Research · arXiv · Zenodo · ORCID
Measured, not guessed.
I don’t have opinions about AI visibility and LLM reliability — I measure them. Published, peer-reviewed work with open data and code.
Publications
Peer-reviewed and pre-registered work.
arXiv 2606.21595 · cs.CL + cs.IR
PEBM — Per-Entity Bias Mapping for AI Visibility
A map of AI hallucination per entity: 100 entities, 1,400 measurements. Tier-1 hallucination rate 52.7%, tier-3 37.9%. Published 2026-06-23; under review at IJHCS (Elsevier, IF 5.1).
RICE-P1 · pre-registered (OSF)
Quality Challenges Do Not Escalate Source Fabrication in LLMs
1,400 observations, 2×5×7 factorial. Source-fabrication rate 0.075 for ChatGPT vs 0.008 for Perplexity — a 9.2× gap. Code and data public.
Essays & positions
Ownable frameworks, not opinions.
Zenodo
Constitutive canon / Chamber Hub
Delegated agency and the state as a constitutive canon — an essay on the readability and contestability of power.
Zenodo · LinkedIn
The Exemption Zone
The “anatomy of omission” of EU AI Act Art. 49(4) — what it means when a high-risk system is left out of the registry.
Researcher identity
Cross-validated profile
ORCID 0009-0003-1020-834X · Zenodo · OSF · GitHub. An entity “exists” for a model only if it is cross-validated.