hbuldurgan/authority-to-action
Authority-to-Action Evaluation Research question. When relevant context is present, does a tool-using language-model system distinguish evidence from permission to act? This dataset contains the 100 synthetic cases and the transcript-free, 600-row results ledger behind the LatentAtlas Authority-to-Action study. Paper (preprint): doi:10.5281/zenodo.21957491 Code, Inspect evaluation and deterministic verifiers: github.com/latentatlas/latentatlas-evidence-evals BENCHMARK DATA… See the full description on the dataset page: https://huggingface.co/datasets/hbuldurgan/authority-to-action.
055
Add Authority-to-Action cases, results ledger and dataset card
initial commit
