Leakage-controlled subset of cds-jb/synthcognition
for 1930-reader experiments (talkie AV-NTA campaign): rows kept only when BOTH sentence and
cognitive_state contain nothing that did not exist or was not in use before 1931 — vocabulary and
slang judged by existence of usage (not tone), plus technologies, objects, institutions, occupations,
practices, and named entities.
Filter: claude-haiku-4-5-20251001 via the Message Batches API, 20 rows/request… See the full description on the dataset page:
https://huggingface.co/datasets/cds-jb/synthcognition-1930.
