Cleaned version of christinakopi/thinkprm-1k-verification-cots-numina-style.
Cleaning rule: COTs are reconstructed from prefix_steps and gt_step_labels so every verifier target has sequential Step N: labels and every step ends with exactly The step is \boxed{correct} or The step is \boxed{incorrect}. Final judgments are canonicalized from prefix_label.
Validation summary:
{
"num_rows": 1000,
"has_steps": 1000,
"sequential": 1000… See the full description on the dataset page:
https://huggingface.co/datasets/christinakopi/thinkprm-1k-verification-cots-numina-style-clean.
