curt is a general-purpose, machine-first programming language designed
for AI agents, with output-token cost as its primary design directive. This
dataset is the complete evaluation record for the language at version 0.2:
two held-out benchmark suites, every model-generated program produced during
evaluation (frozen verbatim), the reference corpus with Python twins, the
formal grammars, the documentation used as model context, and the exact… See the full description on the dataset page:
https://huggingface.co/datasets/therikkening/curt-benchmarks.
