort study testing whether general-purpose compression detects model-associated signatures across held-out prompt sets in short continuations from 127 model endpoints and configurations. The results support candidate-ranking structure beyond exact response length, while character n-gram TF-IDF is stronger and the evidence does not establish open-set or forensic attribution.
An independent evaluation of an EvoForest architecture reimplementation on PhysioNet/CinC 2017 atrial-fibrillation detection. The study finds transferable signal from graph evolution, while a fixed random-convolution representation performs better and exposes redundancy, proposal failures, and computational costs.
A preprint introducing a tiered audit benchmark for quantum artificial-life claims, separating artificial-life evidence, quantum-resource relevance, and computational nonclassicality.
A preprint and research artifact studying whether LLM-driven adversarial program evolution changes when the evolving object is a population rather than a single champion. The repository contains the manuscript, Core War pilot and benchmark harnesses, generated warrior artifacts, figures, tables, and reproducibility scripts.