Reproducibility package for "Algorithmic Correctability and Structural Failure Modes in LLM-Assisted AHP-DEMATEL: A Large-Scale Multi-Model Empirical Study"

This archive contains the filtered experiment dataset, final processed analysis outputs, manuscript figures, supplementary materials, and code required to reproduce the reported analyses from local raw results. The packaged article dataset covers four models from three architectural families across three decision domains and 3,120 recorded API-call entries: 360 criteria-generation runs, 360 DEMATEL runs, 360 AHP Phase A runs, 360 AHP Phase C runs, and 1,680 alternative-rating entries. The final analysis uses an AHP correction budget of 4 and excludes kimi-k2 outputs from the broader production run. Contents: raw experiment outputs used for the article; final processed analysis outputs; manuscript figures; supplementary materials PDF; reproducibility code, prompts, configurations, and reference chains. This Zenodo archive contains mixed-rights materials. The code in the code/ directory is released under the MIT License. Non-code materials are provided for scholarly inspection, verification, and reproducibility purposes as described in the included top-level LICENSE file. Source production run identifier: real-experiment-20260308_183614-resumedVersion: 1.0.1