Title Truth-of-the-Future: Weaponised Generative AI and the Optimisation of Fake News Amplification (Code and Reproducibility Artefacts)

This repository contains the complete code and reproducibility artefacts supporting the study "Truth-of-the-Future: Weaponised Generative AI and the Optimisation of Fake News Amplification." It reproduces every reported result in a deterministic environment with a single fixed random seed across Python, NumPy and PyTorch. The pipeline comprises: text normalisation and exact (SHA-256) plus near-duplicate (MinHash LSH) deduplication; leakage-safe, group-stratified train/validation/test split construction (by speaker, event, publisher or item); a text-only DistilRoBERTa detection baseline with calibrated inference, threshold tuning, and post-hoc calibration fitted only on validation data; a full predictive-validity battery (accuracy, balanced accuracy, macro and minority F1, MCC, ROC-AUC, PR-AUC, Brier, Brier skill score, ECE and MCE) with 95% stratified bootstrap confidence intervals (R = 500); the Weaponisation Index Pipeline (Linguistic Manipulation Index, Multimodal Incongruence Score, Amplification Efficiency Ratio, Early Capture Advantage) and the Unified Weaponisation Risk Profile; a robustness-degradation evaluation under paraphrase and cue-injection perturbations; and all figure-generation scripts.