UPLOADED NON-SYNTHESIZED SOURCES · DIRECT GOOGLE CAPTIONS

Cleaned-up captions (v2)

Confidence-only fields removed; sparse details consolidated into descriptive text. Transcript wording and timing values are preserved. Cleaned-up does not mean human-reviewed.

1,610 uploaded source records · 1,609 unique video captions. One byte-identical pair shares a caption. Synthesized videos are excluded. Collection and upload counts are unchanged.

Gemini 3.8 · requested 2 fps / high spatial resolution · short captions, reconstruction, identity requirements and spoken content. Completed means structurally validated model output, still awaiting annotation review. These captions are unreviewed model annotations. The public export omits internal storage paths and operational credentials.