ActionPace

Ensemble Divergence as a Pre-Reproduction Audit

For readers coming from LinkedIn: the LinkedIn post is only a teaser. This note gives the fuller framing and links to the runnable notebook.

This is a working demo, not a finished benchmark. It explores whether underspecified research methods can be detected before an AI reproduction agent writes code.

The first Colab section should run on the free Colab tier with a Google account and no API key. Larger-model sections require A100 runtime.

• Short technical note: https://github.com/ActionPace/ensemble-divergence/blob/main/social_science_extension.md
• Runnable notebook: https://github.com/ActionPace/ensemble-divergence/blob/main/ensemble_divergence_social_science.ipynb

ActionPace

#Ubuntu 16.04

sudo apt-get -y update
sudo apt-get -y upgrade
sudo apt-get -y install unzip
sudo apt-get -y install cmake

#Install Anaconda
wget https://repo.continuum.io/archive/Anaconda2-4.3.1-Linux-x86_64.sh
sudo bash Anaconda2-4.3.1-Linux-x86_64.sh -b -p /usr/local/anaconda2