What this dataset tests
Whether a model can generate clinically plausible counterfactual decision pointsanchored to a real patient timeline. Required outputs

decision_point_id
counterfactual_action
plausibility_score_0_100
branching_constraints

Counterfactual types

timing_shift
choice_substitution
disposition_change
intensity_change

Typical failures

proposing actions not linked to the real timeline
omitting safety constraints
producing fantasy improvements without… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/clinical-counterfactual-branch-construction-v0.1.