Companion dataset to the independent ThinkStream research stack
(github.com/Khadgar123/ThinkStream-Custom).
Code and data versions are matched through release_manifest.json and the Git
commit recorded for the release.
Portable source-selection manifests per batch (metadata).
Per-batch pipeline artifacts: evidence_1a/1b, rollout, task_cards,
placements, samples_3c, verified, final.
Rendered SFT trajectories… See the full description on the dataset page:
https://huggingface.co/datasets/hheric/WTI.
