Fix zero-shot hang/crash: don't pickle embedding model via cache_data

Zero-shot questionnaire presets (MPE92, 11D-ASC) + savable category sets; clearer centered-cosine threshold help, default 0.50

Zero-shot model match, LLM model picker, prompt persistence + tweaks, export label

Reframe condition-comparison similarity as a correlation

Fix matplotlib 3.9+ crash: replace removed cm.get_cmap with plt.get_cmap

Add 'Load precomputed embeddings' data-source option

Fix condition comparison (mean-center embeddings), heatmap show-all toggle, CLI for run_embeddings, README local-embedding guide

Add embed-only script, random seed, editable LLM prompts, sharper map, top-N heatmap

Deploy app2 on the Space and support Qwen 4B embeddings

Fix CoherenceModel crash from tokenization mismatch in app2

Add app2.py (alternative interface) and its deps (plotly, seaborn, scipy)