Data release for When do large language models recommend leaving? Accountability-to-action asymmetries in relationship advice
(PNAS Nexus, under review). Code:
https://github.com/tomvannuenen/llm-accountabilitylabels/ — judge-coded construct labels (two-judge strict consensus,
Llama 4 Maverick + GLM-4.5-Air): T1/T2/T4 uptake constructs, refined
action categories, deontic packaging, and post-level… See the full description on the dataset page:
https://huggingface.co/datasets/tvannuenen/llm-accountability.
