The Automated Consensus Pipeline Engine: Replications and Flat Dataset Metrics

This repository archives the official replication dataset, processing source code, and empirical assets for the paper titled "The Automated Consensus: A Hierarchical Evaluation of Task-Sensitive Semantic Relatedness in Aligned Large Language Models" (Mehta, 2026). The payload contains the complete 240-observation empirical text matrix aggregated across twenty unique operational crisis scenarios (N=20), evaluating OpenAI's GPT-5.5 engine and Google's Gemini 3.5 Flash (Fast) platform under adversarial system constraints. Included Archive Files: dataset.csv: The unified, flat database tracking text entries against strict structural metadata variables [model, scenario, task, run, text]. automated_consensus_engine.py: The production pipeline script that executes 768-dimensional sentence-vector coordinate tracking (all-mpnet-base-v2), moving-average token diversity mapping (MATTR-50), zero-shot heuristic classification (BART-Large-MNLI), and the 10,000-iteration paired hierarchical block bootstrap difference test. README.md: Comprehensive environmental dependency manual and execution instructions. Open-Science Compliance: This repository is maintained for open-access scientific verification, computational humanities replication, and algorithmic audit tracking under the Creative Commons Attribution 4.0 International (CC-BY-4.0) license.本仓库归档保存了论文《自动化共识：对齐大语言模型（Large Language Model, LLM）中任务敏感语义相关性的层级评估》（Mehta，2026）的官方复现数据集、处理源代码与实证素材。 本数据集包含横跨20种独特运营危机场景（N=20）的完整240条观测实证文本矩阵，在对抗性系统约束条件下，对OpenAI的GPT-5.5引擎与Google的Gemini 3.5 Flash（极速版）平台开展了评估。 归档文件包含： dataset.csv：统一扁平化数据库，用于记录文本条目与严格的结构化元数据变量[模型(model)、场景(scenario)、任务(task)、运行次数(run)、文本(text)]。 automated_consensus_engine.py：生产级流水线执行脚本，可实现768维句向量坐标追踪（基于all-mpnet-base-v2模型）、移动平均Token多样性映射（MATTR-50）、零样本（Zero-shot）启发式分类（BART-Large-MNLI），以及10000次迭代的配对层级块Bootstrap差异检验。 README.md：涵盖完整的环境依赖手册与运行执行指南。 开放科学合规说明： 本仓库遵循知识共享署名4.0国际许可协议（Creative Commons Attribution 4.0 International, CC-BY-4.0），用于支持开放获取的科学验证、计算人文领域的复现研究，以及算法审计追踪。