AAB Dataset — The Collapse Curve: Quantifying the Dose-Dependent Decay of Affective Reasoning in Large Language Models (v1.0)

This dataset accompanies the article “The Collapse Curve: Quantifying the Dose-Dependent Decay of Affective Reasoning in Large Language Models” (DOI: 10.5281/zenodo.17425689). It provides the full reproducibility package for the study of Algorithmic Affective Blunting (AAB), a phenomenon describing the degradation of affective interpretation under semantic stress in large language models (LLMs). Contents:- Core data (Phase 1–4 ADI evaluations)- Calibration logs and evaluation schema- Statistical tables (CLMM summaries, ablation splits, simulated base-aligned conditions)- LaTeX exports for tables and figures- Full manifest and README for reproducibility All data were collected empirically or derived through explicit simulation (Phase 4). No synthetic or placeholder values are included. This release supports open verification of the AAB framework introduced in the associated paper.