Research

Two programmes. Empirical measurement at scale.

MTCP Programme

184,387 evaluations, 32 models, 13 providers, 12 languages
DOI: 10.17605/OSF.IO/DXGK5
Dataset: huggingface.co/datasets/aa8899/mtcp-boundary-500

Measures whether LLMs maintain compliance with explicit constraints across structured correction sequences.

ARCS Programme

2,607 multi-agent data points, 6 chain configurations
DOI: 10.5281/zenodo.20386024 (Papers 1-9), 10.5281/zenodo.20820033 (Papers 10-19)
Dataset: huggingface.co/datasets/aa8899/arcs-authority-vulnerability

Measures architectural alignment failures in multi-agent chains. Authority vulnerability that persists regardless of intervention.

Registered DOIs

DOI Title Access
10.17605/OSF.IO/DXGK5 MTCP Programme Public
10.5281/zenodo.20386024 ARCS Papers 1-9 Public
10.5281/zenodo.20820033 ARCS Papers 10-19 Public
10.5281/zenodo.20820273 Authority Vulnerability Index Q2 2026 Public
10.5281/zenodo.20820281 Framework Inventory v1.0 Restricted
10.5281/zenodo.20820337 Multi-Agent Constraint Assurance Public