This site is under construction. If you find an error or bug, please talk to Charlotte in person or email sleeplesssometimes@gmail.com. Please do not use the channels on our contact page.

Myopic Delirium

The public interface has been delayed due to unforeseen circumstances. Appropriate adjustments have been made and we expect release prior to relocation in June 2026. We reserve all intellectual property rights. The Tholos set has not been and will not be submitted for publication to any journal. Previews of shortened articles will be released in our artifacts section beginning April 20, 2026 and June 1, 2026.

Parameters

Metric Pressure0.30

Scarcity0.35

Predators3

Simplified visualization — February 27, 2026

OUR MISSION

Myopic Delirium is a student-led research group focused on enriching agent architectures for large-scale social simulation. In these environments, agents are fundamentally representative of people, but across the board they are being treated as convenient characters. Despite notions of rigor and unbiased intent, the dominant use of models for social simulations at scale is to support and justify frameworks that institutions need to believe to maximize success. Influential banks create agents with shallow incentives based primarily on monetary goals, while leisure and any inefficiencies serve solely as frictions. Niche religions and spiritual organizations use rewrite assumptions of human nature and capabilities to develop utopic simulations that demonstrate the validity and absolutely true nature of their ideology. In each case the principles of the group are always directly reflected in the predictably supportive results they yield.

As an initiative, we assert ourselves on experimentation devoid of necessary structure and ideology, focused on the development of dynamic, emergent systems rather than fixed incentives. The rigor and ambition of our systems is accompanied by an uncertainty in their outcomes that gave us our name.

THE TECHNICAL

Our initiative developed with a small core team, most of which had worked in the development of limited models. That includes RBC / New Keynesian DSGE Models, CGE and IAM policy workhorses, macroprudential scenario model, and shallow, goal-seeking ABM models. Our best definition of our core model is a hybrid cognitive-institutional ABM, treating bounded cognition as the state space and using ensemble runs with sensitivity and ablation to map regime structures of social emergence.

Our experiments are designed for reproducible batch execution: we run ensembles on multi-core CPU infrastructure (workstations and HPC-style environments), with storage and logging optimized for large parameter sweeps. We are extremely grateful to the Stevens Institute of Technology’s Research Computing Services for access to the JARVIS cluster, and the Hanlon Financial Systems Center for providing compute resources that supported this work. We would also like to thank KTH Royal Institute of Technology and the PDC Center for High Performance Computing (NAISS) for providing initial access and support that enabled us to begin this work on high-performance computing infrastructure.

We have been grateful to receive input from some of the foremost experts across critical domains, but currently maintain 21 core members and have formed four teams: Agent Cognition & Behavioral Mechanics (ABBY); Research Software & Reproducibility (RODY); Empirical Methods, Metrics & Validation (EMMY); High-Performance Simulation Systems (HOBY). Our collective experience is growing, including members and former members of the MEB Biostatistics group at the Karolinska Institutet, Vienna Center for Experimental Economics at Universität Wien, along with participants at Treehacks 2026 and Hack the North in Waterloo.