RogueGPT: A controlled stimulus generator for AI news authenticity research. (arXiv:2601.21963 and arXiv:2601.22871)

This is your the only entry point to deeply understand controlled experiments and begin to use them in practice.

A Claude Code skill that forces agents to quote real code before writing new code. Backed by 930 controlled runs: 0/40 → 40/40 correctness.

GEO Attribution Framework (GAF): causal inference for AI search visibility. A reproducible controlled-experiment template on top of Peec MCP for measuring how LLMs cite brands in any category. MIT.

Replication package, supplementary materials, and analysis pipeline for our ICPC'26 paper.

Replication package of the study titled "The Impact of Knowledge Distillation on the Energy Consumption and Runtime Efficiency of NLP Models" presented at CAIN 2024

Prompt-TDD · Prompt 对照实验方法论案例手册 — SOP、分析工具与两个真实对照实验（含阴性结果）。A casebook for controlled prompt-engineering experiments. CC BY 4.0

Replication package for controlled experiment on investigating effectiveness, efficiency and preference of a controlled natural-language notation versus a key-value notation for the Storyboard Language DSL

AI协作项目全生命周期框架 · A methodology framework for full-lifecycle human-AI collaboration — reviewed through multi-model independent review, 68K Chinese characters · CC BY 4.0

对 Karpathy「不要问 LLM 你怎么想」建议的实证批评。2天单人实验证明：Karpathy 说对了机制原文，但说错了使用建议。附量化测量框架和可复现方法。

A field-tested multi-model academic pipeline: 8-stage workflow from research design through blind peer review, defense simulation & controlled open/closed-book experiment. | 多模型协同学术流水线——论文是测试用例，流水线是核心交付物。

Protótipo experimental para comparar o desempenho front-end de duas versões de uma aplicação web, com coleta de métricas pelo Lighthouse.