ClarusC64/

boundary_collapse_thermometer_v01

A micro-benchmark for sequential boundary integrity in LLMs. Most evals test a single prompt.BCT tests the slope of collapse — how models degrade over turns. This dataset measures: • boundary acknowledgement• policy memory across conversation• resistance to relational pressure• consistency under emotional or identity bait• collapse shape (early-warning → collapse_event → recovery) Columns:

turn_number: sequential interaction order… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/boundary_collapse_thermometer_v01.